CN114387061A - Product push method, device, electronic device and readable storage medium - Google Patents

Abstract

The invention relates to an artificial intelligence technology, and discloses a product pushing method, which comprises the following steps: obtaining historical voice data of a target user, identifying user emotion from the historical voice data according to a preset emotion identification model, extracting user attribute features and user behavior features from historical user information, constructing user figures according to the user emotion, the user attribute features and the user behavior features, clustering the user figures to obtain user clustering labels, performing text screening on product information to obtain product labels, performing similarity clustering processing on the user clustering labels and the product labels, performing relevance arrangement on clustering results, and performing product pushing on the target user according to arrangement results. Furthermore, the invention relates to a blockchain technique, the permutation result being storable in a node of the blockchain. The invention also provides a product pushing method and device, electronic equipment and a computer readable storage medium. The invention can solve the problem of low product pushing accuracy.

Description

Translated fromChinese

产品推送方法、装置、电子设备及可读存储介质Product push method, device, electronic device and readable storage medium

技术领域technical field

本发明涉及人工智能技术领域，尤其涉及一种产品推送方法、装置、电子设备及计算机可读存储介质。The present invention relates to the technical field of artificial intelligence, and in particular, to a product push method, device, electronic device and computer-readable storage medium.

背景技术Background technique

智能推荐、千人千面已经运用在了典型的电商、短视频、广告等场景，根据用户的实时特征和需求精准推送信息，能够提升信息推送的准确度。然而在一些特定领域(如金融领域)尚未全面开展，例如在证券APP的使用场景中，目前主流的证券APP页面复杂，堆积了大量的数据、文章、消息，用户每天需要面临海量信息，但其中很难有真正关注和感兴趣的产品信息直接呈现给用户，导致产品、服务等消息的推送不准确。Intelligent recommendation, thousands of people and thousands of faces have been used in typical e-commerce, short video, advertising and other scenarios. It can accurately push information according to the real-time characteristics and needs of users, which can improve the accuracy of information push. However, in some specific fields (such as the financial field), it has not been fully developed. For example, in the use scenario of securities APP, the current mainstream securities APP pages are complex, accumulating a large amount of data, articles, and news, and users need to face a large amount of information every day, but among them It is difficult to directly present product information that is truly concerned and interesting to users, resulting in inaccurate push of information about products and services.

发明内容SUMMARY OF THE INVENTION

本发明提供一种产品推送方法、装置、设备及存储介质，其主要目的在于解决产品推送准确率较低的问题。The present invention provides a product push method, device, device and storage medium, the main purpose of which is to solve the problem of low product push accuracy.

为实现上述目的，本发明提供的一种产品推送方法，包括：To achieve the above purpose, a product push method provided by the present invention includes:

获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；Obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify user emotions from the voice data according to a preset emotion recognition model;

获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；Obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;

根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签；Construct a user portrait of the target user according to the user emotion, the user attribute feature and the user behavior feature, and perform clustering processing on the user portrait to obtain a user clustering label;

获取产品信息，对所述产品信息进行文本筛选，得到产品标签；Obtain product information, perform text screening on the product information, and obtain a product label;

对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。Perform similarity clustering processing on the user cluster tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and push products to the target users according to the ranking results .

可选地，所述对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，包括：Optionally, performing voice separation on the historical voice data to obtain the voice data of the target user, including:

对所述历史语音数据进行声道判断，并抽取所述历史语音数据中的目标声道的语音数据；Perform channel judgment on the historical voice data, and extract the voice data of the target channel in the historical voice data;

对所述目标声道的语音数据进行端点选取，并根据预设的时间长度和选取的端点对所述语音数据进行数据切割，得到所述目标用户的语音数据。Endpoint selection is performed on the voice data of the target channel, and data cutting is performed on the voice data according to a preset time length and the selected endpoint to obtain the voice data of the target user.

可选地，所述根据预设的情绪识别模型从所述语音数据中识别出用户情绪之前，所述方法还包括：Optionally, before identifying the user's emotion from the voice data according to a preset emotion recognition model, the method further includes:

获取原始训练集合，对所述原始训练集合进行数据增强处理，得到标准训练集合；Obtaining an original training set, performing data enhancement processing on the original training set, and obtaining a standard training set;

利用所述标准训练集合训练预构建的第一网络模型，得到原始模型；Utilize the standard training set to train the pre-built first network model to obtain the original model;

将所述原始模型的参数作为预构建的第二网络模型的初始化参数，并利用所述标准训练集合训练所述第二网络模型，得到所述情绪识别模型。The parameters of the original model are used as initialization parameters of the pre-built second network model, and the second network model is trained by using the standard training set to obtain the emotion recognition model.

可选地，所述从所述历史用户信息中提取用户属性特征及用户行为特征，包括：Optionally, the extraction of user attribute features and user behavior features from the historical user information includes:

获取预设的用户属性标签及用户行为标签；Get the preset user attribute labels and user behavior labels;

将所述历史用户信息中所述用户属性标签对应的特征作为所述用户属性特征，以及将所述历史用户信息中所述用户行为标签对应的特征作为所述用户行为特征。The feature corresponding to the user attribute tag in the historical user information is used as the user attribute feature, and the feature corresponding to the user behavior tag in the historical user information is used as the user behavior feature.

可选地，所述对所述用户画像进行聚类处理，得到用户聚类标签，包括：Optionally, performing clustering processing on the user portraits to obtain user clustering labels, including:

利用预设的语言模型对所述用户画像中的特征进行向量映射，得到特征向量集合；Use a preset language model to perform vector mapping on the features in the user portrait to obtain a feature vector set;

从所述特征向量集合中随机选取预设个数的用户样本作为聚类中心；Randomly select a preset number of user samples from the feature vector set as cluster centers;

依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离，并将所述每个用户样本分到距离最小的聚类中心对应的类别中，得到多个类别簇；Calculate the distance from each user sample in the feature vector set to the cluster center in turn, and classify each user sample into the category corresponding to the cluster center with the smallest distance to obtain a plurality of category clusters;

重新计算每个类别簇的聚类中心，并返回所述依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离的步骤，直至所述多个类别簇的聚类中心收敛，将收敛的类别簇对应的类别作为所述用户聚类标签。Recalculate the cluster center of each category cluster, and return to the step of sequentially calculating the distance from each user sample in the feature vector set to the cluster center, until the cluster centers of the multiple category clusters converge , and take the category corresponding to the converged category cluster as the user clustering label.

可选地，所述对所述产品信息进行文本筛选，得到产品标签，包括：Optionally, performing text screening on the product information to obtain a product label, including:

利用预设的分词算法对所述产品信息进行分词处理，得到产品分词文本；Using a preset word segmentation algorithm to perform word segmentation processing on the product information to obtain product word segmentation text;

利用预设的专有词汇表对所述产品分词文本进行初筛，得到分词筛选文本；Preliminarily screen the word segmentation text of the product by using a preset proprietary vocabulary to obtain the word segmentation screening text;

对所述分词筛选文本中的分词文本进行重要度筛选，并将重要度筛选后的分词文本作为所述产品标签。Perform importance screening on the word segmentation text in the word segmentation screening text, and use the word segmentation text after importance screening as the product label.

可选地，所述对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送，包括：Optionally, performing similarity clustering processing on the user clustering tags and the product tags to obtain a clustering result, performing at least two correlation rankings on the clustering results, and arranging all the clustering results according to the ranking results. The target users mentioned above for product push, including:

将所述用户聚类标签作为用户聚类中心；Taking the user clustering label as the user clustering center;

计算所述产品标签到所述用户聚类中心的距离，将所述距离小于预设距离阈值的产品标签作为所述用户聚类中心的聚类结果；Calculate the distance from the product label to the user clustering center, and use the product label whose distance is less than a preset distance threshold as the clustering result of the user clustering center;

对所述聚类结果中产品标签对应的产品及所述用户聚类标签对应的目标用户进行粗排处理，得到第一相关排列结果；Perform a rough sorting process on the products corresponding to the product labels in the clustering results and the target users corresponding to the user clustering labels to obtain a first related ranking result;

对所述第一相关排列结果中的所述目标用户及产品进行细排处理，得到第二相关排列结果；Performing fine-arranging processing on the target users and products in the first relevant arrangement result, to obtain a second relevant arrangement result;

从所述第二相关排列结果中选取预设个数的产品推送给所述第二相关排列结果中的所述目标用户。A preset number of products are selected from the second relevant arrangement result and pushed to the target user in the second relevant arrangement result.

为了解决上述问题，本发明还提供一种产品推送装置，所述装置包括：In order to solve the above problems, the present invention also provides a product pushing device, the device includes:

用户情绪识别模块，用于获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；The user emotion recognition module is used to obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify the user from the voice data according to a preset emotion recognition model mood;

用户特征提取模块，用于获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；a user feature extraction module, configured to obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;

标签构建模块，用于根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签，获取产品信息，对所述产品信息进行文本筛选，得到产品标签；The label building module is used to construct a user portrait of the target user according to the user emotion, the user attribute characteristics and the user behavior characteristics, and perform clustering processing on the user portrait to obtain user clustering labels, and obtain products. information, performing text screening on the product information to obtain a product label;

产品推送模块，用于对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。The product push module is used to perform similarity clustering processing on the user clustering tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and rank all the clustering results according to the ranking results. Push the product to the target users.

为了解决上述问题，本发明还提供一种电子设备，所述电子设备包括：In order to solve the above problems, the present invention also provides an electronic device, the electronic device includes:

存储器，存储至少一个计算机程序；及a memory that stores at least one computer program; and

处理器，执行所述存储器中存储的计算机程序以实现上述所述的产品推送方法。The processor executes the computer program stored in the memory to implement the product pushing method described above.

为了解决上述问题，本发明还提供一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一个计算机程序，所述至少一个计算机程序被电子设备中的处理器执行以实现上述所述的产品推送方法。In order to solve the above problems, the present invention also provides a computer-readable storage medium, where at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is executed by a processor in an electronic device to realize the above-mentioned The product delivery method described above.

本发明通过预设的情绪识别模型从用户语音数据中识别出目标用户的用户情绪，并根据用户情绪、用户属性特征及用户行为特征构建目标用户的用户画像，用户画像中的用户特征更加丰富，因此对用户画像进行聚类处理，可以更准确的对目标用户进行分类。并且，通过对用户聚类标签及产品标签进行相似度聚类处理，从用户及产品两方面出发进行聚类，并对聚类结果进行至少两次相关性排列，进一步提高了用户及产品的相关性，使得对产品的推送更加准确。因此本发明提出的产品推送方法、装置、电子设备及计算机可读存储介质，可以解决产品推送准确率较低的问题。The present invention identifies the user emotion of the target user from the user voice data through a preset emotion recognition model, and constructs the user portrait of the target user according to the user emotion, user attribute characteristics and user behavior characteristics, and the user characteristics in the user portrait are more abundant, Therefore, clustering the user portraits can more accurately classify the target users. In addition, by performing similarity clustering processing on user cluster tags and product tags, clustering is performed from both users and products, and the clustering results are correlated at least twice, which further improves the correlation between users and products. , which makes the push of products more accurate. Therefore, the product push method, device, electronic device and computer-readable storage medium proposed by the present invention can solve the problem of low product push accuracy.

附图说明Description of drawings

图1为本发明一实施例提供的产品推送方法的流程示意图；1 is a schematic flowchart of a product push method provided by an embodiment of the present invention;

图2为本发明一实施例提供的产品推送装置的功能模块图；2 is a functional block diagram of a product push device provided by an embodiment of the present invention;

图3为本发明一实施例提供的实现所述产品推送方法的电子设备的结构示意图。FIG. 3 is a schematic structural diagram of an electronic device implementing the product push method according to an embodiment of the present invention.

本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。The realization, functional characteristics and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments.

具体实施方式Detailed ways

应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

本申请实施例提供一种产品推送方法。所述产品推送方法的执行主体包括但不限于服务端、终端等能够被配置为执行本申请实施例提供的该方法的电子设备中的至少一种。换言之，所述产品推送方法可以由安装在终端设备或服务端设备的软件或硬件来执行，所述软件可以是区块链平台。所述服务端包括但不限于：单台服务器、服务器集群、云端服务器或云端服务器集群等。所述服务器可以是独立的服务器，也可以是提供云服务、云数据库、云计算、云函数、云存储、网络服务、云通信、中间件服务、域名服务、安全服务、内容分发网络(ContentDelivery Network，CDN)、以及大数据和人工智能平台等基础云计算服务的云服务器。The embodiment of the present application provides a method for pushing a product. The execution body of the product push method includes, but is not limited to, at least one of electronic devices that can be configured to execute the method provided by the embodiments of the present application, such as a server, a terminal, and the like. In other words, the product push method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like. The server can be an independent server, or can provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (Content Delivery Network) , CDN), and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms.

参照图1所示，为本发明一实施例提供的产品推送方法的流程示意图。Referring to FIG. 1 , it is a schematic flowchart of a method for pushing a product according to an embodiment of the present invention.

在本实施例中，所述产品推送方法包括：In this embodiment, the product push method includes:

S1、获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪。S1. Acquire historical voice data of a target user, perform voice separation on the historical voice data, obtain voice data of the target user, and identify user emotions from the voice data according to a preset emotion recognition model.

本发明实施例中，所述历史语音数据是指用户在交易或售后服务时与客服的语音数据，例如，在金融领域，所述历史语音数据可以为证券公司已有的用户关于证券交易及客服人员沟通的语音数据。所述目标用户可以为从用户信息表中圈选出来的目标人群，例如，从银行的用户信息表中圈选10万人或某个人作为目标用户。In the embodiment of the present invention, the historical voice data refers to the voice data of the user and the customer service during transactions or after-sales service. For example, in the financial field, the historical voice data may be the existing users of the securities company about securities transactions and customer service. Voice data of personnel communication. The target user may be a target group selected from the user information table, for example, 100,000 people or a certain person is selected from the user information table of the bank as the target user.

具体地，所述对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，包括：Specifically, the voice separation of the historical voice data to obtain the voice data of the target user includes:

对所述历史语音数据进行声道判断，并抽取所述历史语音数据中的目标声道的语音数据；Perform channel judgment on the historical voice data, and extract the voice data of the target channel in the historical voice data;

对所述目标声道的语音数据进行端点选取，并根据预设的时间长度和选取的端点对所述语音数据进行数据切割，得到所述目标用户的语音数据。Endpoint selection is performed on the voice data of the target channel, and data cutting is performed on the voice data according to a preset time length and the selected endpoint to obtain the voice data of the target user.

本发明实施例中，所述目标声道可以为左声道或者右声道。In this embodiment of the present invention, the target channel may be a left channel or a right channel.

例如，在银行领域中，因为所述历史语音数据是用户与客服的录音，仅为两方通话，且音频为双声道音频，其中左声道为坐席(客服)音频，右声道为用户音频，所以直接选取右声道为目标声道，并抽取右声道语音数据即可。For example, in the banking field, because the historical voice data is a recording of the user and the customer service, it is only a call between two parties, and the audio is two-channel audio, where the left channel is the audio of the agent (customer service), and the right channel is the user's audio. Audio, so directly select the right channel as the target channel, and extract the voice data of the right channel.

可选的，可以采用语音端点检测(Voice Activity Detection，VAD)技术对右声道语音数据进行语音端点选取，在实际应用中，待检测的语音数据往往会包含无效的声音，例如噪声、他人说话声音等，VAD技术可以从带有噪声的语音中准确的定位出语音的开始和结束点，即把静音和噪声作为干扰信号从原始数据中去除。Optionally, the voice endpoint detection (Voice Activity Detection, VAD) technology can be used to select the voice endpoint for the right channel voice data. In practical applications, the voice data to be detected often contains invalid sounds, such as noise, others speaking. Voice, etc., VAD technology can accurately locate the start and end points of speech from speech with noise, that is, remove silence and noise as interference signals from the original data.

本发明实施例中，所述预设的时间长度可以为1s，以1s为单位，对所述右声道语音数据进行切割，切割成若干片，其中，若音频的最后一片不足1s，则基于该片音频的平均值进行填充，直至长度为1s。In the embodiment of the present invention, the preset time length may be 1s, and the voice data of the right channel is cut in units of 1s into several pieces. If the last piece of audio is less than 1s, the audio The average value of the piece of audio is padded until the length is 1s.

详细地，所述根据预设的情绪识别模型从所述语音数据中识别出用户情绪之前，所述方法还包括：In detail, before identifying the user's emotion from the voice data according to the preset emotion recognition model, the method further includes:

获取原始训练集合，对所述原始训练集合进行数据增强处理，得到标准训练集合；Obtaining an original training set, performing data enhancement processing on the original training set, and obtaining a standard training set;

利用所述标准训练集合训练预构建的第一网络模型，得到原始模型；Utilize the standard training set to train the pre-built first network model to obtain the original model;

将所述原始模型的参数作为预构建的第二网络模型的初始化参数，并利用所述标准训练集合训练所述第二网络模型，得到所述情绪识别模型。The parameters of the original model are used as initialization parameters of the pre-built second network model, and the second network model is trained by using the standard training set to obtain the emotion recognition model.

本发明一可选实施例中，所述原始训练集合可以为CASIA汉语情感语料库，并且可以使用混类(Mixup)增强方法，对原始训练集合进行数据增强。所述第一网络模型可以为ResNet50网络，所述第二网络模型可以为改进的ResNet50网络，改进之处在于：去除ResNet50网络的第一层和最后一层，并在ResNet50网络之前加入一层批量训练(BatchNormalization)层、一层卷积层(激活函数为relu)、一层平均池化层，在ResNet50网络后加入一层全连接层(激活函数为relu)、一层批量训练(BatchNormalization)层和最后的全连接层，基于上述改进，可以加速模型训练，同时使得到的情绪识别模型更加适用于音频数据，提高情绪识别的准确度。In an optional embodiment of the present invention, the original training set may be the CASIA Chinese sentiment corpus, and a mixed-class (Mixup) enhancement method may be used to perform data enhancement on the original training set. The first network model can be a ResNet50 network, and the second network model can be an improved ResNet50 network. The improvement lies in: removing the first layer and the last layer of the ResNet50 network, and adding a layer of batches before the ResNet50 network A training (BatchNormalization) layer, a convolutional layer (activation function is relu), an average pooling layer, a fully connected layer (activation function is relu) and a batch training (BatchNormalization) layer are added after the ResNet50 network And the final fully connected layer, based on the above improvements, can speed up model training, and at the same time make the resulting emotion recognition model more suitable for audio data and improve the accuracy of emotion recognition.

S2、获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征。S2. Obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information.

本发明实施例中，所述历史用户信息包括用户的基本信息及交易行为信息。In the embodiment of the present invention, the historical user information includes basic information and transaction behavior information of the user.

详细地，所述从所述历史用户信息中提取用户属性特征及用户行为特征，包括：In detail, the extraction of user attribute features and user behavior features from the historical user information includes:

获取预设的用户属性标签及用户行为标签；Get the preset user attribute labels and user behavior labels;

将所述历史用户信息中所述用户属性标签对应的特征作为所述用户属性特征，以及将所述历史用户信息中所述用户行为标签对应的特征作为所述用户行为特征。The feature corresponding to the user attribute tag in the historical user information is used as the user attribute feature, and the feature corresponding to the user behavior tag in the historical user information is used as the user behavior feature.

本发明一可选实施例中，所述用户属性标签用来表示用户的社会属性，例如，金融领域中，根据用户开立证券账户时的信息，所述用户属性标签包括：性别、年龄、教育程度等。In an optional embodiment of the present invention, the user attribute label is used to represent the social attribute of the user. For example, in the financial field, according to the information when the user opens a securities account, the user attribute label includes: gender, age, education degree, etc.

所述用户行为标签用来表示用户的交易属性，例如，金融领域中，根据用户的交易数据，所述用户行为标签包括：市场板块偏好、交易频率、交易额度、用户风险偏好等。The user behavior tag is used to represent the user's transaction attributes. For example, in the financial field, according to the user's transaction data, the user behavior tag includes: market sector preference, transaction frequency, transaction amount, and user risk preference.

可选的，例如，用户A的用户属性特征包括：性别：男、年龄：24、教育程度：本科，用户交易特征包括：市场板块偏好：医疗、交易频率：高、交易额度：1000、用户风险偏好：中等。Optionally, for example, user attribute characteristics of user A include: gender: male, age: 24, education level: undergraduate, user transaction characteristics include: market sector preference: medical, transaction frequency: high, transaction amount: 1000, user risk Preference: Moderate.

S3、根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签。S3. Construct a user portrait of the target user according to the user emotion, the user attribute feature, and the user behavior feature, and perform clustering processing on the user portrait to obtain a user clustering label.

本发明实施例中，所述用户画像包括用户的情绪特征、用户的属性特征及用户的行为特征，因此对用户的描述更加准确，能够提高对用户进行产品推荐的准确度。In the embodiment of the present invention, the user portrait includes the user's emotional characteristics, the user's attribute characteristics, and the user's behavior characteristics, so the description of the user is more accurate, and the accuracy of product recommendation to the user can be improved.

具体地，所述对所述用户画像进行聚类处理，得到用户聚类标签，包括：Specifically, performing clustering processing on the user portraits to obtain user clustering labels, including:

利用预设的语言模型对所述用户画像中的特征进行向量映射，得到特征向量集合；Use a preset language model to perform vector mapping on the features in the user portrait to obtain a feature vector set;

从所述特征向量集合中随机选取预设个数的用户样本作为聚类中心；Randomly select a preset number of user samples from the feature vector set as cluster centers;

依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离，并将所述每个用户样本分到距离最小的聚类中心对应的类别中，得到多个类别簇；Calculate the distance from each user sample in the feature vector set to the cluster center in turn, and classify each user sample into the category corresponding to the cluster center with the smallest distance to obtain a plurality of category clusters;

重新计算每个类别簇的聚类中心，并返回所述依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离的步骤，直至所述多个类别簇的聚类中心收敛，将收敛的类别簇对应的类别作为所述用户聚类标签。Recalculate the cluster center of each category cluster, and return to the step of sequentially calculating the distance from each user sample in the feature vector set to the cluster center, until the cluster centers of the multiple category clusters converge , and take the category corresponding to the converged category cluster as the user clustering label.

本发明实施例中，所述预设的语言模型可以为bert模型、RoBERTa模型等。所述距离可以为欧氏距离、曼哈顿距离及切比雪夫距离等。In this embodiment of the present invention, the preset language model may be a bert model, a RoBERTa model, or the like. The distance can be Euclidean distance, Manhattan distance, Chebyshev distance, etc.

详细地，所述计算每个类别簇的聚类中心，包括：In detail, the calculation of the cluster center of each category cluster includes:

通过下述聚类公式计算每个类别簇的聚类中心：The cluster center of each category cluster is calculated by the following clustering formula:

其中，E_i为第i个聚类中心，C_i为第i个类别簇，x为类别簇中的用户样本。Among them, E_i is the ith cluster center, C_i is the ith category cluster, and x is the user sample in the category cluster.

本发明另一可选实施例中，还可以使用均值漂移聚类方法、基于密度的聚类方法(DBSCAN)、用高斯混合模型(GMM)的最大期望(EM)聚类方法等来进行聚类处理。例如，在金融领域，所述用户聚类标签包括爱打新、喜爱短线操作、喜爱小盘概念股、喜爱买基金等。In another optional embodiment of the present invention, a mean shift clustering method, a density-based clustering method (DBSCAN), an expectation maximum (EM) clustering method using Gaussian Mixture Model (GMM), etc. can also be used for clustering deal with. For example, in the financial field, the user cluster tags include likes to play new, like short-term operations, like small-cap concept stocks, like to buy funds, and the like.

进一步地，由于所述用户画像中包括更加丰富的用户特征，由此通过聚类处理，可以更加准确地对用户进行分类，从而提高产品推送的准确率。Further, since the user portraits include more abundant user characteristics, users can be classified more accurately through clustering processing, thereby improving the accuracy of product push.

S4、获取产品信息，对所述产品信息进行文本筛选，得到产品标签。S4. Obtain product information, and perform text screening on the product information to obtain a product label.

本发明实施例中，所述产品信息包括产品的类别、产品描述等。In this embodiment of the present invention, the product information includes a product category, a product description, and the like.

具体地，所述对所述产品信息进行文本筛选，得到产品标签，包括：Specifically, performing text screening on the product information to obtain a product label, including:

利用预设的分词算法对所述产品信息进行分词处理，得到产品分词文本；Using a preset word segmentation algorithm to perform word segmentation processing on the product information to obtain product word segmentation text;

利用预设的专有词汇表对所述产品分词文本进行初筛，得到分词筛选文本；Preliminarily screen the word segmentation text of the product by using a preset proprietary vocabulary to obtain the word segmentation screening text;

对所述分词筛选文本中的分词文本进行重要度筛选，并将重要度筛选后的分词文本作为所述产品标签。Perform importance screening on the word segmentation text in the word segmentation screening text, and use the word segmentation text after importance screening as the product label.

本发明实施例中，所述预设的分词算法可以为现有的任何分词方法，如中科院计算所NLPIR、哈工大的LTP、清华大学THULAC、结巴分词、斯坦福分词器等等。In the embodiment of the present invention, the preset word segmentation algorithm may be any existing word segmentation method, such as NLPIR of the Institute of Computing Technology of the Chinese Academy of Sciences, LTP of Harbin Institute of Technology, THULAC of Tsinghua University, stuttering word segmentation, Stanford word segmentation device and so on.

可选的，所述预设的专有词汇表可以为金融领域专家整理的金融产品专有词汇表格。并且，利用TF-IDF算法计算所述分词筛选文本中每个分词的重要度，并按照重要度对分词文本进行排序，选取预设比例的分词文本作为所述产品标签。Optionally, the preset proprietary vocabulary may be a financial product-specific vocabulary compiled by experts in the financial field. Furthermore, the TF-IDF algorithm is used to calculate the importance of each segment in the segmented screening text, the segmented texts are sorted according to their importance, and a preset proportion of segmented texts is selected as the product label.

本发明实施例中，通过分词算法进行分词，并利用专有词汇表进行初筛，以及使用TF-IDF算法进行重要度筛选，可以提高产品标签选取的准确性。In the embodiment of the present invention, word segmentation is performed by a word segmentation algorithm, a proprietary vocabulary is used for preliminary screening, and a TF-IDF algorithm is used for importance screening, which can improve the accuracy of product label selection.

S5、对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。S5. Perform similarity clustering processing on the user clustering tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and perform a ranking on the target users according to the ranking results. Product push.

本发明实施例中，根据所述用户聚类标签及所述产品标签进行来对用户及产品进行匹配，通过同时从用户及产品两方面进行聚类及相关性排列，可以提高产品推送的准确率。In the embodiment of the present invention, users and products are matched according to the user cluster tags and the product tags, and the accuracy of product push can be improved by performing clustering and correlation arrangement from both users and products at the same time. .

其中，相关性是指与用户聚类标签的关联程度。具体的，相关性排列是指对所述用户聚类标签及所述产品标签进行语义理解，将与用户聚类标签关联性较大的产品标签按照从大到小的顺序进行排列，从而可以根据相关性大小的序列，从高到低向用户推送产品，提高产品推送的准确度。Among them, correlation refers to the degree of association with user cluster labels. Specifically, the relevancy arrangement refers to semantically understanding the user cluster tags and the product tags, and arranging the product tags that are more related to the user cluster tags in descending order, so that the tags can be sorted according to the The sequence of correlation size, push products to users from high to low, and improve the accuracy of product push.

详细地，所述对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送，包括：In detail, the similarity clustering process is performed on the user cluster tags and the product tags to obtain a clustering result, the clustering results are correlated at least twice, and the clustering results are arranged according to the sorting results. Target users for product push, including:

将所述用户聚类标签作为用户聚类中心；Taking the user clustering label as the user clustering center;

计算所述产品标签到所述用户聚类中心的距离，将所述距离小于预设距离阈值的产品标签作为所述用户聚类中心的聚类结果；Calculate the distance from the product label to the user clustering center, and use the product label whose distance is less than a preset distance threshold as the clustering result of the user clustering center;

对所述聚类结果中产品标签对应的产品及所述用户聚类标签对应的目标用户进行粗排处理，得到第一相关排列结果；Perform a rough sorting process on the products corresponding to the product labels in the clustering results and the target users corresponding to the user clustering labels to obtain a first related ranking result;

对所述第一相关排列结果中的所述目标用户及产品进行细排处理，得到第二相关排列结果；Performing fine-arranging processing on the target users and products in the first relevant arrangement result, to obtain a second relevant arrangement result;

从所述第二相关排列结果中选取预设个数的产品推送给所述第二相关排列结果中的所述目标用户。A preset number of products are selected from the second relevant arrangement result and pushed to the target user in the second relevant arrangement result.

本发明实施例中，所述粗排是指从万级产品中粗选出百级数量产品，例如，在推送广告场景中，由于存在万级别的广告数，直接进行推送不仅准确率不高且大量数据的处理效率也较低，因此需要从上万个广告中筛选出几百个相关性较高的目标广告进行排列。其中，粗排可以为基于向量内积的深度模型，一般为双塔结构，两侧分别输入用户特征和产品特征，经过深度网络计算后，分别产出用户向量和产品向量，再通过内积等运算计算得到相关排序分数，根据相关排序分数得到用户及产品的第一相关序列结果。In the embodiment of the present invention, the rough ranking refers to the rough selection of hundreds of products from 10,000-level products. For example, in a push advertisement scenario, because there are 10,000-level advertisements, direct push not only has a low accuracy rate but also The processing efficiency of a large amount of data is also low, so it is necessary to screen out hundreds of highly relevant target advertisements from tens of thousands of advertisements for arrangement. Among them, the rough row can be a depth model based on vector inner product, which is generally a double-tower structure. User features and product features are input on both sides. After deep network calculation, the user vector and product vector are respectively produced, and then through the inner product, etc. The relevant ranking score is obtained by operation calculation, and the first relevant sequence result of the user and the product is obtained according to the relevant ranking score.

所述细排是指对粗排后的序列进一步进行精细排序，所述细排处理可以通过LR、DT、SVM、CTR模型等算法进行排列。The fine sorting refers to further fine sorting the sequence after the rough sorting, and the fine sorting processing may be performed by algorithms such as LR, DT, SVM, and CTR model.

例如，用户聚类标签及其对应的用户包括：label1：user1、label2：user2、label3：user3及label4：user4，产品标签对应的产品包括：item1、item2、item3及item4，将所述label1、label2、label3及label4依次作为用户聚类中心，分别计算item1、item2、item3及item4到这四个标签的距离(余弦距离等)，得到聚类结果：label1：item1、item2、item3及item4；label2：item1、item2、item3及item4；label3：item1、item2、item3及item4；label4：item1、item2、item3及item4，对每个聚类结果中的用户聚类标签和对应的产品标签进行粗排，则通过粗排处理后的第一相关排列结果为：user1：item1、item2、item3；user2：item2、item3、item4；user3：item3、item4、item1；user4：item4、item1、item2。再对第一相关排列结果中的用户和产品标签进行细排，通过细排可以进一步对用户进行更加精准的产品相关排列，例如，通过细排处理后的第二相关排列结果包括：user1：item1、item2；user2：item2、item3；user3：item3；user4：item4、item1。则可以分别向user1推送item1或item2、向user2推送item2或item3、向user3推送item3及向user4推送item4或item1。For example, the user cluster labels and their corresponding users include: label1: user1, label2: user2, label3: user3 and label4: user4, the products corresponding to the product labels include: item1, item2, item3 and item4, the label1, label2 , label3 and label4 are used as user clustering centers in turn, calculate the distances (cosine distance, etc.) from item1, item2, item3 and item4 to these four labels respectively, and get the clustering results: label1: item1, item2, item3 and item4; label2: item1, item2, item3, and item4; label3: item1, item2, item3, and item4; label4: item1, item2, item3, and item4. Roughly arrange the user clustering labels and corresponding product labels in each clustering result, then The first related arrangement results after the rough arrangement are: user1: item1, item2, item3; user2: item2, item3, item4; user3: item3, item4, item1; user4: item4, item1, item2. The user and product labels in the first related arrangement result are then finely arranged, and the users can be further arranged in a more accurate product related arrangement through the fine arrangement. For example, the second related arrangement result after the fine arrangement processing includes: user1: item1 , item2; user2: item2, item3; user3: item3; user4: item4, item1. Then you can push item1 or item2 to user1, push item2 or item3 to user2, push item3 to user3, and push item4 or item1 to user4, respectively.

详细地，所述计算所述产品标签到所述用户聚类中心的距离，包括：Specifically, the calculating the distance from the product label to the user cluster center includes:

将所述用户聚类中心作为目标标签，依次选择所述产品标签中任意一个标签作为比较标签，对所述目标标签及所述比较标签进行分词处理，得到目标列表及比较列表；Taking the user clustering center as a target label, selecting any one of the product labels as a comparison label in turn, and performing word segmentation processing on the target label and the comparison label to obtain a target list and a comparison list;

根据所述目标列表及所述比较列表构建编码字典；build an encoding dictionary according to the target list and the comparison list;

利用所述编码字典对所述目标列表及所述比较列表进行向量编码，得到目标向量及比较向量；Perform vector encoding on the target list and the comparison list by using the encoding dictionary to obtain a target vector and a comparison vector;

利用预设的余弦相似度计算公式计算所述目标向量与所述比较向量的目标相似度，确定所述目标相似度为所述产品标签到所述用户聚类中心的距离。The target similarity between the target vector and the comparison vector is calculated by using a preset cosine similarity calculation formula, and the target similarity is determined as the distance from the product label to the user cluster center.

本发明一可选实施例中，利用下述公式计算相似度：In an optional embodiment of the present invention, the similarity is calculated using the following formula:

其中，a为所述目标向量，b为所述比较向量。Among them, a is the target vector, and b is the comparison vector.

本发明实施例中，通过用户聚类标签及产品标签进行聚类，相当于对用户及产品进行初步匹配，可以减少不相关产品的数量，大大提高了产品推荐的速度。并且通过粗排、细排算法对产品及用户进行相关性排列，进一步提高了产品推送的准确率。In the embodiment of the present invention, clustering through user clustering tags and product tags is equivalent to preliminary matching of users and products, which can reduce the number of irrelevant products and greatly improve the speed of product recommendation. And through the rough sorting and fine sorting algorithms, the correlation between products and users is arranged, which further improves the accuracy of product push.

本发明通过预设的情绪识别模型从用户语音数据中识别出目标用户的用户情绪，并根据用户情绪、用户属性特征及用户行为特征构建目标用户的用户画像，用户画像中的用户特征更加丰富，因此对用户画像进行聚类处理，可以更准确的对目标用户进行分类。并且，通过对用户聚类标签及产品标签进行相似度聚类处理，从用户及产品两方面出发进行聚类，并对聚类结果进行至少两次相关性排列，进一步提高了用户及产品的相关性，使得对产品的推送更加准确。因此本发明提出的产品推送方法，可以解决产品推送准确率较低的问题。The present invention identifies the user emotion of the target user from the user voice data through a preset emotion recognition model, and constructs the user portrait of the target user according to the user emotion, user attribute characteristics and user behavior characteristics, and the user characteristics in the user portrait are more abundant, Therefore, clustering the user portraits can more accurately classify the target users. In addition, by performing similarity clustering processing on user clustering tags and product tags, clustering is carried out from two aspects of users and products, and the clustering results are correlated at least twice, which further improves the correlation between users and products. It makes the push of products more accurate. Therefore, the product push method proposed by the present invention can solve the problem of low product push accuracy.

如图2所示，是本发明一实施例提供的产品推送装置的功能模块图。As shown in FIG. 2 , it is a functional block diagram of a product pushing device provided by an embodiment of the present invention.

本发明所述产品推送装置100可以安装于电子设备中。根据实现的功能，所述产品推送装置100可以包括用户情绪识别模块101、用户特征提取模块102、标签构建模块103及产品推送模块104。本发明所述模块也可以称之为单元，是指一种能够被电子设备处理器所执行，并且能够完成固定功能的一系列计算机程序段，其存储在电子设备的存储器中。Theproduct pushing device 100 of the present invention can be installed in an electronic device. According to the implemented functions, theproduct push device 100 may include a useremotion recognition module 101 , a userfeature extraction module 102 , alabel construction module 103 and aproduct push module 104 . The modules in the present invention can also be called units, which refer to a series of computer program segments that can be executed by the electronic device processor and can perform fixed functions, and are stored in the memory of the electronic device.

在本实施例中，关于各模块/单元的功能如下：In this embodiment, the functions of each module/unit are as follows:

所述用户情绪识别模块101，用于获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；The useremotion recognition module 101 is configured to acquire the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and extract the voice data from the voice data according to a preset emotion recognition model. Identify user emotions;

所述用户特征提取模块102，用于获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；The userfeature extraction module 102 is configured to obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;

所述标签构建模块103，用于根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签，获取产品信息，对所述产品信息进行文本筛选，得到产品标签；Thelabel construction module 103 is configured to construct a user portrait of the target user according to the user emotion, the user attribute characteristics and the user behavior characteristics, and perform clustering processing on the user portrait to obtain a user clustering label. , obtain product information, perform text screening on the product information, and obtain product labels;

所述产品推送模块104，用于对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。Theproduct push module 104 is configured to perform similarity clustering processing on the user cluster tags and the product tags to obtain a clustering result, and perform at least two correlation arrangements on the clustering results, and according to the arrangement As a result, product push is performed to the target user.

详细地，所述产品推送装置100各模块的具体实施方式如下：In detail, the specific implementation of each module of theproduct pushing device 100 is as follows:

步骤一、获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪。Step 1: Obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify user emotions from the voice data according to a preset emotion recognition model.

本发明实施例中，所述历史语音数据是指用户在交易或售后服务时与客服的语音数据，例如，在金融领域，所述历史语音数据可以为证券公司已有的用户关于证券交易及客服人员沟通的语音数据。所述目标用户可以为从用户信息表中圈选出来的目标人群，例如，从银行的用户信息表中圈选10万人或某个人作为目标用户。In the embodiment of the present invention, the historical voice data refers to the voice data of the user and the customer service during transactions or after-sales service. For example, in the financial field, the historical voice data may be the existing users of the securities company about securities transactions and customer service. Voice data of personnel communication. The target user may be a target group selected from the user information table, for example, 100,000 people or a certain person is selected from the user information table of the bank as the target user.

具体地，所述对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，包括：Specifically, the voice separation of the historical voice data to obtain the voice data of the target user includes:

对所述历史语音数据进行声道判断，并抽取所述历史语音数据中的目标声道的语音数据；Perform channel judgment on the historical voice data, and extract the voice data of the target channel in the historical voice data;

对所述目标声道的语音数据进行端点选取，并根据预设的时间长度和选取的端点对所述语音数据进行数据切割，得到所述目标用户的语音数据。Endpoint selection is performed on the voice data of the target channel, and data cutting is performed on the voice data according to a preset time length and the selected endpoint to obtain the voice data of the target user.

本发明实施例中，所述目标声道可以为左声道或者右声道。In this embodiment of the present invention, the target channel may be a left channel or a right channel.

例如，在银行领域中，因为所述历史语音数据是用户与客服的录音，仅为两方通话，且音频为双声道音频，其中左声道为坐席(客服)音频，右声道为用户音频，所以直接选取右声道为目标声道，并抽取右声道语音数据即可。For example, in the banking field, because the historical voice data is a recording of the user and the customer service, it is only a call between two parties, and the audio is two-channel audio, where the left channel is the audio of the agent (customer service), and the right channel is the user's audio. Audio, so directly select the right channel as the target channel, and extract the voice data of the right channel.

可选的，可以采用语音端点检测(Voice Activity Detection，VAD)技术对右声道语音数据进行语音端点选取，在实际应用中，待检测的语音数据往往会包含无效的声音，例如噪声、他人说话声音等，VAD技术可以从带有噪声的语音中准确的定位出语音的开始和结束点，即把静音和噪声作为干扰信号从原始数据中去除。Optionally, the voice endpoint detection (Voice Activity Detection, VAD) technology can be used to select the voice endpoint for the right channel voice data. In practical applications, the voice data to be detected often contains invalid sounds, such as noise, others speaking. Voice, etc., VAD technology can accurately locate the start and end points of speech from speech with noise, that is, remove silence and noise as interference signals from the original data.

本发明实施例中，所述预设的时间长度可以为1s，以1s为单位，对所述右声道语音数据进行切割，切割成若干片，其中，若音频的最后一片不足1s，则基于该片音频的平均值进行填充，直至长度为1s。In the embodiment of the present invention, the preset time length may be 1s, and the voice data of the right channel is cut in units of 1s into several pieces. If the last piece of audio is less than 1s, the audio The average value of the piece of audio is padded until the length is 1s.

详细地，所述根据预设的情绪识别模型从所述语音数据中识别出用户情绪之前，所述方法还包括：In detail, before identifying the user's emotion from the voice data according to the preset emotion recognition model, the method further includes:

获取原始训练集合，对所述原始训练集合进行数据增强处理，得到标准训练集合；Obtaining an original training set, performing data enhancement processing on the original training set, and obtaining a standard training set;

利用所述标准训练集合训练预构建的第一网络模型，得到原始模型；Utilize the standard training set to train the pre-built first network model to obtain the original model;

将所述原始模型的参数作为预构建的第二网络模型的初始化参数，并利用所述标准训练集合训练所述第二网络模型，得到所述情绪识别模型。The parameters of the original model are used as initialization parameters of the pre-built second network model, and the second network model is trained by using the standard training set to obtain the emotion recognition model.

本发明一可选实施例中，所述原始训练集合可以为CASIA汉语情感语料库，并且可以使用混类(Mixup)增强方法，对原始训练集合进行数据增强。所述第一网络模型可以为ResNet50网络，所述第二网络模型可以为改进的ResNet50网络，改进之处在于：去除ResNet50网络的第一层和最后一层，并在ResNet50网络之前加入一层批量训练(BatchNormalization)层、一层卷积层(激活函数为relu)、一层平均池化层，在ResNet50网络后加入一层全连接层(激活函数为relu)、一层批量训练(BatchNormalization)层和最后的全连接层，基于上述改进，可以加速模型训练，同时使得到的情绪识别模型更加适用于音频数据，提高情绪识别的准确度。In an optional embodiment of the present invention, the original training set may be the CASIA Chinese sentiment corpus, and a mixed-class (Mixup) enhancement method may be used to perform data enhancement on the original training set. The first network model can be a ResNet50 network, and the second network model can be an improved ResNet50 network. The improvement lies in: removing the first layer and the last layer of the ResNet50 network, and adding a layer of batches before the ResNet50 network A training (BatchNormalization) layer, a convolutional layer (activation function is relu), an average pooling layer, a fully connected layer (activation function is relu) and a batch training (BatchNormalization) layer are added after the ResNet50 network And the final fully connected layer, based on the above improvements, can speed up model training, and at the same time make the resulting emotion recognition model more suitable for audio data and improve the accuracy of emotion recognition.

步骤二、获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征。Step 2: Obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information.

本发明实施例中，所述历史用户信息包括用户的基本信息及交易行为信息。In the embodiment of the present invention, the historical user information includes basic information and transaction behavior information of the user.

详细地，所述从所述历史用户信息中提取用户属性特征及用户行为特征，包括：In detail, the extraction of user attribute features and user behavior features from the historical user information includes:

获取预设的用户属性标签及用户行为标签；Get the preset user attribute labels and user behavior labels;

将所述历史用户信息中所述用户属性标签对应的特征作为所述用户属性特征，以及将所述历史用户信息中所述用户行为标签对应的特征作为所述用户行为特征。The feature corresponding to the user attribute tag in the historical user information is used as the user attribute feature, and the feature corresponding to the user behavior tag in the historical user information is used as the user behavior feature.

本发明一可选实施例中，所述用户属性标签用来表示用户的社会属性，例如，金融领域中，根据用户开立证券账户时的信息，所述用户属性标签包括：性别、年龄、教育程度等。In an optional embodiment of the present invention, the user attribute label is used to represent the social attribute of the user. For example, in the financial field, according to the information when the user opens a securities account, the user attribute label includes: gender, age, education degree, etc.

所述用户行为标签用来表示用户的交易属性，例如，金融领域中，根据用户的交易数据，所述用户行为标签包括：市场板块偏好、交易频率、交易额度、用户风险偏好等。The user behavior tag is used to represent the user's transaction attributes. For example, in the financial field, according to the user's transaction data, the user behavior tag includes: market sector preference, transaction frequency, transaction amount, and user risk preference.

可选的，例如，用户A的用户属性特征包括：性别：男、年龄：24、教育程度：本科，用户交易特征包括：市场板块偏好：医疗、交易频率：高、交易额度：1000、用户风险偏好：中等。Optionally, for example, user attribute characteristics of user A include: gender: male, age: 24, education level: undergraduate, user transaction characteristics include: market sector preference: medical, transaction frequency: high, transaction amount: 1000, user risk Preference: Moderate.

步骤三、根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签。Step 3: Construct a user portrait of the target user according to the user emotion, the user attribute feature and the user behavior feature, and perform a clustering process on the user portrait to obtain a user clustering label.

本发明实施例中，所述用户画像包括用户的情绪特征、用户的属性特征及用户的行为特征，因此对用户的描述更加准确，能够提高对用户进行产品推荐的准确度。In the embodiment of the present invention, the user portrait includes the user's emotional characteristics, the user's attribute characteristics, and the user's behavior characteristics, so the description of the user is more accurate, and the accuracy of product recommendation to the user can be improved.

具体地，所述对所述用户画像进行聚类处理，得到用户聚类标签，包括：Specifically, performing clustering processing on the user portraits to obtain user clustering labels, including:

利用预设的语言模型对所述用户画像中的特征进行向量映射，得到特征向量集合；Use a preset language model to perform vector mapping on the features in the user portrait to obtain a feature vector set;

从所述特征向量集合中随机选取预设个数的用户样本作为聚类中心；Randomly select a preset number of user samples from the feature vector set as cluster centers;

依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离，并将所述每个用户样本分到距离最小的聚类中心对应的类别中，得到多个类别簇；Calculate the distance from each user sample in the feature vector set to the cluster center in turn, and classify each user sample into the category corresponding to the cluster center with the smallest distance to obtain a plurality of category clusters;

重新计算每个类别簇的聚类中心，并返回所述依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离的步骤，直至所述多个类别簇的聚类中心收敛，将收敛的类别簇对应的类别作为所述用户聚类标签。Recalculate the cluster center of each category cluster, and return to the step of sequentially calculating the distance from each user sample in the feature vector set to the cluster center, until the cluster centers of the multiple category clusters converge , and take the category corresponding to the converged category cluster as the user clustering label.

本发明实施例中，所述预设的语言模型可以为bert模型、RoBERTa模型等。所述距离可以为欧氏距离、曼哈顿距离及切比雪夫距离等。In this embodiment of the present invention, the preset language model may be a bert model, a RoBERTa model, or the like. The distance can be Euclidean distance, Manhattan distance, Chebyshev distance, etc.

详细地，所述计算每个类别簇的聚类中心，包括：In detail, the calculation of the cluster center of each category cluster includes:

通过下述聚类公式计算每个类别簇的聚类中心：The cluster center of each category cluster is calculated by the following clustering formula:

其中，E_i为第i个聚类中心，C_i为第i个类别簇，x为类别簇中的用户样本。Among them, E_i is the ith cluster center, C_i is the ith category cluster, and x is the user sample in the category cluster.

本发明另一可选实施例中，还可以使用均值漂移聚类方法、基于密度的聚类方法(DBSCAN)、用高斯混合模型(GMM)的最大期望(EM)聚类方法等来进行聚类处理。例如，在金融领域，所述用户聚类标签包括爱打新、喜爱短线操作、喜爱小盘概念股、喜爱买基金等。In another optional embodiment of the present invention, a mean shift clustering method, a density-based clustering method (DBSCAN), an expectation maximum (EM) clustering method using Gaussian Mixture Model (GMM), etc. can also be used for clustering deal with. For example, in the financial field, the user cluster tags include likes to play new, like short-term operations, like small-cap concept stocks, like to buy funds, and the like.

进一步地，由于所述用户画像中包括更加丰富的用户特征，由此通过聚类处理，可以更加准确地对用户进行分类，从而提高产品推送的准确率。Further, since the user portraits include more abundant user characteristics, users can be classified more accurately through clustering processing, thereby improving the accuracy of product push.

步骤四、获取产品信息，对所述产品信息进行文本筛选，得到产品标签。Step 4: Obtain product information, perform text screening on the product information, and obtain a product label.

本发明实施例中，所述产品信息包括产品的类别、产品描述等。In this embodiment of the present invention, the product information includes a product category, a product description, and the like.

具体地，所述对所述产品信息进行文本筛选，得到产品标签，包括：Specifically, performing text screening on the product information to obtain a product label, including:

利用预设的分词算法对所述产品信息进行分词处理，得到产品分词文本；Using a preset word segmentation algorithm to perform word segmentation processing on the product information to obtain product word segmentation text;

利用预设的专有词汇表对所述产品分词文本进行初筛，得到分词筛选文本；Preliminarily screen the word segmentation text of the product by using a preset proprietary vocabulary to obtain the word segmentation screening text;

对所述分词筛选文本中的分词文本进行重要度筛选，并将重要度筛选后的分词文本作为所述产品标签。Perform importance screening on the word segmentation text in the word segmentation screening text, and use the word segmentation text after importance screening as the product label.

本发明实施例中，所述预设的分词算法可以为现有的任何分词方法，如中科院计算所NLPIR、哈工大的LTP、清华大学THULAC、结巴分词、斯坦福分词器等等。In the embodiment of the present invention, the preset word segmentation algorithm may be any existing word segmentation method, such as NLPIR of the Institute of Computing Technology of the Chinese Academy of Sciences, LTP of Harbin Institute of Technology, THULAC of Tsinghua University, stuttering word segmentation, Stanford word segmentation device and so on.

可选的，所述预设的专有词汇表可以为金融领域专家整理的金融产品专有词汇表格。并且，利用TF-IDF算法计算所述分词筛选文本中每个分词的重要度，并按照重要度对分词文本进行排序，选取预设比例的分词文本作为所述产品标签。Optionally, the preset proprietary vocabulary may be a financial product-specific vocabulary compiled by experts in the financial field. Furthermore, the TF-IDF algorithm is used to calculate the importance of each segment in the segmented screening text, the segmented texts are sorted according to their importance, and a preset proportion of segmented texts is selected as the product label.

本发明实施例中，通过分词算法进行分词，并利用专有词汇表进行初筛，以及使用TF-IDF算法进行重要度筛选，可以提高产品标签选取的准确性。In the embodiment of the present invention, word segmentation is performed by a word segmentation algorithm, a proprietary vocabulary is used for preliminary screening, and a TF-IDF algorithm is used for importance screening, which can improve the accuracy of product label selection.

步骤五、对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。Step 5. Perform similarity clustering processing on the user clustering tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and classify the target users according to the ranking results. Product push.

本发明实施例中，根据所述用户聚类标签及所述产品标签进行来对用户及产品进行匹配，通过同时从用户及产品两方面进行聚类及相关性排列，可以提高产品推送的准确率。In the embodiment of the present invention, users and products are matched according to the user cluster tags and the product tags, and the accuracy of product push can be improved by performing clustering and correlation arrangement from both users and products at the same time. .

其中，相关性是指与用户聚类标签的关联程度。具体的，相关性排列是指对所述用户聚类标签及所述产品标签进行语义理解，将与用户聚类标签关联性较大的产品标签按照从大到小的顺序进行排列，从而可以根据相关性大小的序列，从高到低向用户推送产品，提高产品推送的准确度。Among them, correlation refers to the degree of association with user cluster labels. Specifically, the relevancy arrangement refers to semantically understanding the user cluster tags and the product tags, and arranging the product tags that are more related to the user cluster tags in descending order, so that the tags can be sorted according to the The sequence of correlation size, push products to users from high to low, and improve the accuracy of product push.

详细地，所述对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送，包括：In detail, the similarity clustering process is performed on the user cluster tags and the product tags to obtain a clustering result, the clustering results are correlated at least twice, and the clustering results are arranged according to the sorting results. Target users for product push, including:

将所述用户聚类标签作为用户聚类中心；Taking the user clustering label as the user clustering center;

计算所述产品标签到所述用户聚类中心的距离，将所述距离小于预设距离阈值的产品标签作为所述用户聚类中心的聚类结果；Calculate the distance from the product label to the user clustering center, and use the product label whose distance is less than a preset distance threshold as the clustering result of the user clustering center;

对所述聚类结果中产品标签对应的产品及所述用户聚类标签对应的目标用户进行粗排处理，得到第一相关排列结果；Perform a rough sorting process on the products corresponding to the product labels in the clustering results and the target users corresponding to the user clustering labels to obtain a first related ranking result;

对所述第一相关排列结果中的所述目标用户及产品进行细排处理，得到第二相关排列结果；Performing fine-arranging processing on the target users and products in the first relevant arrangement result, to obtain a second relevant arrangement result;

从所述第二相关排列结果中选取预设个数的产品推送给所述第二相关排列结果中的所述目标用户。A preset number of products are selected from the second relevant arrangement result and pushed to the target user in the second relevant arrangement result.

本发明实施例中，所述粗排是指从万级产品中粗选出百级数量产品，例如，在推送广告场景中，由于存在万级别的广告数，直接进行推送不仅准确率不高且大量数据的处理效率也较低，因此需要从上万个广告中筛选出几百个相关性较高的目标广告进行排列。其中，粗排可以为基于向量内积的深度模型，一般为双塔结构，两侧分别输入用户特征和产品特征，经过深度网络计算后，分别产出用户向量和产品向量，再通过内积等运算计算得到相关排序分数，根据相关排序分数得到用户及产品的第一相关序列结果。In the embodiment of the present invention, the rough ranking refers to the rough selection of hundreds of products from 10,000-level products. For example, in a push advertisement scenario, because there are 10,000-level advertisements, direct push not only has a low accuracy rate but also The processing efficiency of a large amount of data is also low, so it is necessary to screen out hundreds of highly relevant target advertisements from tens of thousands of advertisements for arrangement. Among them, the rough row can be a depth model based on vector inner product, which is generally a double-tower structure. User features and product features are input on both sides. After deep network calculation, the user vector and product vector are respectively produced, and then through the inner product, etc. The relevant ranking score is obtained by operation calculation, and the first relevant sequence result of the user and the product is obtained according to the relevant ranking score.

所述细排是指对粗排后的序列进一步进行精细排序，所述细排处理可以通过LR、DT、SVM、CTR模型等算法进行排列。The fine sorting refers to further fine sorting the sequence after the rough sorting, and the fine sorting processing may be performed by algorithms such as LR, DT, SVM, and CTR model.

例如，用户聚类标签及其对应的用户包括：label1：user1、label2：user2、label3：user3及label4：user4，产品标签对应的产品包括：item1、item2、item3及item4，将所述label1、label2、label3及label4依次作为用户聚类中心，分别计算item1、item2、item3及item4到这四个标签的距离(余弦距离等)，得到聚类结果：label1：item1、item2、item3及item4；label2：item1、item2、item3及item4；label3：item1、item2、item3及item4；label4：item1、item2、item3及item4，对每个聚类结果中的用户聚类标签和对应的产品标签进行粗排，则通过粗排处理后的第一相关排列结果为：user1：item1、item2、item3；user2：item2、item3、item4；user3：item3、item4、item1；user4：item4、item1、item2。再对第一相关排列结果中的用户和产品标签进行细排，通过细排可以进一步对用户进行更加精准的产品相关排列，例如，通过细排处理后的第二相关排列结果包括：user1：item1、item2；user2：item2、item3；user3：item3；user4：item4、item1。则可以分别向user1推送item1或item2、向user2推送item2或item3、向user3推送item3及向user4推送item4或item1。For example, the user cluster labels and their corresponding users include: label1: user1, label2: user2, label3: user3 and label4: user4, the products corresponding to the product labels include: item1, item2, item3 and item4, the label1, label2 , label3 and label4 are used as user clustering centers in turn, calculate the distances (cosine distance, etc.) from item1, item2, item3 and item4 to these four labels respectively, and get the clustering results: label1: item1, item2, item3 and item4; label2: item1, item2, item3, and item4; label3: item1, item2, item3, and item4; label4: item1, item2, item3, and item4. Roughly arrange the user clustering labels and corresponding product labels in each clustering result, then The first related arrangement results after the rough arrangement are: user1: item1, item2, item3; user2: item2, item3, item4; user3: item3, item4, item1; user4: item4, item1, item2. The user and product labels in the first related arrangement result are then finely arranged, and the users can be further arranged in a more accurate product related arrangement through the fine arrangement. For example, the second related arrangement result after the fine arrangement processing includes: user1: item1 , item2; user2: item2, item3; user3: item3; user4: item4, item1. Then you can push item1 or item2 to user1, push item2 or item3 to user2, push item3 to user3, and push item4 or item1 to user4, respectively.

详细地，所述计算所述产品标签到所述用户聚类中心的距离，包括：Specifically, the calculating the distance from the product label to the user cluster center includes:

将所述用户聚类中心作为目标标签，依次选择所述产品标签中任意一个标签作为比较标签，对所述目标标签及所述比较标签进行分词处理，得到目标列表及比较列表；Taking the user clustering center as a target label, selecting any one of the product labels as a comparison label in turn, and performing word segmentation processing on the target label and the comparison label to obtain a target list and a comparison list;

根据所述目标列表及所述比较列表构建编码字典；build an encoding dictionary according to the target list and the comparison list;

利用所述编码字典对所述目标列表及所述比较列表进行向量编码，得到目标向量及比较向量；Perform vector encoding on the target list and the comparison list by using the encoding dictionary to obtain a target vector and a comparison vector;

利用预设的余弦相似度计算公式计算所述目标向量与所述比较向量的目标相似度，确定所述目标相似度为所述产品标签到所述用户聚类中心的距离。The target similarity between the target vector and the comparison vector is calculated by using a preset cosine similarity calculation formula, and the target similarity is determined as the distance from the product label to the user cluster center.

本发明一可选实施例中，利用下述公式计算相似度：In an optional embodiment of the present invention, the similarity is calculated using the following formula:

其中，a为所述目标向量，b为所述比较向量。Among them, a is the target vector, and b is the comparison vector.

本发明实施例中，通过用户聚类标签及产品标签进行聚类，相当于对用户及产品进行初步匹配，可以减少不相关产品的数量，大大提高了产品推荐的速度。并且通过粗排、细排算法对产品及用户进行相关性排列，进一步提高了产品推送的准确率。In the embodiment of the present invention, clustering through user clustering tags and product tags is equivalent to preliminary matching of users and products, which can reduce the number of irrelevant products and greatly improve the speed of product recommendation. And through the rough sorting and fine sorting algorithms, the correlation between products and users is arranged, which further improves the accuracy of product push.

本发明通过预设的情绪识别模型从用户语音数据中识别出目标用户的用户情绪，并根据用户情绪、用户属性特征及用户行为特征构建目标用户的用户画像，用户画像中的用户特征更加丰富，因此对用户画像进行聚类处理，可以更准确的对目标用户进行分类。并且，通过对用户聚类标签及产品标签进行相似度聚类处理，从用户及产品两方面出发进行聚类，并对聚类结果进行至少两次相关性排列，进一步提高了用户及产品的相关性，使得对产品的推送更加准确。因此本发明提出的产品推送装置，可以解决产品推送准确率较低的问题。The present invention identifies the user emotion of the target user from the user voice data through a preset emotion recognition model, and constructs the user portrait of the target user according to the user emotion, user attribute characteristics and user behavior characteristics, and the user characteristics in the user portrait are more abundant. Therefore, clustering the user portraits can more accurately classify the target users. In addition, by performing similarity clustering processing on user cluster tags and product tags, clustering is performed from both users and products, and the clustering results are correlated at least twice, which further improves the correlation between users and products. , which makes the push of products more accurate. Therefore, the product push device proposed by the present invention can solve the problem of low product push accuracy.

如图3所示，是本发明一实施例提供的实现产品推送方法的电子设备的结构示意图。As shown in FIG. 3 , it is a schematic structural diagram of an electronic device for implementing a product push method provided by an embodiment of the present invention.

所述电子设备可以包括处理器10、存储器11、通信接口12和总线13，还可以包括存储在所述存储器11中并可在所述处理器10上运行的计算机程序，如产品推送程序。The electronic device may include aprocessor 10, amemory 11, acommunication interface 12, and abus 13, and may also include a computer program stored in thememory 11 and executable on theprocessor 10, such as a product push program.

其中，所述存储器11至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、移动硬盘、多媒体卡、卡型存储器(例如：SD或DX存储器等)、磁性存储器、磁盘、光盘等。所述存储器11在一些实施例中可以是电子设备的内部存储单元，例如该电子设备的移动硬盘。所述存储器11在另一些实施例中也可以是电子设备的外部存储设备，例如电子设备上配备的插接式移动硬盘、智能存储卡(Smart Media Card，SMC)、安全数字(SecureDigital，SD)卡、闪存卡(Flash Card)等。进一步地，所述存储器11还可以既包括电子设备的内部存储单元也包括外部存储设备。所述存储器11不仅可以用于存储安装于电子设备的应用软件及各类数据，例如产品推送程序的代码等，还可以用于暂时地存储已经输出或者将要输出的数据。Wherein, thememory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, CD etc. In some embodiments, thememory 11 may be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, thememory 11 may also be an external storage device of the electronic device, such as a pluggable mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) equipped on the electronic device. Cards, Flash Cards, etc. Further, thememory 11 may also include both an internal storage unit of an electronic device and an external storage device. Thememory 11 can not only be used to store application software installed on the electronic device and various data, such as the code of a product push program, etc., but also can be used to temporarily store data that has been output or will be output.

所述处理器10在一些实施例中可以由集成电路组成，例如可以由单个封装的集成电路所组成，也可以是由多个相同功能或不同功能封装的集成电路所组成，包括一个或者多个中央处理器(Central Processing unit，CPU)、微处理器、数字处理芯片、图形处理器及各种控制芯片的组合等。所述处理器10是所述电子设备的控制核心(Control Unit)，利用各种接口和线路连接整个电子设备的各个部件，通过运行或执行存储在所述存储器11内的程序或者模块(例如产品推送程序等)，以及调用存储在所述存储器11内的数据，以执行电子设备的各种功能和处理数据。In some embodiments, theprocessor 10 may be composed of integrated circuits, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits packaged with the same function or different functions, including one or more integrated circuits. Central processing unit (Central Processing Unit, CPU), microprocessor, digital processing chip, graphics processor and combination of various control chips, etc. Theprocessor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect the various components of the entire electronic device, by running or executing the program or module (such as a product) stored in thememory 11. Push programs, etc.), and call data stored in thememory 11 to perform various functions of the electronic device and process data.

所述通信接口12用于上述电子设备与其他设备之间的通信，包括网络接口和用户接口。可选地，所述网络接口可以包括有线接口和/或无线接口(如WI-FI接口、蓝牙接口等)，通常用于在该电子设备与其他电子设备之间建立通信连接。所述用户接口可以是显示器(Display)、输入单元(比如键盘(Keyboard))，可选地，用户接口还可以是标准的有线接口、无线接口。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。其中，显示器也可以适当的称为显示屏或显示单元，用于显示在电子设备中处理的信息以及用于显示可视化的用户界面。Thecommunication interface 12 is used for communication between the above electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and/or a wireless interface (eg, a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, and the like. The display may also be appropriately called a display screen or a display unit, which is used for displaying information processed in the electronic device and for displaying a visual user interface.

所述总线13可以是外设部件互连标准(peripheral component interconnect，简称PCI)总线或扩展工业标准结构(extended industry standard architecture，简称EISA)总线等。该总线13可以分为地址总线、数据总线、控制总线等。所述总线13被设置为实现所述存储器11以及至少一个处理器10等之间的连接通信。Thebus 13 may be a peripheral component interconnect (PCI for short) bus or an extended industry standard architecture (extended industry standard architecture, EISA for short) bus or the like. Thebus 13 can be divided into an address bus, a data bus, a control bus, and the like. Thebus 13 is configured to implement connection communication between thememory 11 and at least oneprocessor 10 and the like.

图3仅示出了具有部件的电子设备，本领域技术人员可以理解的是，图3示出的结构并不构成对所述电子设备的限定，可以包括比图示更少或者更多的部件，或者组合某些部件，或者不同的部件布置。FIG. 3 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 3 does not constitute a limitation on the electronic device, and may include fewer or more components than those shown in the drawings. , or a combination of certain components, or a different arrangement of components.

例如，尽管未示出，所述电子设备还可以包括给各个部件供电的电源(比如电池)，优选地，电源可以通过电源管理装置与所述至少一个处理器10逻辑相连，从而通过电源管理装置实现充电管理、放电管理、以及功耗管理等功能。电源还可以包括一个或一个以上的直流或交流电源、再充电装置、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。所述电子设备还可以包括多种传感器、蓝牙模块、Wi-Fi模块等，在此不再赘述。For example, although not shown, the electronic device may also include a power source (such as a battery) for powering the various components, preferably, the power source may be logically connected to the at least oneprocessor 10 through a power management device, so that the power source can be logically connected through the power management device. Implement functions such as charge management, discharge management, and power management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components. The electronic device may further include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

进一步地，所述电子设备还可以包括网络接口，可选地，所述网络接口可以包括有线接口和/或无线接口(如WI-FI接口、蓝牙接口等)，通常用于在该电子设备与其他电子设备之间建立通信连接。Further, the electronic device may also include a network interface, optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.) Establish a communication connection between other electronic devices.

可选地，该电子设备还可以包括用户接口，用户接口可以是显示器(Display)、输入单元(比如键盘(Keyboard))，可选地，用户接口还可以是标准的有线接口、无线接口。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。其中，显示器也可以适当的称为显示屏或显示单元，用于显示在电子设备中处理的信息以及用于显示可视化的用户界面。Optionally, the electronic device may further include a user interface, and the user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, and the like. The display may also be appropriately called a display screen or a display unit, which is used for displaying information processed in the electronic device and for displaying a visual user interface.

应该了解，所述实施例仅为说明之用，在专利申请范围上并不受此结构的限制。It should be understood that the embodiments are only used for illustration, and are not limited by this structure in the scope of the patent application.

所述电子设备中的所述存储器11存储的产品推送程序是多个指令的组合，在所述处理器10中运行时，可以实现：The product push program stored in thememory 11 in the electronic device is a combination of multiple instructions, and when running in theprocessor 10, it can achieve:

获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；Obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify user emotions from the voice data according to a preset emotion recognition model;

获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；Obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;

根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签；Construct a user portrait of the target user according to the user emotion, the user attribute feature and the user behavior feature, and perform clustering processing on the user portrait to obtain a user clustering label;

获取产品信息，对所述产品信息进行文本筛选，得到产品标签；Obtain product information, perform text screening on the product information, and obtain a product label;

对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。Perform similarity clustering processing on the user cluster tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and push products to the target users according to the ranking results .

具体地，所述处理器10对上述指令的具体实现方法可参考附图对应实施例中相关步骤的描述，在此不赘述。Specifically, for the specific implementation method of the above-mentioned instruction by theprocessor 10, reference may be made to the description of the relevant steps in the embodiment corresponding to the accompanying drawings, which will not be repeated here.

进一步地，所述电子设备集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读存储介质中。所述计算机可读存储介质可以是易失性的，也可以是非易失性的。例如，所述计算机可读介质可以包括：能够携带所述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM，Read-Only Memory)。Further, if the modules/units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. The computer-readable storage medium may be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM, Read-Only). Memory).

本发明还提供一种计算机可读存储介质，所述可读存储介质存储有计算机程序，所述计算机程序在被电子设备的处理器所执行时，可以实现：The present invention also provides a computer-readable storage medium, where the readable storage medium stores a computer program, and when executed by a processor of an electronic device, the computer program can realize:

获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；Obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify user emotions from the voice data according to a preset emotion recognition model;

获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；Obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;

根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签；Construct a user portrait of the target user according to the user emotion, the user attribute feature and the user behavior feature, and perform clustering processing on the user portrait to obtain a user clustering label;

获取产品信息，对所述产品信息进行文本筛选，得到产品标签；Obtain product information, perform text screening on the product information, and obtain a product label;

对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。Perform similarity clustering processing on the user cluster tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and push products to the target users according to the ranking results .

在本发明所提供的几个实施例中，应该理解到，所揭露的设备，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and there may be other division manners in actual implementation.

所述作为分离部件说明的模块可以是或者也可以不是物理上分开的，作为模块显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

另外，在本发明各个实施例中的各功能模块可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用硬件加软件功能模块的形式实现。In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software function modules.

对于本领域技术人员而言，显然本发明不限于上述示范性实施例的细节，而且在不背离本发明的精神或基本特征的情况下，能够以其他的具体形式实现本发明。It will be apparent to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, but that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics of the invention.

因此，无论从哪一点来看，均应将实施例看作是示范性的，而且是非限制性的，本发明的范围由所附权利要求而不是上述说明限定，因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本发明内。不应将权利要求中的任何附关联图标记视为限制所涉及的权利要求。Therefore, the embodiments are to be regarded in all respects as illustrative and not restrictive, and the scope of the invention is to be defined by the appended claims rather than the foregoing description, which are therefore intended to fall within the scope of the claims. All changes within the meaning and range of the equivalents of , are included in the present invention. Any reference signs in the claims shall not be construed as limiting the involved claim.

本申请实施例可以基于人工智能技术对相关的数据进行获取和处理。其中，人工智能(Artificial Intelligence，AI)是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能，感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。The embodiments of the present application may acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. .

人工智能基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、机器人技术、生物识别技术、语音处理技术、自然语言处理技术以及机器学习/深度学习等几大方向。The basic technologies of artificial intelligence generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning/deep learning.

本发明所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。The blockchain referred to in the present invention is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

此外，显然“包括”一词不排除其他单元或步骤，单数不排除复数。系统权利要求中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第二等词语用来表示名称，而并不表示任何特定的顺序。Furthermore, it is clear that the word "comprising" does not exclude other units or steps and the singular does not exclude the plural. Several units or means recited in the system claims can also be implemented by one unit or means by means of software or hardware. Second-class terms are used to denote names and do not denote any particular order.

最后应说明的是，以上实施例仅用以说明本发明的技术方案而非限制，尽管参照较佳实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，可以对本发明的技术方案进行修改或等同替换，而不脱离本发明技术方案的精神和范围。Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be Modifications or equivalent substitutions can be made without departing from the spirit and scope of the technical solutions of the present invention.

Claims (10)

Translated fromChinese

1.一种产品推送方法，其特征在于，所述方法包括：1. a product push method, is characterized in that, described method comprises:获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；Obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify user emotions from the voice data according to a preset emotion recognition model;获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；Obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签；Construct a user portrait of the target user according to the user emotion, the user attribute feature and the user behavior feature, and perform clustering processing on the user portrait to obtain a user clustering label;获取产品信息，对所述产品信息进行文本筛选，得到产品标签；Obtain product information, perform text screening on the product information, and obtain a product label;对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。Perform similarity clustering processing on the user clustering tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and push products to the target users according to the ranking results .2.如权利要求1所述的产品推送方法，其特征在于，所述对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，包括：2. The product push method according to claim 1, wherein the voice separation is performed on the historical voice data to obtain the voice data of the target user, comprising:对所述历史语音数据进行声道判断，并抽取所述历史语音数据中的目标声道的语音数据；Perform channel judgment on the historical voice data, and extract the voice data of the target channel in the historical voice data;对所述目标声道的语音数据进行端点选取，并根据预设的时间长度和选取的端点对所述语音数据进行数据切割，得到所述目标用户的语音数据。Endpoint selection is performed on the voice data of the target channel, and data cutting is performed on the voice data according to a preset time length and the selected endpoint to obtain the voice data of the target user.3.如权利要求2中所述的产品推送方法，其特征在于，所述根据预设的情绪识别模型从所述语音数据中识别出用户情绪之前，所述方法还包括：3. The product push method according to claim 2, wherein before the user's emotion is identified from the voice data according to a preset emotion recognition model, the method further comprises:获取原始训练集合，对所述原始训练集合进行数据增强处理，得到标准训练集合；Obtaining an original training set, performing data enhancement processing on the original training set, and obtaining a standard training set;利用所述标准训练集合训练预构建的第一网络模型，得到原始模型；Utilize the standard training set to train the pre-built first network model to obtain the original model;将所述原始模型的参数作为预构建的第二网络模型的初始化参数，并利用所述标准训练集合训练所述第二网络模型，得到所述情绪识别模型。The parameters of the original model are used as initialization parameters of the pre-built second network model, and the second network model is trained by using the standard training set to obtain the emotion recognition model.4.如权利要求1所述的产品推送方法，其特征在于，所述从所述历史用户信息中提取用户属性特征及用户行为特征，包括：4. The product push method according to claim 1, wherein the extraction of user attribute features and user behavior features from the historical user information comprises:获取预设的用户属性标签及用户行为标签；Get the preset user attribute labels and user behavior labels;将所述历史用户信息中所述用户属性标签对应的特征作为所述用户属性特征，以及将所述历史用户信息中所述用户行为标签对应的特征作为所述用户行为特征。The feature corresponding to the user attribute tag in the historical user information is used as the user attribute feature, and the feature corresponding to the user behavior tag in the historical user information is used as the user behavior feature.5.如权利要求4所述的产品推送方法，其特征在于，所述对所述用户画像进行聚类处理，得到用户聚类标签，包括：5. The product push method according to claim 4, wherein the described user portraits are clustered to obtain user clustering labels, comprising:利用预设的语言模型对所述用户画像中的特征进行向量映射，得到特征向量集合；Use a preset language model to perform vector mapping on the features in the user portrait to obtain a feature vector set;从所述特征向量集合中随机选取预设个数的用户样本作为聚类中心；Randomly select a preset number of user samples from the feature vector set as cluster centers;依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离，并将所述每个用户样本分到距离最小的聚类中心对应的类别中，得到多个类别簇；Calculate the distance from each user sample in the feature vector set to the cluster center in turn, and classify each user sample into the category corresponding to the cluster center with the smallest distance to obtain a plurality of category clusters;重新计算每个类别簇的聚类中心，并返回所述依次计算所述特征向量集合中每个用户样本到所述聚类中心的距离的步骤，直至所述多个类别簇的聚类中心收敛，将收敛的类别簇对应的类别作为所述用户聚类标签。Recalculate the cluster center of each category cluster, and return to the step of sequentially calculating the distance from each user sample in the feature vector set to the cluster center, until the cluster centers of the multiple category clusters converge , and take the category corresponding to the converged category cluster as the user clustering label.6.如权利要求1所述的产品推送方法，其特征在于，所述对所述产品信息进行文本筛选，得到产品标签，包括：6. The method for pushing a product according to claim 1 , wherein the product information is subjected to text screening to obtain a product label, comprising:利用预设的分词算法对所述产品信息进行分词处理，得到产品分词文本；Using a preset word segmentation algorithm to perform word segmentation processing on the product information to obtain product word segmentation text;利用预设的专有词汇表对所述产品分词文本进行初筛，得到分词筛选文本；Preliminarily screen the word segmentation text of the product by using a preset proprietary vocabulary to obtain the word segmentation screening text;对所述分词筛选文本中的分词文本进行重要度筛选，并将重要度筛选后的分词文本作为所述产品标签。Perform importance screening on the word segmentation text in the word segmentation screening text, and use the word segmentation text after importance screening as the product label.7.如权利要求1所述的产品推送方法，其特征在于，所述对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送，包括：7. The product push method according to claim 1, wherein the similarity clustering process is performed on the user clustering label and the product label to obtain a clustering result, and the clustering result is performed on the clustering result. Relevance rankings are performed at least twice, and products are pushed to the target users according to the ranking results, including:将所述用户聚类标签作为用户聚类中心；Taking the user clustering label as the user clustering center;计算所述产品标签到所述用户聚类中心的距离，将所述距离小于预设距离阈值的产品标签作为所述用户聚类中心的聚类结果；Calculate the distance from the product label to the user clustering center, and use the product label whose distance is less than a preset distance threshold as the clustering result of the user clustering center;对所述聚类结果中产品标签对应的产品及所述用户聚类标签对应的目标用户进行粗排处理，得到第一相关排列结果；Perform rough sorting on the product corresponding to the product label in the clustering result and the target user corresponding to the user clustering label, to obtain a first related ranking result;对所述第一相关排列结果中的所述目标用户及产品进行细排处理，得到第二相关排列结果；Performing fine-arranging processing on the target users and products in the first relevant arrangement result to obtain a second relevant arrangement result;从所述第二相关排列结果中选取预设个数的产品推送给所述第二相关排列结果中的所述目标用户。A preset number of products are selected from the second relevant arrangement result and pushed to the target user in the second relevant arrangement result.8.一种产品推送装置，其特征在于，所述装置包括：8. A product pushing device, wherein the device comprises:用户情绪识别模块，用于获取目标用户的历史语音数据，对所述历史语音数据进行语音分离，得到所述目标用户的语音数据，根据预设的情绪识别模型从所述语音数据中识别出用户情绪；The user emotion recognition module is used to obtain the historical voice data of the target user, perform voice separation on the historical voice data, obtain the voice data of the target user, and identify the user from the voice data according to a preset emotion recognition model mood;用户特征提取模块，用于获取所述目标用户的历史用户信息，从所述历史用户信息中提取用户属性特征及用户行为特征；a user feature extraction module, configured to obtain historical user information of the target user, and extract user attribute features and user behavior features from the historical user information;标签构建模块，用于根据所述用户情绪、所述用户属性特征及所述用户行为特征构建所述目标用户的用户画像，对所述用户画像进行聚类处理，得到用户聚类标签，获取产品信息，对所述产品信息进行文本筛选，得到产品标签；A label building module is used to construct a user portrait of the target user according to the user emotion, the user attribute characteristics and the user behavior characteristics, and perform clustering processing on the user portrait to obtain user clustering labels, and obtain products. information, performing text screening on the product information to obtain a product label;产品推送模块，用于对所述用户聚类标签及所述产品标签进行相似度聚类处理，得到聚类结果，对所述聚类结果进行至少两次相关性排列，并根据排列结果对所述目标用户进行产品推送。The product push module is used to perform similarity clustering processing on the user clustering tags and the product tags to obtain a clustering result, perform at least two correlation rankings on the clustering results, and rank all the clustering results according to the ranking results. Push the product to the target users.9.一种电子设备，其特征在于，所述电子设备包括：9. An electronic device, characterized in that the electronic device comprises:至少一个处理器；以及，at least one processor; and,与所述至少一个处理器通信连接的存储器；其中，a memory communicatively coupled to the at least one processor; wherein,所述存储器存储有可被所述至少一个处理器执行的计算机程序，所述计算机程序被所述至少一个处理器执行，以使所述至少一个处理器能够执行如权利要求1至7中任意一项所述的产品推送方法。The memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform any one of claims 1 to 7 The product push method described in item.10.一种计算机可读存储介质，存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至7中任意一项所述的产品推送方法。10 . A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the method for pushing a product according to any one of claims 1 to 7 is implemented. 11 .

Images