技术领域technical field
本发明涉及IPTV技术领域和推荐技术领域，具体涉及一种利用人工神经网络推荐IPTV直播频道的方法及系统。The present invention relates to the technical field of IPTV and the technical field of recommendation, in particular to a method and system for recommending IPTV live channels by using an artificial neural network.
背景技术Background technique
随着技术的不断进步，电视产业蓬勃发展，电视节目内容日渐丰富，朝着多元化的方向发展。其中，IPTV以其鲜明的交互特性深受广大用户的青睐。与此同时，问题出现了，用户享受观看节目的愉悦之后，需要在浩如烟海的频道中不断切换来选中自己感兴趣的下一个频道，长时间的操作降低了用户体验，且十分浪费资源。虽然有的系统已经设计了检索功能，但用户对自己想看的节目仍然没有一个确定的概念，仅仅局限于检索电视台提供的少数节目，检索功能得不到充分利用。因此，频道推荐成为用户的重要需求和亟待解决的问题。With the continuous advancement of technology, the TV industry is booming, and the content of TV programs is becoming more and more abundant, and it is developing in a diversified direction. Among them, IPTV is favored by the majority of users for its distinctive interactive features. At the same time, a problem arises. After enjoying the pleasure of watching a program, the user needs to constantly switch among a vast number of channels to select the next channel he is interested in. The long-term operation reduces the user experience and is a waste of resources. Although some systems have designed a search function, users still don't have a definite concept of the programs they want to watch. They are only limited to searching a few programs provided by TV stations, and the search function is not fully utilized. Therefore, channel recommendation has become an important demand of users and an urgent problem to be solved.
申请号为201610258946.7的中国发明专利公开了一种基于电视机顶盒的视频推荐系统，其中的推荐引擎混合使用了协同过滤和内容过滤，从用户的历史行为中分析其兴趣，将符合用户兴趣的视频推荐给用户。申请号为201310684807.7的中国发明专利公开的视频推荐系统及方法中，系统需要通过信息获取模块获取用户个人信息、社交网络信息等多源数据，以此分析用户的性格、情绪和喜好，从而为用户推荐其感兴趣的频道。申请号为200410083188.7的中国发明专利公开的电视频道推荐系统及推荐方法，根据记录的收看时间和频道序号来计算每个频道的推荐分值，向用户推荐分值最大的频道。申请号为201110203368.4的中国发明专利公开了一种电视频道推荐系统及方法，其中包括了前端系统和终端系统，前端系统包括用户行为采集服务器和智能分析服务器，智能分析服务器将信息进行一系列处理后得到用户行为轨迹，经过进一步处理得到汇总信息，而后将节目库中与之匹配的节目推荐给用户。但现有技术大多是针对IPTV点播频道进行推荐，且需要过多的用户属性，例如用户观看的节目信息等，适用范围较窄。The Chinese invention patent with application number 201610258946.7 discloses a video recommendation system based on a TV set-top box, in which the recommendation engine uses collaborative filtering and content filtering to analyze users' interests from their historical behavior and recommend videos that match users' interests. to the user. In the video recommendation system and method disclosed in the Chinese invention patent with the application number of 201310684807.7, the system needs to obtain multi-source data such as user personal information and social network information through the information acquisition module, so as to analyze the user's personality, emotion and preferences, so as to serve the user Recommend channels they are interested in. The Chinese invention patent application number 200410083188.7 discloses a TV channel recommendation system and recommendation method. According to the recorded viewing time and channel number, the recommendation score of each channel is calculated, and the channel with the highest score is recommended to the user. The Chinese invention patent with application number 201110203368.4 discloses a TV channel recommendation system and method, which includes a front-end system and a terminal system. The front-end system includes a user behavior collection server and an intelligent analysis server. The intelligent analysis server performs a series of processing on the information Obtain the user's behavior trajectory, obtain summary information after further processing, and then recommend the matching program in the program library to the user. However, most of the existing technologies are recommended for IPTV on-demand channels, and too many user attributes are required, such as program information watched by the user, and the scope of application is relatively narrow.
发明内容Contents of the invention
本发明的目的是针对现有技术的不足，提供了一种利用人工神经网络推荐IPTV直播频道的方法，所述方法融合了多种人工神经网络为用户进行推荐，不需要过多的用户属性，适用范围更广。The purpose of the present invention is to address the deficiencies in the prior art and provide a method for using artificial neural networks to recommend IPTV live channels. The method combines multiple artificial neural networks to recommend for users without requiring too many user attributes. The scope of application is wider.
本发明的另一目的在于提供一种利用人工神经网络推荐IPTV直播频道的系统。Another object of the present invention is to provide a system for recommending IPTV live channels using an artificial neural network.
本发明的目的可以通过如下技术方案实现：The purpose of the present invention can be achieved through the following technical solutions:
一种利用人工神经网络推荐IPTV直播频道的方法，所述方法包括以下步骤：A method utilizing an artificial neural network to recommend an IPTV live channel, said method comprising the following steps:
S1、选取滑动窗口天数ΔT，并取[t-ΔT,t]时间窗口内的数据作为原始训练数据，其中t表示训练数据的结束日期，时间t不得大于等于用户当前观看的日期，原始训练数据包含以下数据结构<设备号，进入观看时刻，观看频道，观看时长>，其中设备号不局限于机顶盒设备号；S1. Select the sliding window days ΔT, and take the data in the [t-ΔT,t] time window as the original training data, where t represents the end date of the training data, and the time t must not be greater than or equal to the date the user is currently viewing. The original training data Contains the following data structure <device number, viewing time, viewing channel, viewing duration>, where the device number is not limited to the set-top box device number;
S2、对提取的原始训练数据进行数据清洗，去除用户因为快速切换频道所产生的噪声数据，清洗后的数据能够表现该用户的行为特征；S2. Perform data cleaning on the extracted original training data to remove the noise data generated by the user due to fast channel switching, and the cleaned data can represent the user's behavioral characteristics;
S3、对清洗后的数据进行训练，针对每个设备号得到各自设备的训练模型，每个设备在t+1天使用与其对应的训练模型进行预测，设备实时采集用户当前观看频道的信息，并将该信息送入已训练的模型中进行预测，为相应设备进行推送；S3. Perform training on the cleaned data, and obtain the training model of each device for each device number. Each device uses its corresponding training model to make predictions on day t+1. The device collects the information of the user's current viewing channel in real time, and Send this information into the trained model for prediction, and push it to the corresponding device;
S4、在当前日期结束时，将原始训练数据窗口结束日期设置为t+1，重复步骤S1。S4. At the end of the current date, set the end date of the original training data window to t+1, and repeat step S1.
进一步地，步骤S2的具体过程为：将原始训练数据中用户观看时长小于10秒及大于3小时的记录进行删除，去除原始训练数据中用户观看时的无关属性，保留的观看属性为<设备号，当前观看频道，下一观看频道，日期>，表示设备号在该日期内所观看的频道及观看该频道后的下一频道，日期由原始训练数据中进入观看的时刻得来，且清洗后的数据按照用户的观看时间的升序进行排列。Further, the specific process of step S2 is: delete the records in the original training data that the user watched for less than 10 seconds and longer than 3 hours, remove the irrelevant attributes of the user's viewing in the original training data, and keep the viewing attributes <device number , the current viewing channel, the next viewing channel, the date>, indicating the channel that the device number watched within the date and the next channel after watching this channel, the date is obtained from the moment of viewing in the original training data, and after cleaning The data for is sorted in ascending order of users' watch time.
进一步地，步骤S3中对清洗后的数据进行训练能够采取三种方法：循环神经网络推荐法、反向神经网络推荐法或多网络冷热频道混合推荐法。Further, three methods can be adopted for training the cleaned data in step S3: recurrent neural network recommendation method, reverse neural network recommendation method or multi-network mixed hot and cold channel recommendation method.
进一步地，步骤S3中对清洗后的数据进行训练采用的是循环神经网络推荐法，具体过程为：Further, in step S3, the training of the cleaned data adopts the recurrent neural network recommendation method, and the specific process is as follows:
S3.1、将清洗后的数据根据每台设备号进行划分，并得到每个用户的观看序列α＝{C1，C2，…，Cn}，观看序列按照用户的观看时间先后进行排列，Ci表示该设备在训练窗口ΔT内第i次观看的频道；S3.1. Divide the cleaned data according to each device number, and obtain each user's viewing sequence α={C1 , C2 ,...,Cn }, and the viewing sequence is arranged according to the user's viewing time , Ci represents the channel that the device watches for the i-th time within the training window ΔT;
S3.2、对每个用户的观看序列进行重构，即将观看序列α从C1到Cm按照序列长度为m向后滑动，其中C1至Cm构成第一个序列，Cm+1为该序列的标签，最终将观看序列α分割为n-m+1个长度为m的序列，且每个序列的标签为该序列末尾频道的下一频道，切分后的每个序列内按照该用户的观看次序进行排列；S3.2. Reconstruct the viewing sequence of each user, that is, slide the viewing sequence α backwards from C1 to Cm according to the sequence length m, where C1 to Cm constitute the first sequence, and Cm+1 is the label of the sequence, and finally divide the viewing sequence α into n-m+1 sequences of length m, and the label of each sequence is the next channel of the channel at the end of the sequence. In each sequence after segmentation, according to Arrange the viewing order of the user;
S3.3、按照用户，将每个序列输入至循环神经网络中进行训练，最终得到每个序列的预测模型；S3.3. According to the user, input each sequence into the cyclic neural network for training, and finally obtain the prediction model of each sequence;
S3.4、设备在t+1天实时采集用户的观看信息，包括以下结构：<设备号，当前频道，日期>，将当前频道以序列长度为m输入至已训练好的该设备预测模型当中，并将预测模型输出的频道号向对应设备进行推送。S3.4. The device collects the user's viewing information in real time on day t+1, including the following structure: <device number, current channel, date>, input the current channel into the trained prediction model of the device with a sequence length of m , and push the channel number output by the prediction model to the corresponding device.
进一步地，步骤S3中对清洗后的数据进行训练采用的是反向神经网络推荐法，具体过程为：Further, in step S3, the cleaned data is trained using the reverse neural network recommendation method, and the specific process is as follows:
S3.1、将清洗后的数据根据每台设备号进行划分，并得到每个用户的观看序列α＝{C1，C2，…，Cn},观看序列按照用户的观看时间先后进行排列，Ci表示该设备在训练窗口ΔT内第i次观看的频道；S3.1. Divide the cleaned data according to each device number, and obtain each user's viewing sequence α={C1 , C2 ,...,Cn }, and the viewing sequence is arranged according to the user's viewing time , Ci represents the channel that the device watches for the i-th time within the training window ΔT;
S3.2、训练数据集的构造，为每个用户的观看序列进行标签标注，其中频道Ci的标签为Ci+1，Ci+1为用户U观看频道Ci的下一个频道，针对每个用户的观看序列α＝{C1，C2，…，Cn}得出标签序列L＝{C2，C3…，Cn+1}，即观看序列α中的频道C1对应标签为C2，频道Cn对应标签为Cn+1，若Cn为当天观看的最后一个频道，则标签Cn+1设为0，表示关机；S3.2. The construction of the training data set, labeling the viewing sequence of each user, where the label of the channel Ci is Ci+1 , and Ci+1 is the next channel of the channel Ci watched by the user U, for Each user’s viewing sequence α={C1 , C2 ,…,Cn } gives the label sequence L={C2 ,C3 …,Cn+1 }, that is, the channel C1 in the viewing sequence α corresponds to The label is C2 , and the corresponding label of channel Cn is Cn+1 , if Cn is the last channel watched that day, then the label Cn+1 is set to 0, indicating shutdown;
S3.3、将每个用户的观看序列α及其对应的标签序列L输入进反向神经网络当中，得到针对每个用户的训练模型；S3.3. Input each user's viewing sequence α and its corresponding label sequence L into the reverse neural network to obtain a training model for each user;
S3.4、设备在t+1天实时采集用户的观看信息，包括以下结构：<设备号，当前频道>，按照设备号将当前频道输入至已训练好的该设备预测模型当中，并将模型输出的频道号向对应设备进行推送。S3.4. The device collects the user's viewing information in real time on day t+1, including the following structure: <device number, current channel>, input the current channel into the trained prediction model of the device according to the device number, and input the model The output channel number is pushed to the corresponding device.
进一步地，步骤S3中对清洗后的数据进行训练采用的是多网络冷热频道混合推荐法，具体过程为：Further, in step S3, the cleaned data is trained using a multi-network hot and cold channel mixed recommendation method, and the specific process is as follows:
S3.1、将清洗后的数据按照设备号进行划分，并统计每个用户在时间窗口[t-ΔT,t]内每个频道的观看频率，设定冷门频道阈值为ρ％，观看频率小于等于ρ％的频道认定为冷门频道，得到冷门频道集合Coldi＝{C1，C2……,Cx}，Coldi表示用户i的冷门频道集合，观看频率大于ρ％的频道认定为热门频道，得到热门频道集合Hoti＝{C1，C2……,Cy}，Hoti表示用户i的热门频道集合；S3.1. Divide the cleaned data according to the device number, and count the viewing frequency of each channel for each user in the time window [t-ΔT,t], set the threshold of unpopular channels as ρ%, and the viewing frequency is less than Channels equal to ρ% are identified as unpopular channels, and the set of unpopular channels Coldi = {C1 , C2 ..., Cx }, Coldi represents the set of unpopular channels of user i, and channels whose viewing frequency is greater than ρ% are identified as popular channel, get the hot channel set Hoti = {C1 , C2 ..., Cy }, Hoti represents the hot channel set of user i;
S3.2、训练每个用户的冷门频道预测模型和热门频道预测模型，即针对每个用户的冷门频道，采用反向神经网络法，得到每个用户的冷门频道预测模型；针对每个用户的热门频道，采用循环神经网络法，得到每个用户的热门频道预测模型；S3.2. Train the unpopular channel prediction model and popular channel prediction model for each user, that is, for each user’s unpopular channel, adopt the reverse neural network method to obtain each user’s unpopular channel prediction model; for each user’s For popular channels, use the cyclic neural network method to obtain the popular channel prediction model for each user;
S3.3、设备在t+1天实时采集用户观看信息，包括以下结构：<设备号，当前频道，日期>，将当前信息输入到已经训练好的该设备模型中，若当前观看的频道在该用户的冷门频道集合Coldi内，则使用该用户的冷门频道预测模型进行预测；若当前观看的频道在该用户的热门频道集合Hoti内，则使用该用户的热门频道预测模型进行预测；将预测出的前N个频道向相应用户进行推送。S3.3. The device collects user viewing information in real time on day t+1, including the following structure: <device number, current channel, date>, input the current information into the trained device model, if the currently watched channel is in In the user's unpopular channel collection Coldi , then use the user's unpopular channel prediction model to predict; if the currently watched channel is in the user's popular channel collection Hoti , then use the user's popular channel prediction model to predict; Push the predicted top N channels to corresponding users.
进一步地，所述方法采用缓存器存储每次为每个用户推荐的频道，若用户为开机用户，则为该用户推荐存储在缓存器中最近一次推荐的频道。Further, the method uses a buffer to store the channel recommended for each user each time, and if the user is a power-on user, recommends the last recommended channel stored in the buffer for the user.
本发明的另一目的可以通过如下技术方案实现：Another object of the present invention can be achieved through the following technical solutions:
一种利用人工神经网络推荐IPTV直播频道的系统，包括数据清洗模块、推荐模块和推送模块，原始训练数据在数据清洗模块中进行数据清洗后，通过推荐模块进行推荐，其中，推荐模块包括区分每个用户冷热频道的冷热频道区分器、对清洗后的原始训练数据进行训练的训练器和对每个用户的热门频道进行实时预测的预测器，推荐结果送入推荐模块为用户进行推送。A system for recommending IPTV live broadcast channels using an artificial neural network, including a data cleaning module, a recommendation module, and a push module. After the original training data is cleaned in the data cleaning module, it is recommended through the recommendation module. A hot and cold channel distinguisher for each user's hot and cold channels, a trainer for training the cleaned original training data, and a predictor for real-time prediction of each user's popular channels, and the recommendation results are sent to the recommendation module for users to push.
本发明与现有技术相比，具有如下优点和有益效果：Compared with the prior art, the present invention has the following advantages and beneficial effects:
本发明所述的利用人工神经网络推荐IPTV直播频道的方法，仅仅需要获取用户的频道观看记录，不需要其他视频数据和描述，就可以通过长短期记忆网络建立训练模型，实时向用户提供可能感兴趣的推荐频道列表，大大提升了用户体验；另外，本发明所提出的个性化实时频道推荐系统，利用滑动窗口不断地更新用户观看数据，从而建立新的推荐模型，可以实时向用户推荐其感兴趣的频道，提高了命中率和用户体验。The method of using the artificial neural network to recommend IPTV live channels according to the present invention only needs to obtain the user's channel viewing records without other video data and descriptions, and can establish a training model through the long-term and short-term memory network to provide users with a sense of possibility in real time. The list of recommended channels of interest greatly improves the user experience; in addition, the personalized real-time channel recommendation system proposed by the present invention uses the sliding window to continuously update the user's viewing data, thereby establishing a new recommendation model, which can recommend to users in real time. Channels of interest, improving the hit rate and user experience.
附图说明Description of drawings
图1为本发明利用人工神经网络推荐IPTV直播频道的方法的流程图。FIG. 1 is a flowchart of a method for recommending IPTV live channels using an artificial neural network according to the present invention.
图2为本发明采用循环神经网络推荐法对清洗后的数据进行训练的示意图。Fig. 2 is a schematic diagram of training the cleaned data by adopting the recurrent neural network recommendation method in the present invention.
图3为本发明采用反向神经网络推荐法对清洗后的数据进行训练的示意图。FIG. 3 is a schematic diagram of the present invention using the reverse neural network recommendation method to train the cleaned data.
图4为本发明采用多网络冷热频道混合推荐法对清洗后的数据进行训练的示意图。FIG. 4 is a schematic diagram of the present invention using the multi-network hot and cold channel mixed recommendation method to train the cleaned data.
具体实施方式Detailed ways
下面结合实施例及附图对本发明作进一步详细的描述，但本发明的实施方式不限于此。The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.
实施例1：Example 1:
本实施例提供了一种利用人工神经网络推荐IPTV直播频道的方法，所述方法的流程图如图1所示，包括以下步骤：The present embodiment provides a kind of method utilizing artificial neural network to recommend IPTV live channel, the flow chart of described method is as shown in Figure 1, comprises the following steps:
S1、选取滑动窗口天数ΔT，并取[t-ΔT,t]时间窗口内的数据作为原始训练数据，其中t表示训练数据的结束日期，时间t不得大于等于用户当前观看的日期，原始训练数据包含以下数据结构<设备号，进入观看时刻，观看频道，观看时长>，其中设备号不局限于机顶盒设备号；S1. Select the sliding window days ΔT, and take the data in the [t-ΔT,t] time window as the original training data, where t represents the end date of the training data, and the time t must not be greater than or equal to the date the user is currently viewing. The original training data Contains the following data structure <device number, viewing time, viewing channel, viewing duration>, where the device number is not limited to the set-top box device number;
S2、对提取的原始训练数据进行数据清洗，去除用户因为快速切换频道所产生的噪声数据，清洗后的数据能够表现该用户的行为特征；S2. Perform data cleaning on the extracted original training data to remove the noise data generated by the user due to fast channel switching, and the cleaned data can represent the user's behavioral characteristics;
具体过程为：将原始训练数据中用户观看时长小于10秒及大于3小时的记录进行删除，去除原始训练数据中用户观看时的无关属性，保留的观看属性为<设备号，当前观看频道，下一观看频道，日期>，表示设备号在该日期内所观看的频道及观看该频道后的下一频道，日期由原始训练数据中进入观看的时刻得来，且清洗后的数据按照用户的观看时间的升序进行排列。The specific process is: delete the records in the original training data that the user’s viewing time is less than 10 seconds and greater than 3 hours, remove the irrelevant attributes of the user’s viewing time in the original training data, and retain the viewing attributes as <device number, current viewing channel, download One watched channel, date>, indicating the channel watched by the device number within the date and the next channel after watching the channel, the date is obtained from the moment when the original training data enters the viewing, and the cleaned data is in accordance with the user's viewing Sort in ascending order of time.
S3、对清洗后的数据进行训练，针对每个设备号得到各自设备的训练模型，每个设备在t+1天使用与其对应的训练模型进行预测，设备实时采集用户当前观看频道的信息，并将该信息送入已训练的模型中进行预测，为相应设备进行推送；S3. Perform training on the cleaned data, and obtain the training model of each device for each device number. Each device uses its corresponding training model to make predictions on day t+1. The device collects the information of the user's current viewing channel in real time, and Send this information into the trained model for prediction, and push it to the corresponding device;
其中，对清洗后的数据进行训练能够采取三种方法：循环神经网络推荐法、反向神经网络推荐法或多网络冷热频道混合推荐法。Among them, three methods can be adopted for training the cleaned data: recurrent neural network recommendation method, reverse neural network recommendation method or multi-network hot and cold channel mixed recommendation method.
采用循环神经网络推荐法的示意图如图2所示，具体过程为：The schematic diagram of the recommendation method using the cyclic neural network is shown in Figure 2, and the specific process is as follows:
S3.1、将清洗后的数据根据每台设备号进行划分，并得到每个用户的观看序列α＝{C1，C2，…，Cn}，观看序列按照用户的观看时间先后进行排列，Ci表示该设备在训练窗口ΔT内第i次观看的频道；S3.1. Divide the cleaned data according to each device number, and obtain each user's viewing sequence α={C1 , C2 ,...,Cn }, and the viewing sequence is arranged according to the user's viewing time , Ci represents the channel that the device watches for the i-th time within the training window ΔT;
S3.2、对每个用户的观看序列进行重构，即将观看序列α从C1到Cm按照序列长度为m向后滑动，其中C1至Cm构成第一个序列，Cm+1为该序列的标签，最终将观看序列α分割为n-m+1个长度为m的序列，且每个序列的标签为该序列末尾频道的下一频道，切分后的每个序列内按照该用户的观看次序进行排列；S3.2. Reconstruct the viewing sequence of each user, that is, slide the viewing sequence α backwards from C1 to Cm according to the sequence length m, where C1 to Cm constitute the first sequence, and Cm+1 is the label of the sequence, and finally divide the viewing sequence α into n-m+1 sequences of length m, and the label of each sequence is the next channel of the channel at the end of the sequence. In each sequence after segmentation, according to Arrange the viewing order of the user;
S3.3、按照用户，将每个序列输入至循环神经网络中进行训练，最终得到每个序列的预测模型；S3.3. According to the user, input each sequence into the cyclic neural network for training, and finally obtain the prediction model of each sequence;
S3.4、设备在t+1天实时采集用户的观看信息，包括以下结构：<设备号，当前频道，日期>，将当前频道以序列长度为m输入至已训练好的该设备预测模型当中，并将预测模型输出的频道号向对应设备进行推送。S3.4. The device collects the user's viewing information in real time on day t+1, including the following structure: <device number, current channel, date>, input the current channel into the trained prediction model of the device with a sequence length of m , and push the channel number output by the prediction model to the corresponding device.
采用反向神经网络推荐法的示意图如图3所示，具体过程为：The schematic diagram of the reverse neural network recommendation method is shown in Figure 3, and the specific process is as follows:
S3.1、将清洗后的数据根据每台设备号进行划分，并得到每个用户的观看序列α＝{C1，C2，…，Cn},观看序列按照用户的观看时间先后进行排列，Ci表示该设备在训练窗口ΔT内第i次观看的频道；S3.1. Divide the cleaned data according to each device number, and obtain each user's viewing sequence α={C1 , C2 ,...,Cn }, and the viewing sequence is arranged according to the user's viewing time , Ci represents the channel that the device watches for the i-th time within the training window ΔT;
S3.2、训练数据集的构造，为每个用户的观看序列进行标签标注，其中频道Ci的标签为Ci+1，Ci+1为用户U观看频道Ci的下一个频道，针对每个用户的观看序列α＝{C1，C2，…，Cn}得出标签序列L＝{C2，C3…，Cn+1}，即观看序列α中的频道C1对应标签为C2，频道Cn对应标签为Cn+1，若Cn为当天观看的最后一个频道，则标签Cn+1设为0，表示关机；S3.2. The construction of the training data set, labeling the viewing sequence of each user, where the label of the channel Ci is Ci+1 , and Ci+1 is the next channel of the channel Ci watched by the user U, for Each user’s viewing sequence α={C1 , C2 ,…,Cn } gives the label sequence L={C2 ,C3 …,Cn+1 }, that is, the channel C1 in the viewing sequence α corresponds to The label is C2 , and the corresponding label of channel Cn is Cn+1 , if Cn is the last channel watched that day, then the label Cn+1 is set to 0, indicating shutdown;
S3.3、将每个用户的观看序列α及其对应的标签序列L输入进反向神经网络当中，得到针对每个用户的训练模型；S3.3. Input each user's viewing sequence α and its corresponding label sequence L into the reverse neural network to obtain a training model for each user;
S3.4、设备在t+1天实时采集用户的观看信息，包括以下结构：<设备号，当前频道>，按照设备号将当前频道输入至已训练好的该设备预测模型当中，并将模型输出的频道号向对应设备进行推送。S3.4. The device collects the user's viewing information in real time on day t+1, including the following structure: <device number, current channel>, input the current channel into the trained prediction model of the device according to the device number, and input the model The output channel number is pushed to the corresponding device.
采用多网络冷热频道混合推荐法的示意图如图4所示，具体过程为：The schematic diagram of the mixed recommendation method using multi-network hot and cold channels is shown in Figure 4, and the specific process is as follows:
S3.1、将清洗后的数据按照设备号进行划分，并统计每个用户在时间窗口[t-ΔT,t]内每个频道的观看频率，设定冷门频道阈值为ρ％，观看频率小于等于ρ％的频道认定为冷门频道，得到冷门频道集合Coldi＝{C1，C2……,Cx}，Coldi表示用户i的冷门频道集合，观看频率大于ρ％的频道认定为热门频道，得到热门频道集合Hoti＝{C1，C2……,Cy}，Hoti表示用户i的热门频道集合；S3.1. Divide the cleaned data according to the device number, and count the viewing frequency of each channel for each user in the time window [t-ΔT,t], set the threshold of unpopular channels as ρ%, and the viewing frequency is less than Channels equal to ρ% are identified as unpopular channels, and the set of unpopular channels Coldi = {C1 , C2 ..., Cx }, Coldi represents the set of unpopular channels of user i, and channels whose viewing frequency is greater than ρ% are identified as popular channel, get the hot channel set Hoti = {C1 , C2 ..., Cy }, Hoti represents the hot channel set of user i;
S3.2、训练每个用户的冷门频道预测模型和热门频道预测模型，即针对每个用户的冷门频道，采用反向神经网络法，得到每个用户的冷门频道预测模型；针对每个用户的热门频道，采用循环神经网络法，得到每个用户的热门频道预测模型；S3.2. Train the unpopular channel prediction model and popular channel prediction model for each user, that is, for each user’s unpopular channel, adopt the reverse neural network method to obtain each user’s unpopular channel prediction model; for each user’s For popular channels, use the cyclic neural network method to obtain the popular channel prediction model for each user;
S3.3、设备在t+1天实时采集用户观看信息，包括以下结构：<设备号，当前频道，日期>，将当前信息输入到已经训练好的该设备模型中，若当前观看的频道在该用户的冷门频道集合Coldi内，则使用该用户的冷门频道预测模型进行预测；若当前观看的频道在该用户的热门频道集合Hoti内，则使用该用户的热门频道预测模型进行预测；将预测出的前N个频道向相应用户进行推送。S3.3. The device collects user viewing information in real time on day t+1, including the following structure: <device number, current channel, date>, input the current information into the trained device model, if the currently watched channel is in In the user's unpopular channel collection Coldi , then use the user's unpopular channel prediction model to predict; if the currently watched channel is in the user's popular channel collection Hoti , then use the user's popular channel prediction model to predict; Push the predicted top N channels to corresponding users.
S4、在当前日期结束时，将原始训练数据窗口结束日期设置为t+1，重复步骤S1。S4. At the end of the current date, set the end date of the original training data window to t+1, and repeat step S1.
实施例2：Example 2:
本实施例提供了一种实施所述利用人工神经网络推荐IPTV直播频道的方法的系统，包括数据清洗模块(101)、推荐模块(102)和推送模块(103)，原始训练数据在数据清洗模块(101)中进行数据清洗后，通过推荐模块(102)进行推荐，其中，推荐模块(102)包括区分每个用户冷热频道的冷热频道区分器(201)、对清洗后的原始训练数据进行训练的训练器(202)和对每个用户的热门频道进行实时预测的预测器(203)，推荐结果送入推荐模块(102)为用户进行推送。The present embodiment provides a kind of system that implements the method for using artificial neural network to recommend IPTV live channels, including data cleaning module (101), recommendation module (102) and push module (103), original training data in data cleaning module After cleaning the data in (101), it is recommended by the recommendation module (102), wherein the recommendation module (102) includes a hot and cold channel distinguisher (201) for distinguishing each user's hot and cold channels, and the original training data after cleaning A trainer (202) for training and a predictor (203) for real-time prediction of each user's popular channel, the recommendation result is sent to the recommendation module (102) for the user to push.
以上所述，仅为本发明专利较佳的实施例，但本发明专利的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明专利所公开的范围内，根据本发明专利的技术方案及其发明专利构思加以等同替换或改变，都属于本发明专利的保护范围。The above is only a preferred embodiment of the patent of the present invention, but the scope of protection of the patent of the present invention is not limited thereto. The equivalent replacement or change of the technical solution and its invention patent concept all belong to the protection scope of the invention patent.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810157400.1ACN108347652B (en) | 2018-02-24 | 2018-02-24 | Method and system for recommending IPTV live broadcast channel by using artificial neural network |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810157400.1ACN108347652B (en) | 2018-02-24 | 2018-02-24 | Method and system for recommending IPTV live broadcast channel by using artificial neural network |
| Publication Number | Publication Date |
|---|---|
| CN108347652Atrue CN108347652A (en) | 2018-07-31 |
| CN108347652B CN108347652B (en) | 2020-01-14 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810157400.1AActiveCN108347652B (en) | 2018-02-24 | 2018-02-24 | Method and system for recommending IPTV live broadcast channel by using artificial neural network |
| Country | Link |
|---|---|
| CN (1) | CN108347652B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110225361A (en)* | 2019-06-03 | 2019-09-10 | 武汉瓯越网视有限公司 | Direct broadcasting room recommended method, storage medium, electronic equipment and system |
| CN113343097A (en)* | 2021-06-24 | 2021-09-03 | 中山大学 | Sequence recommendation method and system based on fragment and self-attention mechanism |
| CN113986954A (en)* | 2021-12-30 | 2022-01-28 | 深圳市明源云科技有限公司 | User event acquisition method and device, intelligent terminal and readable storage medium |
| CN114285508A (en)* | 2021-12-22 | 2022-04-05 | 展讯通信(天津)有限公司 | Frequency modulation channel playing method and device, storage medium and terminal equipment |
| CN114528434A (en)* | 2022-01-19 | 2022-05-24 | 华南理工大学 | IPTV live channel fusion recommendation method based on self-attention mechanism |
| CN116320511A (en)* | 2023-02-03 | 2023-06-23 | 华南理工大学 | Cross-domain fusion recommendation method based on graph convolution network |
| CN116614667A (en)* | 2023-04-27 | 2023-08-18 | 福建新大陆通信科技股份有限公司 | A method and system for channel switching of DVB set-top box based on neural network |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020174429A1 (en)* | 2001-03-29 | 2002-11-21 | Srinivas Gutta | Methods and apparatus for generating recommendation scores |
| CN102510529A (en)* | 2011-09-22 | 2012-06-20 | 中国科学技术大学 | Method for performing on-demand play quantity prediction and memory scheduling on programs |
| CN103491441A (en)* | 2013-09-09 | 2014-01-01 | 东软集团股份有限公司 | Recommendation method and system of live television programs |
| CN104539981A (en)* | 2014-11-26 | 2015-04-22 | 四川长虹电器股份有限公司 | Real-time hot television channel recommending system and method |
| US20150178265A1 (en)* | 2013-12-20 | 2015-06-25 | Google Inc. | Content Recommendation System using a Neural Network Language Model |
| CN105138541A (en)* | 2015-07-08 | 2015-12-09 | 腾讯科技（深圳）有限公司 | Audio fingerprint matching query method and device |
| CN105791913A (en)* | 2016-04-27 | 2016-07-20 | 青岛海信传媒网络技术有限公司 | Multi-user telephone configuration information setting method, multi-user telephone configuration information setting system and cloud server |
| CN106101839A (en)* | 2016-06-20 | 2016-11-09 | 徐汕 | A kind of method identifying that television user gathers |
| US20170041670A1 (en)* | 2015-08-03 | 2017-02-09 | At&T Intellectual Property I, L.P. | Cross-platform analysis |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020174429A1 (en)* | 2001-03-29 | 2002-11-21 | Srinivas Gutta | Methods and apparatus for generating recommendation scores |
| CN102510529A (en)* | 2011-09-22 | 2012-06-20 | 中国科学技术大学 | Method for performing on-demand play quantity prediction and memory scheduling on programs |
| CN103491441A (en)* | 2013-09-09 | 2014-01-01 | 东软集团股份有限公司 | Recommendation method and system of live television programs |
| US20150178265A1 (en)* | 2013-12-20 | 2015-06-25 | Google Inc. | Content Recommendation System using a Neural Network Language Model |
| CN104539981A (en)* | 2014-11-26 | 2015-04-22 | 四川长虹电器股份有限公司 | Real-time hot television channel recommending system and method |
| CN105138541A (en)* | 2015-07-08 | 2015-12-09 | 腾讯科技（深圳）有限公司 | Audio fingerprint matching query method and device |
| US20170041670A1 (en)* | 2015-08-03 | 2017-02-09 | At&T Intellectual Property I, L.P. | Cross-platform analysis |
| CN105791913A (en)* | 2016-04-27 | 2016-07-20 | 青岛海信传媒网络技术有限公司 | Multi-user telephone configuration information setting method, multi-user telephone configuration information setting system and cloud server |
| CN106101839A (en)* | 2016-06-20 | 2016-11-09 | 徐汕 | A kind of method identifying that television user gathers |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110225361A (en)* | 2019-06-03 | 2019-09-10 | 武汉瓯越网视有限公司 | Direct broadcasting room recommended method, storage medium, electronic equipment and system |
| CN110225361B (en)* | 2019-06-03 | 2021-10-15 | 武汉瓯越网视有限公司 | Live broadcast room recommendation method, storage medium, electronic device and system |
| CN113343097A (en)* | 2021-06-24 | 2021-09-03 | 中山大学 | Sequence recommendation method and system based on fragment and self-attention mechanism |
| CN113343097B (en)* | 2021-06-24 | 2023-01-13 | 中山大学 | Sequence recommendation method and system based on segment and self-attention mechanism |
| CN114285508A (en)* | 2021-12-22 | 2022-04-05 | 展讯通信(天津)有限公司 | Frequency modulation channel playing method and device, storage medium and terminal equipment |
| CN113986954A (en)* | 2021-12-30 | 2022-01-28 | 深圳市明源云科技有限公司 | User event acquisition method and device, intelligent terminal and readable storage medium |
| CN114528434A (en)* | 2022-01-19 | 2022-05-24 | 华南理工大学 | IPTV live channel fusion recommendation method based on self-attention mechanism |
| CN116320511A (en)* | 2023-02-03 | 2023-06-23 | 华南理工大学 | Cross-domain fusion recommendation method based on graph convolution network |
| CN116320511B (en)* | 2023-02-03 | 2024-10-11 | 华南理工大学 | A cross-domain fusion recommendation method based on graph convolutional network |
| CN116614667A (en)* | 2023-04-27 | 2023-08-18 | 福建新大陆通信科技股份有限公司 | A method and system for channel switching of DVB set-top box based on neural network |
| Publication number | Publication date |
|---|---|
| CN108347652B (en) | 2020-01-14 |
| Publication | Publication Date | Title |
|---|---|---|
| CN108347652B (en) | Method and system for recommending IPTV live broadcast channel by using artificial neural network | |
| CA2924065C (en) | Content based video content segmentation | |
| CN103747343B (en) | The method and apparatus that resource is recommended at times | |
| CN103491441B (en) | Recommendation method and system of live television programs | |
| CN107197368B (en) | Determine user to the method and system of multimedia content degree of concern | |
| CN106156360B (en) | A kind of application method of multimedia player | |
| CN102917269B (en) | A kind of television program recommendation system and method | |
| WO2020007470A1 (en) | Dynamic viewer prediction system for advertisement scheduling | |
| CN102265276A (en) | Context-based recommender system | |
| CN105187941B (en) | A kind of television terminal and control method of intelligent management favorites | |
| CN102473174A (en) | A method and arrangements for analysing multimedia content | |
| CN103686375A (en) | Video sharing method and device | |
| CN101764661A (en) | Data fusion based video program recommendation system | |
| CN101833552A (en) | Method for marking and recommending streaming media | |
| CN110493654A (en) | The recommendation of video and playback method and device in a kind of list of videos | |
| CN114339417B (en) | Video recommendation method, terminal equipment and readable storage medium | |
| CN110177306A (en) | Video broadcasting method, device, mobile terminal and medium based on mobile terminal | |
| CN111385606A (en) | Video preview method and device and intelligent terminal | |
| CN111062527B (en) | Video traffic collection prediction method and device | |
| CN102695086B (en) | Content pushing method and device for interactive network protocol television | |
| Yu et al. | Follow me: Personalized IPTV channel switching guide | |
| CN112784069B (en) | IPTV content intelligent recommendation system and method | |
| CN102890680A (en) | Man-machine interaction operation platform of intelligent associative network | |
| Hromic et al. | Event panning in a stream of big data | |
| Thawani et al. | Viewing characteristics based personalized ad streaming in an interactive TV environment |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |