技术领域technical field
本申请涉及计算机技术领域，具体涉及一种在线系统个性化推荐的方法；相应于上述方法，本申请同时涉及一种在线个性化推荐的装置以及电子设备，以及一种在线系统个性化推荐的系统。This application relates to the field of computer technology, in particular to a method for personalized recommendation in an online system; corresponding to the above method, this application also relates to a device and electronic equipment for personalized recommendation online, and a system for personalized recommendation in an online system .
背景技术Background technique
互联网技术的迅猛发展使人类进入了信息爆炸的时代。海量信息的同时呈现，一方面使信息获得者很难从中发现自己感兴趣的部分，另一方面也使得大量少人问津的信息无法被一般用户获取，以上情况严重阻碍了信息效用的充分发挥。The rapid development of Internet technology has brought mankind into the era of information explosion. The simultaneous presentation of massive amounts of information, on the one hand, makes it difficult for information acquirers to find the part they are interested in, and on the other hand, it also makes it impossible for ordinary users to obtain a large amount of information that few people care about. The above situation seriously hinders the full use of information utility.
为解决上述问题，目前出现了个性化推荐系统。个性化推荐系统通过建立用户与信息之间的二元关系，利用已有的选择过程或相似性关系挖掘每个用户潜在感兴趣的个性化信息，进而进行个性化推荐，使信息与信息使用者之间具有更高的匹配度。In order to solve the above problems, a personalized recommendation system has emerged. The personalized recommendation system establishes the binary relationship between users and information, uses the existing selection process or similarity relationship to mine the personalized information that each user is potentially interested in, and then performs personalized recommendations, so that information and information users have a higher matching degree.
现有技术下的个性化推荐系统，采用如下方式实现。The personalized recommendation system in the prior art is implemented in the following manner.
首先预先大量搜集用户的行为，提炼出样本数据，根据这些样本数据，通过机器学习的方法，训练得到用户模型，在线根据每个用户的用户特征数据以及上述用户模型，为用户推荐个性化信息，在在线交易系统中，所述个性化信息对应于个性化推荐的实体，一般包括商品、店铺或者品牌等；例如，亚马逊网站根据用户特征数据推荐书籍，就是个性化推荐实体的例子。First, a large number of user behaviors are collected in advance, and sample data is extracted. Based on these sample data, user models are trained through machine learning methods, and personalized information is recommended for users online based on the user characteristic data of each user and the above user models. In an online transaction system, the personalized information corresponds to a personalized recommended entity, generally including commodities, stores, or brands; for example, Amazon.com recommends books based on user characteristic data, which is an example of a personalized recommended entity.
在上述现有技术下，需要通过不同系统搜集大量数据，并将采集到的原始数据，汇总到离线平台。例如，将用户曝光日志、点击日志、成交数据、购物车行为、收藏夹数据等多个不同应用的数据，汇总到离线平台。然后，根据这些原始数据计算所有用户的特征以及特征数据，再根据这些数据中记录的用户反馈行为，如：点击、访问、点赞、收藏、预定、购买等，对样本进行打标，最后在离线大数据平台中进行模型训练。训练完毕，将获得的用户模型上线使用。Under the above existing technologies, it is necessary to collect a large amount of data through different systems, and aggregate the collected raw data to an offline platform. For example, data from multiple different applications, such as user exposure logs, click logs, transaction data, shopping cart behavior, favorites data, etc., are aggregated to the offline platform. Then, calculate the characteristics and characteristic data of all users based on these raw data, and then mark the samples according to the user feedback behaviors recorded in these data, such as: click, visit, like, favorite, reservation, purchase, etc., and finally in the Model training is performed on an offline big data platform. After the training is complete, the obtained user model will be used online.
上述现有技术存在明显缺陷，主要问题在于，需要从多个平台回流数据，然后汇总、计算，这个过程需要耗费较多的计算资源和存储资源；此外，由于各个平台回流数据的效率和时点不一，造成模型训练需要等待较长时间，使训练模型的实时性差。而且，由于一次需要积累的数据量比较大，一般只能使用所搜集的一部分数据记录，很多数据反映的情况无法反映到用户模型中，无法真正落实大数据应用。The above-mentioned existing technologies have obvious defects. The main problem is that data needs to be returned from multiple platforms, and then aggregated and calculated. This process requires a lot of computing resources and storage resources; Inconsistencies cause the model training to wait for a long time, making the real-time performance of the training model poor. Moreover, due to the relatively large amount of data that needs to be accumulated at one time, generally only a part of the collected data records can be used, and many situations reflected by the data cannot be reflected in the user model, and big data applications cannot be truly implemented.
发明内容Contents of the invention
本申请提供一种在线系统个性化推荐的方法，以解决现有技术下资源消耗过多，实时性差的问题，并真正落实大数据应用。本申请还提供一种在线个性化推荐系统，以及在线系统个性化推荐装置；以及一种实现在线系统个性化推荐的电子设备。This application provides a method for personalized recommendation in an online system to solve the problems of excessive resource consumption and poor real-time performance in the prior art, and to truly implement big data applications. The present application also provides an online personalized recommendation system, an online system personalized recommendation device, and an electronic device for realizing online system personalized recommendation.
本申请提供一种在线系统个性化推荐的方法，包括：This application provides a method for personalized recommendation in an online system, including:
接收访问用户的访问请求，并提取所述访问用户的用户特征数据；receiving an access request from an access user, and extracting user characteristic data of the access user;
根据所述访问用户的用户特征数据，以及当前用户模型，提供个性化推荐信息；Provide personalized recommendation information according to the user characteristic data of the visiting user and the current user model;
实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户作出反馈操作的所述个性化推荐信息以及所做的反馈操作，并形成新增用户模型训练条目；Collect the user characteristic data of the visiting user, the personalized recommendation information provided for the visiting user, the personalized recommendation information and the feedback operation made by the visiting user in real time, and form a newly added User model training items;
将所述新增用户模型训练条目加入训练样本集；adding the newly added user model training item to the training sample set;
以当前训练样本集进行用户模型训练；Use the current training sample set for user model training;
所述用户模型训练完成后，将所获得的更新的用户模型作为当前用户模型。After the training of the user model is completed, the obtained updated user model is used as the current user model.
可选的，所述用户特征数据至少包括：用户身份标识，并包括下列任何用户特征数据中的至少一个：性别，年龄，交易记录。Optionally, the user feature data includes at least: a user identity and at least one of any of the following user feature data: gender, age, and transaction records.
可选的，所述个性化推荐信息包括下列个性化推荐实体的至少一种：商品、店铺、品牌。Optionally, the personalized recommendation information includes at least one of the following personalized recommendation entities: commodities, stores, and brands.
可选的，所述访问用户作出反馈操作包括如下操作之一：点击，访问，点赞，收藏，预定，购买。Optionally, the feedback operation performed by the visiting user includes one of the following operations: click, visit, like, bookmark, reserve, purchase.
可选的，所述训练数据条目中还包括以下数据的一种或者两种：获取该记录的时间点、资源位信息。Optionally, the training data entry further includes one or both of the following data: the time point at which the record was obtained, and resource bit information.
可选的，将积累的所述新增用户模型训练条目加入训练样本集之后，所述以当前训练样本集进行用户模型训练的步骤之前，执行下述步骤：Optionally, after adding the accumulated new user model training items to the training sample set, before the step of using the current training sample set to perform user model training, perform the following steps:
判断所述新增用户模型训练样本是否达到预定的阈值；若是，则进入下一步骤。Judging whether the newly added user model training sample reaches a predetermined threshold; if so, proceed to the next step.
可选的，所述用户模型训练采用机器学习方法。Optionally, the training of the user model adopts a machine learning method.
可选的，所述机器学习方法采用逻辑回归方法或者梯度提升决策树方法。Optionally, the machine learning method adopts a logistic regression method or a gradient boosting decision tree method.
可选的，所述新增用户模型训练条目采用日志方式记录。Optionally, the newly added user model training entry is recorded in a log mode.
相应的，本申请还提供一种在线系统个性化推荐的装置，包括：Correspondingly, the present application also provides an online system personalized recommendation device, including:
用户特征数据提取单元，用于接收访问用户的访问请求，并提取所述访问用户的用户特征数据；A user feature data extraction unit, configured to receive an access request from an access user, and extract user feature data of the access user;
个性化推荐信息提供单元，用于根据所述访问用户的用户特征数据，以及当前用户模型，提供个性化推荐信息；A personalized recommendation information providing unit, configured to provide personalized recommendation information according to the user characteristic data of the visiting user and the current user model;
用户模型训练条目形成单元，用于实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户作出反馈操作的所述个性化推荐信息以及所做的反馈操作，并形成新增用户模型训练条目；A user model training item formation unit, configured to collect in real time the user characteristic data of the visiting user, the personalized recommendation information provided for the visiting user, the personalized recommendation information of the visiting user performing feedback operations, and all Feedback operations done, and form new user model training items;
训练样本集搜集单元，用于将所述新增用户模型训练条目加入训练样本集；A training sample set collection unit, configured to add the newly added user model training item to the training sample set;
用户模型训练单元，用于以当前训练样本集进行用户模型训练；A user model training unit, configured to perform user model training with the current training sample set;
用户模型更新单元，用于在所述用户模型训练完成后，将所获得的更新的用户模型作为当前用户模型。The user model updating unit is configured to use the obtained updated user model as the current user model after the training of the user model is completed.
相应的，本申请还提供一种电子设备，包括：Correspondingly, the present application also provides an electronic device, including:
显示器；monitor;
处理器；processor;
存储器，用于存储实现在线系统个性化推荐的方法的程序，该设备通电并运行该在线系统个性化推荐的方法的程序后，执行下述步骤：The memory is used to store the program for implementing the method for personalized recommendation in the online system, and after the device is powered on and runs the program for the method for personalized recommendation in the online system, the following steps are performed:
接收访问用户的访问请求，并提取所述访问用户的用户特征数据；receiving an access request from an access user, and extracting user characteristic data of the access user;
根据所述访问用户的用户特征数据，以及当前用户模型，提供个性化推荐信息；Provide personalized recommendation information according to the user characteristic data of the visiting user and the current user model;
实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户作出反馈操作的所述个性化推荐信息以及所做的反馈操作，并将为访问用户的一次访问-推荐过程形成的上述内容组合为一条新增的用户模型训练条目；Collect the user characteristic data of the visiting user, the personalized recommendation information provided for the visiting user, the personalized recommendation information and the feedback operation made by the visiting user in real time, and provide information for the visiting user The above content formed by a user's visit-recommendation process is combined into a new user model training item;
将所述新增的用户模型训练条目加入训练样本集；adding the newly added user model training item to the training sample set;
以当前训练样本集进行用户模型训练；Use the current training sample set for user model training;
所述用户模型训练完成后，将所获得的更新的用户模型作为当前用户模型。After the training of the user model is completed, the obtained updated user model is used as the current user model.
相应的，本申请还提供一种在线个性化推荐的系统，包括：在线子系统、离线子系统；Correspondingly, this application also provides an online personalized recommendation system, including: an online subsystem and an offline subsystem;
所述在线子系统，用于接收用户的访问请求，并提取用户特征数据，并根据所提取的用户特征数据以及当前用户模型，向提出访问请求的用户提供个性化推荐实体；以及，实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户对所述个性化推荐信息的反馈信息，形成新增用户模型训练条目并发送；以及，接收所述离线子系统提供的更新的用户模型；The online subsystem is configured to receive a user's access request, extract user characteristic data, and provide a personalized recommendation entity to the user who made the access request according to the extracted user characteristic data and the current user model; The user feature data of the visiting user, the personalized recommendation information provided for the visiting user, the feedback information of the visiting user to the personalized recommendation information, form a new user model training item and send it; and, receive an updated user model provided by the offline subsystem;
所述离线子系统，接收所述在线子系统发送的新增用户模型训练条目，并将所述新增用户模型训练条目加入当前训练样本集；采用所述当前训练样本集进行用户模型训练，将训练完成获得的更新的用户模型向外发送。The offline subsystem receives the newly added user model training entry sent by the online subsystem, and adds the newly added user model training entry to the current training sample set; uses the current training sample set to perform user model training, and The updated user model obtained after training is sent out.
与现有技术相比，本申请提供的在线系统个性化推荐的方法，对用户特征数据、个性化推荐信息和对所述用户作出反馈操作的所述个性化推荐信息进行实时搜集，并形成新增用户模型训练条目，通过积累所述用户模型训练条目形成的训练样本集，训练所述用户模型，并在新的用户模型形成后，立即作为当前用户模型，用于向用户提供个性化推荐信息。Compared with the prior art, the online system personalized recommendation method provided by this application collects user characteristic data, personalized recommendation information and the personalized recommendation information for feedback operations to the user in real time, and forms a new Add user model training items, train the user model by accumulating the training sample set formed by the user model training items, and immediately use it as the current user model after the new user model is formed to provide personalized recommendation information to the user .
使用本申请提供的方法，在向用户提供个性化推荐信息的同时，进行信息搜集并形成用户模型训练条目；这种处理方式，可以有效减少信息汇总产生的计算量，并且减少了存储中间信息的存储空间；因此，本申请提供的方法可以有效减少资源消耗。Using the method provided by this application, while providing personalized recommendation information to users, information is collected and user model training items are formed; this processing method can effectively reduce the amount of calculation generated by information aggregation, and reduce the storage of intermediate information. storage space; therefore, the method provided by this application can effectively reduce resource consumption.
在本申请提供的方法中，用户模型训练条目随时可以用于进行用户模型训练，并且新形成的用户模型可以马上在向用户提供个性化推荐信息时使用。这样，可以随时根据新搜集的数据对用户模型进行调整，并迅速用于向用户提供个性化推荐信息；因此，本申请提供的方法能够及时根据所搜集的数据对用户模型进行调整，比现有技术具有更高实时性。此外，本申请提供的方法，还能够充分利用所搜集到的全部数据，有效实现大数据应用。In the method provided in this application, the user model training items can be used for user model training at any time, and the newly formed user model can be used immediately when providing personalized recommendation information to users. In this way, the user model can be adjusted at any time according to the newly collected data, and quickly used to provide personalized recommendation information to the user; Technology is more real-time. In addition, the method provided in this application can also make full use of all the collected data and effectively implement big data applications.
附图说明Description of drawings
图1是本申请第一实施例提供的一种在线系统个性化推荐的方法流程图；FIG. 1 is a flow chart of a method for personalized recommendation in an online system provided in the first embodiment of the present application;
图2是本申请第二实施例提供的一种在线系统个性化推荐的装置的单元框图；FIG. 2 is a unit block diagram of an online system personalized recommendation device provided in the second embodiment of the present application;
图3是本申请第四实施例提供的一种在线系统个性化推荐的系统示意图。Fig. 3 is a system schematic diagram of an online system personalized recommendation provided by the fourth embodiment of the present application.
具体实施方式detailed description
在下面的描述中阐述了很多具体细节以便于充分理解本申请。但是，本申请能够以很多不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本申请内涵的情况下做类似推广，因此，本申请不受下面公开的具体实施的限制。In the following description, numerous specific details are set forth in order to provide a thorough understanding of the application. However, the present application can be implemented in many other ways different from those described here, and those skilled in the art can make similar promotions without violating the connotation of the present application. Therefore, the present application is not limited by the specific implementations disclosed below.
本申请提供了一种在线系统个性化推荐的方法和装置，以及一种在线个性化推荐的系统；以及在线系统个性化推荐的电子设备，在下面的实施例中逐一进行详细说明。The present application provides a method and device for online system personalized recommendation, a system for online personalized recommendation, and an electronic device for online system personalized recommendation, which will be described in detail in the following embodiments.
为了便于理解本申请的技术方案，首先对提出本申请的背景及本申请的技术方案作简要说明。In order to facilitate the understanding of the technical solution of the present application, first, the background of the application and the technical solution of the present application are briefly explained.
参照图1，其示出了本申请第一实施例提供的一种在线系统个性化推荐的方法处理流程图。Referring to FIG. 1 , it shows a process flowchart of a method for personalized recommendation in an online system provided by the first embodiment of the present application.
在该实施例中，系统根据用户特征数据，以及当前用户模型，提供个性化推荐实体；并且，会记录本次推荐相关的用户特征数据、为用户提供的所述个性化推荐信息，以及用户作出反馈操作的情况，并将其作为训练新的用户模型的数据。In this embodiment, the system provides a personalized recommendation entity according to the user characteristic data and the current user model; and, it will record the user characteristic data related to this recommendation, the personalized recommendation information provided for the user, and the user's actions. Feedback the situation of the operation and use it as data for training a new user model.
例如：当用户在购买一部手机时，系统将根据用户特征数据，推荐适合用户需要的耳机，手机套等商品；用户对某款耳机作出购买操作；上述过程可能形成多条用户模型训练条目，其中一条新增用户模型训练条目，包括如下数据：该用户的用户特征数据、向用户推荐的耳机、以及该用户将对耳机作出购买操作的打标记录，这条用户模型训练条目作为正样本；另外一条新增用户模型训练条目包括如下数据：该用户的用户特征数据，向用户推荐的手机套，以及该用户未对手机套作出操作的打标记录，该条用户模型训练条目可以作为负样本使用。For example: when a user buys a mobile phone, the system will recommend earphones, mobile phone cases and other products suitable for the user's needs according to the user's characteristic data; the user makes a purchase operation on a certain earphone; the above process may form multiple user model training items, One of the new user model training items includes the following data: the user's user characteristic data, the earphones recommended to the user, and the marking record of the user's purchase operation on the earphones. This user model training item is used as a positive sample; Another new user model training entry includes the following data: the user's user characteristic data, the mobile phone case recommended to the user, and the marking record that the user did not operate the mobile phone case. This user model training entry can be used as a negative sample use.
以下结合图1对本实施例提供的一种个性化推荐系统方法进行说明，并且对该方法的各个步骤进行说明。A personalized recommendation system method provided by this embodiment is described below with reference to FIG. 1 , and each step of the method is described.
步骤S101,接收访问用户的访问请求，并提取所述访问用户的用户特征数据。Step S101, receiving an access request from an access user, and extracting user characteristic data of the access user.
所谓访问用户，主要指通过因特网或者移动网在线访问特定站点、网页、服务器的用户。The so-called visiting users mainly refer to users who access specific sites, webpages, and servers online through the Internet or mobile networks.
所述访问用户的用户特征数据，是指反映访问用户本身的各方面特征信息的数据。这些数据可以通过不同渠道获得。The user characteristic data of the visiting user refers to data reflecting various characteristic information of the visiting user itself. These data can be obtained through different sources.
首先，访问用户一般通过浏览器或者APP应用作为中介，实现对特定站点、网页、服务器的访问。登陆这些特定站点、网页或者服务器一般需要通过账户登陆，对应这些账户会记录有用户相关的特征数据。First of all, visiting users generally use browsers or APPs as intermediaries to access specific sites, web pages, and servers. Logging in to these specific sites, webpages, or servers generally requires logging in through an account, and user-related characteristic data will be recorded corresponding to these accounts.
其次，用户特征数据也可以通过用户访问方式、用户访问区域等不需要用户登录账户的方式获得。例如，所述访问用户的访问方式特征包括：互联网访问还是移动互联网访问；移动互联网访问中，还可以进一步判断使用的访问终端的品牌；所述用户访问区域，即LBS信息，即用户进行访问时所在的地域，这些地域可以根据具体地点划分为不同类型，如高校、商务办公地点；如城市、乡镇，每一种类型均反映用户可能具有不同的影响其反馈操作行为的特征。Secondly, user characteristic data can also be obtained through methods such as user access methods and user access areas that do not require users to log in to their accounts. For example, the access mode characteristics of the accessing user include: Internet access or mobile Internet access; in mobile Internet access, the brand of the access terminal used can be further judged; the user access area, that is, LBS information, that is, when the user visits Geography, these regions can be divided into different types according to the specific location, such as colleges and universities, business office locations; such as cities, towns, and each type reflects that users may have different characteristics that affect their feedback operation behavior.
所述用户特征数据，从获得方式划分，包括用户自己提供的数据，以及根据用户登陆访问行为获得的相关数据。The user characteristic data, divided from the way of obtaining, includes the data provided by the user himself and the relevant data obtained according to the user's login and access behavior.
所述用户自己提供的数据，包括用户登陆时提供的用户身份标识，以及注册时提供的性别，年龄等信息；也可以包括用户在使用网站、服务器中关联的邮箱、银行账号等信息。The data provided by the user itself includes the user identity provided by the user when logging in, and information such as gender and age provided when registering; it may also include the user's email address, bank account number and other information associated with the website or server.
所述根据用户登陆访问行为获得的相关数据，包括直接获得的数据和间接获得的数据。The relevant data obtained according to the user's login and access behavior includes directly obtained data and indirectly obtained data.
所述直接获得的数据包括：用户对购物网站的访问、下单情况获得的用户访问记录、用户购买记录以及其他各种用户行为信息，这些信息是对用户历史行为的直接记录。The directly obtained data includes: user visits to shopping websites, user access records obtained from order placement, user purchase records, and various other user behavior information, which are direct records of user historical behavior.
所述间接获得的数据，主要是从上述直接获得的数据中总结出的反映用户特征的数据，例如，根据用户购买的书籍，对用户的文化水平、阅读领域作出的归类等。根据用户购买的商品品牌，对用户的购买力水平做出的判断等。The indirectly obtained data is mainly the data that reflects the characteristics of the user summarized from the above directly obtained data, for example, according to the books purchased by the user, the classification of the user's education level and reading field, etc. According to the product brand purchased by the user, the judgment made on the purchasing power level of the user, etc.
在不同的应用场景下，所述用户特征数据的具体会包含不同具体项目，但不论何种情况，至少包括用户身份标识(ID),用户身份标识可以直接通过用户登陆时提供的信息获得，并可以作为进一步查询获得用户的其它信息记录的依据；例如，用户的性别，年龄,以及交易记录(在网络购物的场景下)等，这些信息如前所述，一般以用户身份标识作为主关键字存储在数据库的记录表中。记录用户特征数据的数据表或者数据库既可以在远程的服务器上，也可能在客户端。In different application scenarios, the user characteristic data will contain different specific items, but in any case, at least including the user identity (ID), the user identity can be obtained directly from the information provided by the user when logging in, and It can be used as a basis for further query to obtain other information records of the user; for example, the user's gender, age, and transaction records (in the case of online shopping), etc., such information, as mentioned above, generally uses the user identity as the primary key stored in the record table of the database. The data table or database that records user characteristic data can be on a remote server or on a client.
以上说明获得用户特征数据的各种渠道和获得方式，在具体实施例中，根据不同的需求，需要根据具体情况搜集所需要的用户特征数据，这些用户特征数据应当是对用户模型训练有用的数据。The above describes the various channels and methods of obtaining user characteristic data. In specific embodiments, according to different needs, it is necessary to collect the required user characteristic data according to specific situations. These user characteristic data should be useful data for user model training. .
一般而言，所述用户特征数据至少包括用户身份标识,并包括下列任何用户特征数据中的至少一个：性别，年龄,交易记录；当然，用户特征数据完全可以包含其他可能相关的数据。随着大数据技术的发展以及数据挖掘技术的进步，越来越多的数据相关性被发现，并且有越来越好的数据模型可以反映各种数据对最终希望提供的个性化推荐信息的作用，因此，可以用于用户模型训练的用户特征数据的类型也会越来越多。Generally speaking, the user characteristic data includes at least the user identity, and at least one of the following user characteristic data: gender, age, transaction records; of course, the user characteristic data can completely contain other possibly related data. With the development of big data technology and the advancement of data mining technology, more and more data correlations have been discovered, and there are better and better data models that can reflect the role of various data on the final personalized recommendation information that we hope to provide , therefore, there will be more and more types of user feature data that can be used for user model training.
本步骤所称的访问请求，可以包括对特定网站、网页的访问浏览，也可以指用户在网站中对商品等的搜索。The access request referred to in this step may include access and browsing of a specific website or web page, or may refer to a user's search for commodities on a website.
步骤S102，根据所述访问用户的用户特征数据，以及当前用户模型，提供个性化推荐信息。Step S102, providing personalized recommendation information according to the user characteristic data of the visiting user and the current user model.
通过前述步骤S101，获得了访问用户的用户特征数据；这些用户特征数据能够作为为用户提供个性化推荐信息的依据；具体获得个性化推荐信息需要通过使用当前用户模型，以所述用户特征数据为依据，带入当前用户模型，推导出所述个性化推荐信息。Through the aforementioned step S101, the user characteristic data of the visiting user are obtained; these user characteristic data can be used as the basis for providing personalized recommendation information for the user; specifically obtaining personalized recommendation information needs to use the current user model, taking the user characteristic data as the basis Based on the current user model, the personalized recommendation information is derived.
所述当前用户模型，是根据所述访问用户访问前搜集的各个用户的用户特征数据，以预定的方式进行训练获得的当前正在使用的用户模型。当前用户模型的作用是，向其提供所需要的特定的用户特征数据后，当前用户模型能够根据用户特征数据反映的用户特性，向该用户提供相应的个性化推荐信息。The current user model is a user model currently in use obtained by training in a predetermined manner according to the user characteristic data of each user collected by the visiting user before the visit. The role of the current user model is to provide the user with corresponding personalized recommendation information according to the user characteristics reflected by the user characteristic data after providing the required specific user characteristic data.
在本实施例中，通过不断搜集数据，能够不断积累新的训练数据，这些训练数据可以用于对所述用户模型的训练，从而获得更新的用户模型。本步骤使用当前用户模型一词，其中“当前”即说明用户模型在本实施例中是不断修正、演变的，本步骤使用的用户模型为当前有效的用户模型，但过一段时间，该用户模型可能就已经由于新的训练过程而有所变化。In this embodiment, by continuously collecting data, new training data can be continuously accumulated, and these training data can be used for training the user model, so as to obtain an updated user model. This step uses the word current user model, where "current" means that the user model is constantly revised and evolved in this embodiment. The user model used in this step is the currently valid user model, but after a period of time, the user model It may have changed due to the new training process.
所述个性化推荐信息，是指所述当前用户模型根据特定访问用户的用户特征数据推算出的向该特定访问用户推荐的信息。所述推荐，主要指在该访问用户的访问界面展示，包括视觉、听觉或者其它可能方式的展示；所述个性化，其含义是指与该用户的用户特征相匹配。The personalized recommendation information refers to the information recommended to a specific visiting user calculated by the current user model according to the user characteristic data of the specific visiting user. The recommendation mainly refers to displaying on the accessing user's access interface, including visual, auditory or other possible display; the personalization means matching with the user's user characteristics.
所述个性化推荐信息根据具体情况可以对应不同类型的个性化推荐实体，在本实施例中，假定场景为电商的销售场景，则所述个性化推荐信息可以是如下类型的个性化推荐实体：商品、店铺或者品牌。The personalized recommendation information may correspond to different types of personalized recommendation entities according to specific circumstances. In this embodiment, assuming that the scenario is an e-commerce sales scenario, the personalized recommendation information may be the following types of personalized recommendation entities : commodity, store or brand.
例如，某位访问用户正在访问网上书店，根据其用户特征数据反映的该用户的兴趣范围，可以为其推荐相关的书籍商品；某访问用户访问淘宝网站，搜索笔记本，根据其用户特征数据反映的该用户的价格承受能力或者品牌喜好，可以为其推荐符合其用户特征并出售笔记本的网上店铺；某访问用户正在查询购买服装，根据其用户特征反映的该用户的性别、年龄、以及购买能力，可以为其推荐提供符合其性别、年龄以及消费层次的服装品牌。For example, if a visiting user is visiting an online bookstore, relevant book products can be recommended according to the user's interest range reflected in the user's characteristic data; According to the user's price tolerance or brand preference, an online store that matches the user's characteristics and sells notebooks can be recommended; if a visiting user is inquiring about buying clothes, according to the user's gender, age, and purchasing ability reflected by the user's characteristics, It can recommend clothing brands that match their gender, age, and consumption level.
所述当前用户模型可以包括一个或者多个子模型，分别负责从不同角度为访问用户进行不同方面信息的个性化推荐；并且可以最终在同一个访问界面展现，当然，也可以根据用户访问的具体情况展现使用特定的当前用户模型的子模型，例如，一个已经注册过的用户通过账户登录后，立刻根据其用户特征数据，向其展现其上次登录后新出现的该用户可能关心的商品、店铺或者品牌；或者，用户搜索书籍时，立刻使用负责推荐书籍的当前用户模型的子模型根据用户特征数据为用户推荐适当的书籍。这些子模型使用的具体的用户特征数据，可能分别是该访问用户的所有用户特征数据的一个特定部分。这些不同的子模型也可以视为同一个当前用户模型的不同功能单元。The current user model may include one or more sub-models, which are respectively responsible for making personalized recommendations for different aspects of information for accessing users from different perspectives; and can be finally displayed on the same access interface, of course, it can also be based on the specific circumstances of user access Show a sub-model that uses a specific current user model. For example, after a registered user logs in through an account, immediately based on the user's characteristic data, it will show the new products and stores that the user may care about after the last login. or a brand; or, when a user searches for books, immediately use the sub-model of the current user model responsible for recommending books to recommend appropriate books to the user based on user characteristic data. The specific user characteristic data used by these sub-models may respectively be a specific part of all user characteristic data of the visiting user. These different sub-models can also be regarded as different functional units of the same current user model.
步骤S103，实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户作出反馈操作的所述个性化推荐信息以及所做的反馈操作，并形成新增用户模型训练条目。Step S103, collecting the user feature data of the visiting user, the personalized recommendation information provided for the visiting user, the personalized recommendation information and the feedback operation performed by the visiting user in real time, and Create a new user model training entry.
本步骤用于搜集用户访问、向用户提供个性化推荐信息以及用户对个性化推荐信息进行反馈这一完整的访问-推荐过程形成的所有相关内容，包括：个性化推荐的原始依据，即用户特征数据；为所述访问用户提供的个性化推荐信息；对个性化推荐的反馈，即所述访问用户对何种个性化推荐信息作出了反馈以及具体作出何种反馈。最终将这些内容组合形成数据记录。一次访问-推荐过程形成的所述数据记录可以采用不同的具体形式记录。例如，可以记录为一个完整的数据包，发送到后台，由后台服务器根据模型训练的需求从中抽取出相关数据形成多条新增用户模型训练条目。当然，也可以直接形成新增用户模型训练条目并发送。总之，需要根据用户模型训练的需要以适当的方式搜集和解析一次访问-推荐过程形成的数据记录。This step is used to collect all relevant content formed in the complete visit-recommendation process of user access, providing personalized recommendation information to users, and user feedback on personalized recommendation information, including: the original basis of personalized recommendation, namely user characteristics Data; personalized recommendation information provided for the visiting user; feedback on the personalized recommendation, that is, what kind of personalized recommendation information the visiting user has given feedback on and what kind of feedback has been given. Finally, these contents are combined to form data records. The data records formed during a visit-recommendation process may be recorded in different specific forms. For example, it can be recorded as a complete data packet and sent to the background, and the background server will extract relevant data from it according to the requirements of model training to form multiple new user model training entries. Of course, it is also possible to directly form a new user model training item and send it. In short, it is necessary to collect and analyze the data records formed by a visit-recommendation process in an appropriate manner according to the needs of user model training.
所述新增用户模型训练条目，作为后续的用户模型训练的训练素材。根据需要，所搜集的信息还可以进一步包括获取该记录的时间点、以及资源位信息，所述获取该记录的时间点，即记录形成该新增的用户模型训练条目的信息形成时间；所述资源位信息，即记录相关信息所在的位置，例如，某个具体的个性化推荐信息在浏览器的顶部或者下部，或者侧栏，等等，这些信息对于用户的反馈行为也有重要影响。The newly added user model training item is used as training material for subsequent user model training. According to needs, the collected information may further include the time point of obtaining the record and resource bit information, the time point of obtaining the record, that is, the information formation time of the record forming the newly added user model training item; Resource location information, that is, the location where relevant information is recorded, for example, a specific personalized recommendation information is located at the top or bottom of the browser, or on the sidebar, etc. This information also has an important impact on the user's feedback behavior.
实现本步骤，需要在供访问用户使用的浏览器或者app中植入程序，该程序随时搜集所有向用户提供的个性化推荐信息以及用户对个性化推荐信息的反馈操作，将这些信息和先前作为个性化推荐信息已经获得的用户特征数据对应，并通过网络传送到训练用户模型的服务器一端。具体使用的植入程序可以根据不同环境采用不同的类型，例如，采用Javascript脚本程序在浏览器中对通过浏览器提供的各种信息以及通过浏览器界面发生的各种操作进行记录。To achieve this step, a program needs to be implanted in the browser or app used by the visiting user. The program collects all personalized recommendation information provided to the user and the user's feedback on the personalized recommendation information at any time, and combines this information with the previous information as The personalized recommendation information corresponds to the obtained user characteristic data, and is transmitted to the server side for training the user model through the network. The specific implant program used can be of different types according to different environments, for example, Javascript script programs are used to record various information provided through the browser and various operations occurring through the browser interface in the browser.
所述访问用户的反馈操作，包括访问用户对所述个性化推荐信息作出的各种方式的反馈，例如：点击，访问，点赞，收藏，预定，购买等操作。这些操作反应了访问用户对不同的个性化推荐信息的关心程度的差异，这些差异对于调整向用户推荐的个性化信息有重要意义。The feedback operation of the visiting user includes various feedbacks made by the visiting user on the personalized recommendation information, such as operations such as clicking, visiting, liking, bookmarking, booking, and purchasing. These operations reflect the differences in the degree of concern of visiting users to different personalized recommendation information, and these differences are of great significance for adjusting the personalized information recommended to users.
记录所述访问用户作出反馈操作的所述个性化推荐信息的方法，可以是在这些信息上加上特定标识，即，在这些个性化推荐信息的记录上加上表示对其发生了反馈操作的标识，一般称为打标；例如，在记录表格中对应每个性化推荐信息设置有标识是否在其上发生了反馈操作的字段，这个字段也可以直接记录发生的反馈操作是什么，通过这种记录方式，就在记录个性化推荐信息的同时对访问用户的反馈进行了记录。The method of recording the personalized recommendation information of the feedback operation performed by the accessing user may be to add a specific mark to the information, that is, add a record indicating that a feedback operation has occurred to the records of the personalized recommendation information. Identification, generally called marking; for example, in the record form, corresponding to each personalized recommendation information, there is a field identifying whether a feedback operation has occurred on it. This field can also directly record what the feedback operation occurred. Through this In the recording method, the user's feedback is recorded while recording the personalized recommendation information.
所述将为访问用户的一次访问-推荐过程形成的上述内容形成新增用户模型训练条目，具体可以采用多种方式，一种比较简单的优选方式是，将这些内容以固定的格式形成日志记录，并随时以日志方式发送，该日志的接收方根据日志格式的规定，将其还原为表格形式或其他数据存储形式，并根据用户模型训练的需要进行解析，抽取出一个或者若干个用户模型训练条目，该用户模型训练条目中记录的用户特征数据，可以作为模型训练的输入数据，该用户模型训练条目记录的个性化推荐信息中，该访问用户没有做出反馈的，可以作为用户训练的负样本，该访问用户做出反馈的，可以作为用户训练的正样本。The above-mentioned content that will be formed for a visit-recommendation process of the visiting user will form a new user model training item. Specifically, various methods can be used. A relatively simple and preferred method is to form a log record of these contents in a fixed format , and send it as a log at any time. The receiver of the log restores it to a form or other data storage form according to the log format, and analyzes it according to the needs of user model training, and extracts one or several user model training data. Item, the user feature data recorded in the user model training item can be used as the input data for model training, and among the personalized recommendation information recorded in the user model training item, if the visiting user has not given feedback, it can be used as the negative for user training. The sample, which is the feedback given by the visiting user, can be used as a positive sample for user training.
步骤S104，将所述新增用户模型训练条目加入训练样本集。Step S104, adding the newly added user model training item into the training sample set.
本步骤中，将前一步骤形成的新增用户模型训练条目加入训练样本集。所谓训练样本集，即用于用户模型训练的样本数据的集合；对用户模型进行训练需要搜集大量的用户模型训练条目，从大量的样本数据中才能实现对用户模型的有意义的调整，这些训练样本不断积累，并不断用于用户模型训练，则最终训练获得的用户模型就会更为精确。In this step, the newly added user model training items formed in the previous step are added to the training sample set. The so-called training sample set is a collection of sample data used for user model training; training a user model requires the collection of a large number of user model training items, and meaningful adjustments to the user model can only be achieved from a large number of sample data. As samples are continuously accumulated and used for user model training, the user model obtained through final training will be more accurate.
在实际实现中，新增用户模型训练条目是通过每一次访问用户的访问过程中形成的所有访问相关数据，一次次累计起来，通过不断积累用户模型训练条目，事实上获得了大数据的效果。In the actual implementation, the newly added user model training items are accumulated through all the access-related data formed during each visit to the user. By continuously accumulating user model training items, the effect of big data is actually obtained.
所述训练样本集，存储在服务器一端的训练样本集数据库中；该训练样本集根据模型训练的要求，一般会采用特定的记录格式；所述将新增的用户模型训练条目加入训练样本集的过程，具体包括对原始条目的解析过程，最终将其中的有效数据以符合要求的格式存入训练样本集所在的数据库。The training sample set is stored in the training sample set database on the server side; the training sample set generally adopts a specific record format according to the requirements of model training; the addition of the newly added user model training entry to the training sample set process, specifically including the process of parsing the original entries, and finally storing the valid data in a format that meets the requirements into the database where the training sample set is located.
例如，上一步骤形成的新增用户模型训练条目采用日志形式记录，则将所述日志发送到服务器一端，服务器一端以预定的日志格式为依据，对该条日志中记录的内容进行解析，将该用户模型训练条目中的内容按照训练数据需要的形成用户模型训练条目，并记载到训练样本集数据库的相应位置。For example, if the newly added user model training entry formed in the previous step is recorded in the form of a log, the log will be sent to the server side, and the server side will analyze the content recorded in the log based on the predetermined log format. The content in the user model training entry forms the user model training entry according to the training data requirements, and records it in the corresponding position of the training sample set database.
步骤S105，以当前训练样本集进行用户模型训练。Step S105, perform user model training with the current training sample set.
所述当前训练样本集，是指在启动用户模型训练时搜集到的所有有效的训练样本组成的数据集合。The current training sample set refers to a data set composed of all valid training samples collected when the user model training is started.
在本申请提供的技术方案中，训练数据通过每次用户访问过程不断积累，其训练样本集的内容是一个动态累计的过程。因此，每次启动用户模型训练时，其训练样本集都会发生变化，即样本数据会增加；使用更多的样本数据进行模型训练，则模型的精确度也会提升。当然，训练样本集也可以设定数据淘汰的机制，例如，可以将积累实践超过一定时间的样本数据淘汰；如某些用户购物的相关样本数据以及是一年前的，已经无法准确反映用户的情况，则这些数据可以淘汰掉。In the technical solution provided by this application, the training data is continuously accumulated through each user access process, and the content of the training sample set is a dynamic accumulation process. Therefore, every time user model training is started, its training sample set will change, that is, the sample data will increase; if more sample data is used for model training, the accuracy of the model will also increase. Of course, the training sample set can also set a data elimination mechanism. For example, the sample data that has been accumulated for more than a certain period of time can be eliminated; for example, the relevant sample data of some users’ shopping is a year ago, which can no longer accurately reflect the user’s experience. In this case, these data can be eliminated.
所述用户模型训练样本用于用户模型训练，可以使用将新增的用户模型训练对原有的用户模型进行调整，即进行增量训练，也可以是将新增的用户模型训练样本加入原有的用户模型训练样本中，形成一个全量的用户模型训练样本，并对用户模型从头进行训练，即全量训练。The user model training samples are used for user model training. The newly added user model training can be used to adjust the original user model, that is, incremental training, or the newly added user model training samples can be added to the original user model. In the user model training samples, a full amount of user model training samples is formed, and the user model is trained from scratch, that is, full training.
由于启动用户模型训练后，训练过程需要较长时间，并且占用计算资源，因此，在通常情况下，一般不会每次出现新增用户模型训练样本都启动一次模型训练，而是在符合一定条件的情况下才启动所述模型训练。即在本步骤前，增加一个判断所述新增的用户模型训练样本是否达到预定的阈值的判断过程，若是，则进入本步骤，否则，暂时不进入本步骤。Since user model training starts, the training process takes a long time and takes up computing resources. Therefore, under normal circumstances, it is generally not necessary to start model training every time a new user model training sample appears, but when certain conditions are met. The model training is only started under the condition of That is, before this step, add a judging process for judging whether the newly added user model training samples reach a predetermined threshold, if so, enter this step, otherwise, temporarily skip this step.
所述一定条件，可以包括积累数量条件以及积累时间条件，可以考虑上述两个条件任意一者，也可以同时考虑两个阈值条件。The certain condition may include an accumulation quantity condition and an accumulation time condition, and any one of the above two conditions may be considered, or both threshold conditions may be considered at the same time.
例如，在一个不断产生交易的电商平台中，可以确定适当的时间阈值，例如24小时，当该时间阈值到达后，则启动新一轮用户模型训练，将该段时间内新增的用户模型训练样本用于这一轮训练中。For example, in an e-commerce platform that continuously generates transactions, an appropriate time threshold can be determined, such as 24 hours. When the time threshold is reached, a new round of user model training is started, and the newly added user model within this period The training samples are used in this round of training.
再如，在一个交易比较少的拍卖平台中，可以确定适当的数量阈值，当新增的用户模型训练样本的条目积累数量超过该阈值时，则启动新一轮的用户模型训练，并将新增的用户模型训练样本用于这一轮训练中。For another example, in an auction platform with relatively few transactions, an appropriate quantity threshold can be determined. When the cumulative number of entries of the newly added user model training samples exceeds the threshold, a new round of user model training will be started and the new The increased user model training samples are used in this round of training.
类似的，对于某个平台，可以同时设置积累时间阈值和积累数量阈值，两者都达到条件，则启动用户模型训练；或者，两者任意一个达到条件，则启动用户模型训练。Similarly, for a certain platform, the accumulation time threshold and the accumulation quantity threshold can be set at the same time. If both meet the conditions, user model training will be started; or if either of the two conditions is met, user model training will be started.
所述用户模型训练，在现有技术下可以采用多种方式实现，主要是通过机器学习方法实现用户模型训练，具体采用的机器学习方法例如可以是逻辑回归方法或者梯度提升决策树方法。以下以逻辑回归方法对用户模型训练过程做简要说明。The user model training can be implemented in various ways in the prior art, mainly through the machine learning method to realize the user model training, and the specific machine learning method can be, for example, a logistic regression method or a gradient boosting decision tree method. The following is a brief description of the user model training process using the logistic regression method.
所述逻辑回归方法包括：训练数据收集，特征提取，特征筛选，模型训练。The logistic regression method includes: training data collection, feature extraction, feature screening, and model training.
所述训练数据收集即前述步骤形成的用户模型训练样本集。The training data collection is the user model training sample set formed in the preceding steps.
所述特征提取，即根据上述训练样本集中的数据，收集与拟合目标相关的各种数据。The feature extraction is to collect various data related to the fitting target according to the data in the above training sample set.
所述特征筛选，用相关性衡量方法衡量特征(本实施例中来自用户特征数据)与拟合目标之间的相关性程度，并过滤相关性小于给定阈值的特征；例如，本实施例用于服装电商，可以根据训练样本中的已知数据，衡量用户的特定年龄、性别特征与推荐服装品牌之间的相关性，若相关性达到预设的阈值要求，则将年龄，性别特征作为与服装品牌相关的特征。The feature screening is to measure the degree of correlation between the feature (from user feature data in this embodiment) and the fitting target with a correlation measurement method, and filter the features whose correlation is less than a given threshold; for example, this embodiment uses For clothing e-commerce, based on the known data in the training samples, the correlation between the user's specific age and gender characteristics and the recommended clothing brand can be measured. If the correlation reaches the preset threshold requirements, the age and gender characteristics will be used as Features associated with clothing brands.
所述模型训练阶段，基于前述用户模型训练数据，对具有相关性的用户数据拟合回归模型，使得根据回归模型得到的预测值与训练数据中的目标值的差距最小。即：通过不断调整相关参数，使拟合获得的预测值(即本实施例中的个性化推荐信息)与用户模型训练数据中的目标值(即用户做出反馈的个性化推荐信息)尽可能一致。In the model training phase, based on the aforementioned user model training data, a regression model is fitted to relevant user data, so that the difference between the predicted value obtained according to the regression model and the target value in the training data is the smallest. That is, by continuously adjusting relevant parameters, the predicted value obtained by fitting (that is, the personalized recommendation information in this embodiment) and the target value in the user model training data (that is, the personalized recommendation information that the user makes feedback) are as close as possible unanimous.
除了上述逻辑回归方法外，还可以采用梯度提升决策树方法。In addition to the logistic regression method described above, a gradient boosting decision tree method can also be employed.
上述用户模型训练方法均采用现有技术下成熟的算法，不属于本发明的独创性部分，在此不予以详细描述。The above user model training methods all use mature algorithms in the prior art, which do not belong to the original part of the present invention and will not be described in detail here.
步骤S106，所述用户模型训练完成后，将所获得的更新的用户模型作为当前用户模型。Step S106, after the training of the user model is completed, the obtained updated user model is used as the current user model.
在前一步骤完成用户模型训练后，所获得的用户模型相比原先使用的用户模型，就是更新的用户模型；该更新的用户模型可以马上替代原先的用户模型作为当前用户模型使用。当执行所述步骤S102时，所述当前用户模型就是经过本次更新的用户模型。After the user model training is completed in the previous step, the obtained user model is an updated user model compared with the originally used user model; the updated user model can immediately replace the original user model and be used as the current user model. When the step S102 is executed, the current user model is the updated user model.
采用本方法，可以一边积累数据一边进行模型更新，并在模型更新后立刻投入使用，继而在新的用户模型下积累新的用户模型训练数据，并周而复始，实现数据积累和用户模型的相互之间的正向改善循环，迅速提高用户模型质量，提升个性化推荐信息的精确度。With this method, the model can be updated while accumulating data, and put into use immediately after the model is updated, and then accumulate new user model training data under the new user model, and go round and round to realize the interaction between data accumulation and user models The positive improvement cycle can rapidly improve the quality of user models and improve the accuracy of personalized recommendation information.
上述第一实施例中，涉及到作为访问用户的访问界面的客户端，与进行用户模型训练的服务器；两者之间通过网络联系。所述客户端提供访问用户的身份数据等给所述服务器，所述服务器根据这些数据查询存储的用户特征数据并依据用户特征数据形成个性化推荐信息，并将个性化推荐信息发送到客户端供展示；所述客户端搜集访问用户访问-推荐过程中获得的用户特征数据、个性化推荐信息和反馈操作等信息，形成一个新增用户模型训练条目，并发送给服务器，服务器根据不断积累的用户模型训练条目，在适当时机进行用户模型训练，生成更新的当前用户模型，并开始使用当前用户模型。In the above-mentioned first embodiment, the client as the access interface of the accessing user and the server for training the user model are involved; the two are connected through a network. The client provides the identity data of the visiting user to the server, and the server queries the stored user characteristic data according to these data and forms personalized recommendation information according to the user characteristic data, and sends the personalized recommendation information to the client for Display; the client collects information such as user characteristic data, personalized recommendation information, and feedback operations obtained during the user access-recommendation process to form a new user model training item and send it to the server. Model training entry, perform user model training at an appropriate time, generate an updated current user model, and start using the current user model.
所述步骤S103的用户模型训练条目的形成过程，可以由客户端完成，也可以是在后台由服务器实时搜集每次访问用户的访问过程产生的数据，这些数据中，个性化推荐信息在发送的同时进行搜集，用户特征数据可以在客户端发送访问用户的身份标识时，查询用户数据库获得，访问用户对个性化推荐信息的反馈，来自客户端提供的相关情况；最终，在服务器一侧将某一次访问-推荐过程形成的上述所有数据集合为一个数据条目，并发送给存储训练样本集的数据库存储。The formation process of the user model training items in step S103 can be completed by the client, or in the background, the server can collect in real time the data generated during the visit process of each visiting user. Among these data, the personalized recommendation information is sent At the same time, the user feature data can be obtained by querying the user database when the client sends the identity of the accessing user. The feedback of the accessing user on the personalized recommendation information comes from the relevant information provided by the client; finally, on the server side, a certain All the above-mentioned data sets formed by a visit-recommendation process are a data entry, and are sent to the database storage for storing the training sample set.
本申请第二实施例提供一种实现在线个性化推荐的装置；请参看图2。The second embodiment of the present application provides an apparatus for realizing online personalized recommendation; please refer to FIG. 2 .
本实施例提供的在线系统个性化推荐的装置，包括：用户特征数据提取单元201、个性化推荐信息提供单元202、用户模型训练条目形成单元203、训练样本集搜集单元204、用户模型训练单元205、用户模型更新单元206。The device for personalized recommendation in an online system provided in this embodiment includes: a user characteristic data extraction unit 201, a personalized recommendation information provision unit 202, a user model training item formation unit 203, a training sample set collection unit 204, and a user model training unit 205 , a user model updating unit 206 .
所述用户特征数据提取单元201，用于接收访问用户的访问请求，并提取所述访问用户的用户特征数据。所述用户特征数据至少包括：用户身份标识,并包括下列任何用户特征数据中的至少一个：性别，年龄,交易记录。The user feature data extraction unit 201 is configured to receive an access request from an access user, and extract user feature data of the access user. The user feature data at least include: user identity, and include at least one of any of the following user feature data: gender, age, transaction records.
所述个性化推荐信息提供单元202，用于根据所述访问用户的用户特征数据，以及当前用户模型，提供个性化推荐信息。所述个性化推荐信息包括下列个性化推荐实体的至少一种：商品、店铺、品牌。The personalized recommendation information providing unit 202 is configured to provide personalized recommendation information according to the user characteristic data of the visiting user and the current user model. The personalized recommendation information includes at least one of the following personalized recommendation entities: commodities, stores, and brands.
所述用户模型训练条目形成单元203，用于实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户作出反馈操作的所述个性化推荐信息以及所做的反馈操作，并形成新增用户模型训练条目。所述访问用户作出反馈操作包括如下操作之一：点击，访问，点赞，收藏，预定，购买。所述训练数据条目中还可包括以下数据的一种或者两种：获取该记录的时间点、资源位信息。所述用户模型训练条目优选采用日志方式记录。The user model training item formation unit 203 is configured to collect the user characteristic data of the visiting user in real time, the personalized recommendation information provided for the visiting user, and the personalized recommendation for the visiting user to make a feedback operation Information and the feedback operations done, and form a new user model training item. The feedback operation performed by the visiting user includes one of the following operations: clicking, visiting, liking, bookmarking, booking, and purchasing. The training data entry may also include one or both of the following data: the time point at which the record was obtained, and resource bit information. The user model training entries are preferably recorded in a log mode.
所述训练样本集搜集单元204，用于将所述新增用户模型训练条目加入训练样本集。The training sample set collecting unit 204 is configured to add the newly added user model training item into the training sample set.
所述用户模型训练单元205，用于以当前训练样本集进行用户模型训练。所述用户模型训练采用机器学习方法；具体的机器学习方法例如可以采用逻辑回归方法或者梯度提升决策树方法。The user model training unit 205 is configured to perform user model training with the current training sample set. The user model training adopts a machine learning method; a specific machine learning method may, for example, adopt a logistic regression method or a gradient boosting decision tree method.
所述用户模型更新单元206，用于在所述用户模型训练完成后，将所获得的更新的用户模型作为当前用户模型。The user model updating unit 206 is configured to use the obtained updated user model as the current user model after the training of the user model is completed.
在优选实施方案中，该在线系统个性化推荐的装置还包括阈值判断单用，用于判断所述新增的用户模型训练样本是否达到预定的阈值；若是，则启动所述用户模型训练单元205进行用户模型训练。In a preferred embodiment, the device for personalized recommendation in the online system also includes a threshold judgment unit, which is used to judge whether the newly added user model training sample reaches a predetermined threshold; if so, start the user model training unit 205 Conduct user model training.
本申请第三实施例提供一种电子设备，所述电子设备包括：The third embodiment of the present application provides an electronic device, and the electronic device includes:
显示器；monitor;
处理器；processor;
存储器，用于存储实现在线系统个性化推荐的方法的程序，该设备通电并运行该在线系统个性化推荐的方法的程序后，执行下述步骤：The memory is used to store the program for implementing the method for personalized recommendation in the online system, and after the device is powered on and runs the program for the method for personalized recommendation in the online system, the following steps are performed:
接收访问用户的访问请求，并提取所述访问用户的用户特征数据；receiving an access request from an access user, and extracting user characteristic data of the access user;
根据所述访问用户的用户特征数据，以及当前用户模型，提供个性化推荐信息；Provide personalized recommendation information according to the user characteristic data of the visiting user and the current user model;
实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户作出反馈操作的所述个性化推荐信息以及所做的反馈操作，并将为访问用户的一次访问-推荐过程形成的上述内容组合为一条新增的用户模型训练条目；Collect the user characteristic data of the visiting user, the personalized recommendation information provided for the visiting user, the personalized recommendation information and the feedback operation made by the visiting user in real time, and provide information for the visiting user The above content formed by a user's visit-recommendation process is combined into a new user model training item;
将所述新增的用户模型训练条目加入训练样本集；adding the newly added user model training item to the training sample set;
以当前训练样本集进行用户模型训练；Use the current training sample set for user model training;
所述用户模型训练完成后，将所获得的更新的用户模型作为当前用户模型。After the training of the user model is completed, the obtained updated user model is used as the current user model.
本申请第四实施例提供一种实现上述第一实施例方法的在线个性化推荐的系统；请参看图3。The fourth embodiment of the present application provides an online personalized recommendation system implementing the method of the first embodiment above; please refer to FIG. 3 .
所述在线个性化的系统，包括在线子系统401、离线子系统402。The online personalized system includes an online subsystem 401 and an offline subsystem 402 .
所述在线子系统401，用于接收用户的访问请求，并提取用户特征数据，并根据所提取的用户特征数据以及当前用户模型，向提出访问请求的用户提供个性化推荐实体；以及，实时搜集所述访问用户的用户特征数据、为所述访问用户提供的所述个性化推荐信息、所述访问用户对所述个性化推荐信息的反馈信息，形成新增的用户模型训练条目并发送；以及，接收所述离线子系统提供的更新的用户模型。The online subsystem 401 is configured to receive a user's access request, extract user characteristic data, and provide a personalized recommendation entity to the user who made the access request according to the extracted user characteristic data and the current user model; and, collect in real time The user feature data of the visiting user, the personalized recommendation information provided for the visiting user, and the feedback information of the visiting user to the personalized recommendation information form a newly added user model training item and send it; and , receiving an updated user model provided by the offline subsystem.
所述在线子系统实现实时在线的功能，包括界面展示以及客户端的实时数据搜集等，本系统强调对所有训练数据的实时搜集，一次性完整搜集相关数据形成用户模型训练条目，这样就可以避免离线搜集数据需要访问多个数据库以及不能及时搜集到完整数据的弊端。在线子系统除了客户端外，可以包括通过网络与其连接的服务器。The online subsystem realizes real-time online functions, including interface display and client real-time data collection, etc. This system emphasizes the real-time collection of all training data, and collects relevant data at one time to form user model training items, so as to avoid offline Data collection requires access to multiple databases and the drawbacks of not being able to collect complete data in a timely manner. In addition to clients, the online subsystem may include servers connected to it via a network.
在优选的方案中，所述在线子系统401形成的用户模型训练条目以日志形式记录并发送。In a preferred solution, the user model training entries formed by the online subsystem 401 are recorded and sent in the form of logs.
所述离线子系统402，接收所述在线子系统401发送的新增的用户模型训练条目，并将所述新增的用户模型训练条目加入当前训练样本集；采用所述当前训练样本集进行用户模型训练，将训练完成获得的更新的用户模型作为当前用户模型向外发送。The offline subsystem 402 receives the newly added user model training item sent by the online subsystem 401, and adds the newly added user model training item into the current training sample set; adopts the current training sample set for user Model training, the updated user model obtained after training is sent out as the current user model.
所述离线子系统将所述新增的用户模型训练条目加入当前训练样本集之后，随时判断所述新增的用户模型训练样本是否达到预定的阈值；若是，则开始所述采用所述当前训练样本集进行用户模型训练的过程。After the offline subsystem adds the newly added user model training entry to the current training sample set, it judges at any time whether the newly added user model training sample reaches a predetermined threshold; The sample set is used to train the user model.
该离线子系统402不处理在线工作，随时可以根据需要启动用户模型训练，不影响在线系统的工作。The offline subsystem 402 does not handle online work, and user model training can be started at any time as needed without affecting the work of the online system.
本申请虽然以较佳实施例公开如上，但其并不是用来限定本申请，任何本领域技术人员在不脱离本申请的精神和范围内，都可以做出可能的变动和修改，因此本申请的保护范围应当以本申请权利要求所界定的范围为准。Although the present application is disclosed as above with preferred embodiments, it is not intended to limit the present application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the present application The scope of protection should be based on the scope defined by the claims of this application.
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。Memory may include non-permanent storage in computer readable media, in the form of random access memory (RAM) and/or nonvolatile memory such as read only memory (ROM) or flash RAM. Memory is an example of computer readable media.
1、计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括非暂存电脑可读媒体(transitorymedia)，如调制的数据信号和载波。1. Computer-readable media include permanent and non-permanent, removable and non-removable media. Information storage can be realized by any method or technology. Information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash memory or other memory technology, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridge, tape magnetic disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media excludes non-transitory computer-readable media, such as modulated data signals and carrier waves.
2、本领域技术人员应明白，本申请的实施例可提供为方法、系统或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。2. Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610149438.5ACN107203518A (en) | 2016-03-16 | 2016-03-16 | Method, system and device, the electronic equipment of on-line system personalized recommendation |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610149438.5ACN107203518A (en) | 2016-03-16 | 2016-03-16 | Method, system and device, the electronic equipment of on-line system personalized recommendation |
| Publication Number | Publication Date |
|---|---|
| CN107203518Atrue CN107203518A (en) | 2017-09-26 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610149438.5APendingCN107203518A (en) | 2016-03-16 | 2016-03-16 | Method, system and device, the electronic equipment of on-line system personalized recommendation |
| Country | Link |
|---|---|
| CN (1) | CN107203518A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107705183A (en)* | 2017-09-30 | 2018-02-16 | 深圳乐信软件技术有限公司 | Recommendation method, apparatus, storage medium and the server of a kind of commodity |
| CN108009877A (en)* | 2017-11-24 | 2018-05-08 | 阿里巴巴集团控股有限公司 | Information mining method and device |
| CN108763342A (en)* | 2018-05-14 | 2018-11-06 | 北京比特智学科技有限公司 | Education resource distribution method and device |
| CN108803868A (en)* | 2018-04-13 | 2018-11-13 | 北京诺亦腾科技有限公司 | A kind of VR scenes experience cabin collecting method and device |
| CN108846687A (en)* | 2018-04-02 | 2018-11-20 | 平安科技（深圳）有限公司 | Client segmentation method, apparatus and storage medium |
| CN109146606A (en)* | 2018-07-09 | 2019-01-04 | 广州品唯软件有限公司 | A kind of brand recommended method, electronic equipment, storage medium and system |
| CN109299368A (en)* | 2018-09-29 | 2019-02-01 | 北京思路创新科技有限公司 | A kind of method and system for the intelligent personalized recommendation of environmental information resource AI |
| CN109325170A (en)* | 2018-08-06 | 2019-02-12 | 江西清华泰豪三波电机有限公司 | Material method for pushing and device |
| CN109413175A (en)* | 2018-10-19 | 2019-03-01 | 北京奇艺世纪科技有限公司 | A kind of information processing method, device and electronic equipment |
| WO2019061989A1 (en)* | 2017-09-30 | 2019-04-04 | 平安科技（深圳）有限公司 | Loan risk control method, electronic device and readable storage medium |
| CN110059170A (en)* | 2019-03-21 | 2019-07-26 | 北京邮电大学 | More wheels based on user's interaction talk with on-line training method and system |
| CN110059254A (en)* | 2019-04-24 | 2019-07-26 | 秒针信息技术有限公司 | A kind of message push method and device |
| CN110135979A (en)* | 2019-05-14 | 2019-08-16 | 极智(上海)企业管理咨询有限公司 | The matched method, apparatus of individual demand, medium and electronic equipment |
| CN110163693A (en)* | 2018-01-30 | 2019-08-23 | 哈尔滨学院 | A kind of electric business Method of Commodity Recommendation based on big data |
| WO2019205950A1 (en)* | 2018-04-28 | 2019-10-31 | K11集团有限公司 | User data collection system and information pushing method |
| CN110413888A (en)* | 2019-07-24 | 2019-11-05 | 腾讯科技（深圳）有限公司 | A kind of books recommended method and device |
| CN110428298A (en)* | 2019-07-15 | 2019-11-08 | 阿里巴巴集团控股有限公司 | A kind of shop recommended method, device and equipment |
| CN110472995A (en)* | 2019-07-08 | 2019-11-19 | 汉海信息技术(上海)有限公司 | To shop prediction technique, device, readable storage medium storing program for executing and electronic equipment |
| CN110674408A (en)* | 2019-09-30 | 2020-01-10 | 北京三快在线科技有限公司 | Service platform, and real-time generation method and device of training sample |
| CN110717536A (en)* | 2019-09-30 | 2020-01-21 | 北京三快在线科技有限公司 | Method and device for generating training sample |
| CN111246056A (en)* | 2018-11-29 | 2020-06-05 | 佳能株式会社 | Information processing apparatus and control method thereof, imaging apparatus and storage medium |
| CN111461283A (en)* | 2020-03-18 | 2020-07-28 | 上海携程商务有限公司 | Automatic iteration operation and maintenance method, system, equipment and storage medium of AI model |
| CN111861605A (en)* | 2019-04-28 | 2020-10-30 | 阿里巴巴集团控股有限公司 | Business object recommendation method |
| CN112182359A (en)* | 2019-07-05 | 2021-01-05 | 腾讯科技（深圳）有限公司 | Feature management method and system of recommendation model |
| CN112598487A (en)* | 2021-02-20 | 2021-04-02 | 汇正(广州)企业管理咨询有限公司 | Enterprise informatization management method and system based on artificial intelligence |
| CN113378060A (en)* | 2021-07-01 | 2021-09-10 | 第四范式（北京）技术有限公司 | Resource recommendation method, device, equipment and medium |
| CN113573106A (en)* | 2020-04-28 | 2021-10-29 | 北京达佳互联信息技术有限公司 | Model updating method and device for multimedia information and server |
| CN114154816A (en)* | 2021-11-17 | 2022-03-08 | 鼎捷软件股份有限公司 | Enterprise management system and execution method thereof |
| CN114519144A (en)* | 2022-02-21 | 2022-05-20 | 北京三快在线科技有限公司 | Information recommendation method, device, server and storage medium |
| CN115470397A (en)* | 2021-06-10 | 2022-12-13 | 腾讯科技（深圳）有限公司 | Content recommendation method and device, computer equipment and storage medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101206751A (en)* | 2007-12-25 | 2008-06-25 | 北京科文书业信息技术有限公司 | Customer recommendation system based on data digging and method thereof |
| CN102800006A (en)* | 2012-07-23 | 2012-11-28 | 姚明东 | Real-time goods recommendation method based on customer shopping intention exploration |
| CN103020128A (en)* | 2012-11-19 | 2013-04-03 | 北京奇虎科技有限公司 | Method and device for data interaction with terminal device |
| CN103164434A (en)* | 2011-12-13 | 2013-06-19 | 阿里巴巴集团控股有限公司 | Method, device and system for obtaining real-time data |
| CN103678647A (en)* | 2013-12-20 | 2014-03-26 | Tcl集团股份有限公司 | Method and system for recommending information |
| CN104376058A (en)* | 2014-11-07 | 2015-02-25 | 华为技术有限公司 | User interest model updating method and device |
| CN104851025A (en)* | 2015-05-09 | 2015-08-19 | 湘南学院 | Case-reasoning-based personalized recommendation method for E-commerce website commodity |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101206751A (en)* | 2007-12-25 | 2008-06-25 | 北京科文书业信息技术有限公司 | Customer recommendation system based on data digging and method thereof |
| CN103164434A (en)* | 2011-12-13 | 2013-06-19 | 阿里巴巴集团控股有限公司 | Method, device and system for obtaining real-time data |
| CN102800006A (en)* | 2012-07-23 | 2012-11-28 | 姚明东 | Real-time goods recommendation method based on customer shopping intention exploration |
| CN103020128A (en)* | 2012-11-19 | 2013-04-03 | 北京奇虎科技有限公司 | Method and device for data interaction with terminal device |
| CN103678647A (en)* | 2013-12-20 | 2014-03-26 | Tcl集团股份有限公司 | Method and system for recommending information |
| CN104376058A (en)* | 2014-11-07 | 2015-02-25 | 华为技术有限公司 | User interest model updating method and device |
| CN104851025A (en)* | 2015-05-09 | 2015-08-19 | 湘南学院 | Case-reasoning-based personalized recommendation method for E-commerce website commodity |
| Title |
|---|
| 张轩荣: "让客户的行为留下痕迹", 《大数据时代，营销团队就该这样管》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107705183B (en)* | 2017-09-30 | 2021-04-27 | 深圳乐信软件技术有限公司 | A product recommendation method, device, storage medium and server |
| CN107705183A (en)* | 2017-09-30 | 2018-02-16 | 深圳乐信软件技术有限公司 | Recommendation method, apparatus, storage medium and the server of a kind of commodity |
| WO2019061989A1 (en)* | 2017-09-30 | 2019-04-04 | 平安科技（深圳）有限公司 | Loan risk control method, electronic device and readable storage medium |
| CN108009877A (en)* | 2017-11-24 | 2018-05-08 | 阿里巴巴集团控股有限公司 | Information mining method and device |
| CN110163693A (en)* | 2018-01-30 | 2019-08-23 | 哈尔滨学院 | A kind of electric business Method of Commodity Recommendation based on big data |
| CN108846687A (en)* | 2018-04-02 | 2018-11-20 | 平安科技（深圳）有限公司 | Client segmentation method, apparatus and storage medium |
| CN108803868A (en)* | 2018-04-13 | 2018-11-13 | 北京诺亦腾科技有限公司 | A kind of VR scenes experience cabin collecting method and device |
| CN110415065A (en)* | 2018-04-28 | 2019-11-05 | K11集团有限公司 | User data collection system and information push method |
| WO2019205950A1 (en)* | 2018-04-28 | 2019-10-31 | K11集团有限公司 | User data collection system and information pushing method |
| CN108763342A (en)* | 2018-05-14 | 2018-11-06 | 北京比特智学科技有限公司 | Education resource distribution method and device |
| CN109146606A (en)* | 2018-07-09 | 2019-01-04 | 广州品唯软件有限公司 | A kind of brand recommended method, electronic equipment, storage medium and system |
| CN109146606B (en)* | 2018-07-09 | 2022-02-22 | 广州品唯软件有限公司 | Brand recommendation method, electronic equipment, storage medium and system |
| CN109325170A (en)* | 2018-08-06 | 2019-02-12 | 江西清华泰豪三波电机有限公司 | Material method for pushing and device |
| CN109299368A (en)* | 2018-09-29 | 2019-02-01 | 北京思路创新科技有限公司 | A kind of method and system for the intelligent personalized recommendation of environmental information resource AI |
| CN109413175B (en)* | 2018-10-19 | 2021-07-20 | 北京奇艺世纪科技有限公司 | Information processing method and device and electronic equipment |
| CN109413175A (en)* | 2018-10-19 | 2019-03-01 | 北京奇艺世纪科技有限公司 | A kind of information processing method, device and electronic equipment |
| CN111246056A (en)* | 2018-11-29 | 2020-06-05 | 佳能株式会社 | Information processing apparatus and control method thereof, imaging apparatus and storage medium |
| CN110059170A (en)* | 2019-03-21 | 2019-07-26 | 北京邮电大学 | More wheels based on user's interaction talk with on-line training method and system |
| CN110059170B (en)* | 2019-03-21 | 2022-04-26 | 北京邮电大学 | Multi-round dialogue online training method and system based on user interaction |
| CN110059254A (en)* | 2019-04-24 | 2019-07-26 | 秒针信息技术有限公司 | A kind of message push method and device |
| CN111861605A (en)* | 2019-04-28 | 2020-10-30 | 阿里巴巴集团控股有限公司 | Business object recommendation method |
| CN110135979A (en)* | 2019-05-14 | 2019-08-16 | 极智(上海)企业管理咨询有限公司 | The matched method, apparatus of individual demand, medium and electronic equipment |
| CN112182359B (en)* | 2019-07-05 | 2024-03-15 | 深圳市雅阅科技有限公司 | Feature management method and system of recommendation model |
| CN112182359A (en)* | 2019-07-05 | 2021-01-05 | 腾讯科技（深圳）有限公司 | Feature management method and system of recommendation model |
| CN110472995B (en)* | 2019-07-08 | 2024-11-01 | 汉海信息技术(上海)有限公司 | Store arrival prediction method and device, readable storage medium and electronic equipment |
| CN110472995A (en)* | 2019-07-08 | 2019-11-19 | 汉海信息技术(上海)有限公司 | To shop prediction technique, device, readable storage medium storing program for executing and electronic equipment |
| CN110428298A (en)* | 2019-07-15 | 2019-11-08 | 阿里巴巴集团控股有限公司 | A kind of shop recommended method, device and equipment |
| CN110413888B (en)* | 2019-07-24 | 2024-05-10 | 腾讯科技（深圳）有限公司 | Book recommendation method and device |
| CN110413888A (en)* | 2019-07-24 | 2019-11-05 | 腾讯科技（深圳）有限公司 | A kind of books recommended method and device |
| CN110717536A (en)* | 2019-09-30 | 2020-01-21 | 北京三快在线科技有限公司 | Method and device for generating training sample |
| CN110674408B (en)* | 2019-09-30 | 2021-06-04 | 北京三快在线科技有限公司 | Service platform, and real-time generation method and device of training sample |
| CN110674408A (en)* | 2019-09-30 | 2020-01-10 | 北京三快在线科技有限公司 | Service platform, and real-time generation method and device of training sample |
| CN110717536B (en)* | 2019-09-30 | 2025-04-29 | 北京三快在线科技有限公司 | A method and device for generating training samples |
| CN111461283A (en)* | 2020-03-18 | 2020-07-28 | 上海携程商务有限公司 | Automatic iteration operation and maintenance method, system, equipment and storage medium of AI model |
| CN113573106A (en)* | 2020-04-28 | 2021-10-29 | 北京达佳互联信息技术有限公司 | Model updating method and device for multimedia information and server |
| CN113573106B (en)* | 2020-04-28 | 2023-03-21 | 北京达佳互联信息技术有限公司 | Model updating method and device for multimedia information and server |
| CN112598487A (en)* | 2021-02-20 | 2021-04-02 | 汇正(广州)企业管理咨询有限公司 | Enterprise informatization management method and system based on artificial intelligence |
| CN115470397A (en)* | 2021-06-10 | 2022-12-13 | 腾讯科技（深圳）有限公司 | Content recommendation method and device, computer equipment and storage medium |
| CN115470397B (en)* | 2021-06-10 | 2024-04-05 | 腾讯科技（深圳）有限公司 | Content recommendation method, device, computer equipment and storage medium |
| CN113378060A (en)* | 2021-07-01 | 2021-09-10 | 第四范式（北京）技术有限公司 | Resource recommendation method, device, equipment and medium |
| CN114154816A (en)* | 2021-11-17 | 2022-03-08 | 鼎捷软件股份有限公司 | Enterprise management system and execution method thereof |
| CN114519144A (en)* | 2022-02-21 | 2022-05-20 | 北京三快在线科技有限公司 | Information recommendation method, device, server and storage medium |
| Publication | Publication Date | Title |
|---|---|---|
| CN107203518A (en) | Method, system and device, the electronic equipment of on-line system personalized recommendation | |
| TWI690880B (en) | Recommended method and device for marketing products | |
| TWI694401B (en) | Searching method and device integrating user relationship data | |
| US9928537B2 (en) | Management and storage of distributed bookmarks | |
| TWI615724B (en) | Information push, search method and device based on electronic information-based keyword extraction | |
| JP5897019B2 (en) | Method and apparatus for determining linked list of candidate products | |
| US11275748B2 (en) | Influence score of a social media domain | |
| WO2016101777A1 (en) | Analysis and collection system for user interest data and method therefor | |
| TW201327233A (en) | Personalized information push method and device | |
| CN104391999B (en) | Information recommendation method and device | |
| TW201513019A (en) | Method and system for extracting user behavior features to personalize recommendations | |
| WO2014018781A1 (en) | Data refining engine for high performance analysis system and method | |
| CN102339448B (en) | Group purchase platform information processing method and device | |
| CN105469263A (en) | Commodity recommendation method and device | |
| CN104699725A (en) | Data searching processing method and system | |
| CA3070612A1 (en) | Click rate estimation | |
| US9390446B2 (en) | Consumer centric online product research | |
| JP6560323B2 (en) | Determination device, determination method, and determination program | |
| CN107633416B (en) | Method, device and system for recommending service object | |
| CN110188291B (en) | Document processing based on proxy log | |
| TW201828200A (en) | Data processing method and apparatus increasing the overall display efficiency of the object display environment and decreasing the waste of display resources of each object display environment | |
| CN110796505B (en) | Business object recommendation method and device | |
| CN111858686A (en) | Data display method and device, terminal equipment and storage medium | |
| CN110209944A (en) | A kind of stock analysis teacher recommended method, device, computer equipment and storage medium | |
| CN119179774A (en) | Data processing and digest generation method, medium, computer device, and program product |
| 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 | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20170926 | |
| RJ01 | Rejection of invention patent application after publication |