WO2019134274A1 - 兴趣探索方法、存储介质、电子设备及系统 - Google Patents

兴趣探索方法、存储介质、电子设备及系统 Download PDF

Info

Publication number
WO2019134274A1
WO2019134274A1 PCT/CN2018/081315 CN2018081315W WO2019134274A1 WO 2019134274 A1 WO2019134274 A1 WO 2019134274A1 CN 2018081315 W CN2018081315 W CN 2018081315W WO 2019134274 A1 WO2019134274 A1 WO 2019134274A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
content
exploration
interest
probability
Prior art date
Application number
PCT/CN2018/081315
Other languages
English (en)
French (fr)
Inventor
李龙华
陈少杰
张文明
Original Assignee
武汉斗鱼网络科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 武汉斗鱼网络科技有限公司 filed Critical 武汉斗鱼网络科技有限公司
Publication of WO2019134274A1 publication Critical patent/WO2019134274A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising

Definitions

  • the present invention relates to the field of network technologies, and in particular, to an interest exploration method, a storage medium, a device, and a system.
  • the personalized recommendation system is based on the user's historical behavior. Although such a recommendation system can be personalized, the recommendation based on this model is easy to appear.
  • the recommended content is concentrated in the preferences that users often pay attention to. Over time, users' interest preferences will become narrower and narrower, and even some users' activity will decline or be lost. The main reason for this phenomenon is that in the era of information explosion, users themselves are very difficult to screen information and need tools.
  • the personalized recommendation system helps users to do this, but it does not help users discover new points of interest. In the recommended area, this issue is defined as a surprise. Since the recommendation system is aimed at the content that the user explores, the user does not have the behavior, and has certain blindness, which may cause the final conversion rate to decrease. Therefore, most of the recommendation systems are designed to avoid the negative effects of interest exploration. This feature is not used.
  • the object of the present invention is to overcome the deficiencies of the above background art, and to provide an interest exploration method, a storage medium, a device and a system, and determine whether to explore the interest of the user according to the probability of the user's exploration, and improve the success rate of the interest exploration and the user's Conversion rate.
  • the technical solution adopted by the present invention is to provide an interest exploration method, which comprises the following steps:
  • the recommended access content belongs to multiple content categories
  • the user includes a registered user
  • the calculation method of the search probability p m of the registered user is:
  • n is the total number of content categories visited by the registered user.
  • the user includes a non-registered user
  • the calculation method of the exploration probability ⁇ k of the non-registered user is:
  • ⁇ is the number of times the interest exploration is desired
  • k is the number of times the interest exploration has been performed on the non-registered user.
  • the user's feedback information is extracted from the in-memory database, and the feedback information is obtained from the client and stored in the in-memory database;
  • the feedback information includes the number of exposures of the recommended access content and the number of clicks of the user in each of the content categories.
  • the specific process for filtering the content classification that is continuously pushed from the plurality of content categories according to the feedback information includes:
  • a score of the plurality of content categories is calculated, and the content with the highest score is classified as a content classification that continues to be pushed.
  • the scores of all the content classifications are calculated according to the confidence interval upper bound algorithm, and the calculation method is:
  • I j is the score of the jth content classification
  • T j (n) is the cumulative number of interest exploration times for the jth content
  • n is the number of times the cumulative interest exploration is classified for all the content
  • ⁇ is the weight of the jth content classification
  • c is the user's The number of clicks of the j content categories
  • v is the number of exposures of the jth content classification.
  • the present invention also provides a storage medium having stored thereon a computer program that, when executed by a processor, implements the steps of the above method.
  • the present invention also provides a control presentation device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, the processor implementing the steps of the method when the computer program is executed.
  • the invention also provides an interest exploration system, which comprises a discovery probability calculation module, a feedback information acquisition module and a screening module;
  • the exploration probability calculation module is configured to: calculate, according to historical behavior information of the user, an exploration probability of performing interest exploration for each user;
  • the feedback information obtaining module is configured to: when the user's search probability exceeds a set threshold, obtain feedback information about the latest recommended access content of the user, where the recommended access content belongs to multiple content categories;
  • the screening module is configured to: screen, according to the feedback information, a content classification that is continuously pushed from a plurality of the content categories.
  • the push weights are sorted in the previous content classification, and different content classifications can be explored differently to further improve the success rate of the exploration.
  • the user is divided into a registered user and a non-registered user.
  • the calculation method of the exploration probability of the registered user and the non-registered user is different, so as to give the registered user and the non-registered user different exploration probabilities, thereby further improving the success rate of the exploration.
  • FIG. 1 is a flowchart of an interest exploration method according to an embodiment of the present invention.
  • FIG. 2 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
  • FIG. 3 is a structural block diagram of an interest exploration system according to an embodiment of the present invention.
  • an embodiment of the present invention provides an interest exploration method, which can be applied to online real-time exploration of a user in music, video, news, games, webcasting, or online shopping, and the method includes the following steps:
  • the user first reads the user's historical behavior information (user portrait) to determine the user's preference; and then, according to the configuration, calculates whether the user's request triggers the interest exploration behavior.
  • the user's historical behavior information includes the user's basic information and behavior information.
  • the basic information includes user registration duration, user level, user mailbox authentication status, user mobile phone authentication status, source type, and registration place.
  • Behavior information includes viewing information, login information, recharge information, barrage information, and transaction information. From the time dimension, the behavior information also includes historical behavior information indicators, historical behavior information index volatility and recent behavior information.
  • the historical behavior information of the user may further include user data obtained by one or more of the basic information and the behavior information.
  • the user is divided into a registered user and a non-registered user, and the calculation methods of the search probability of the registered user and the non-registered user are different, so as to give the registered user and the non-registered user different exploration probabilities, thereby further improving the exploration.
  • the success rate specifically, tends to give registered users a smaller probability of exploration and increase the probability of exploration for non-registered users.
  • the calculation method of the exploration probability p m of the registered user is:
  • a is a fixed probability
  • a is a constant selected according to a specific application scenario, 0 ⁇ a ⁇ 1
  • m is the total number of content categories visited by the registered user, and can be statistically obtained from the historical behavior information of the user.
  • the probability of exploration for the registered user is the largest (1).
  • the probability of exploring the interest of the registered user is smaller.
  • the calculation probability of the non-registered user's exploration probability ⁇ k is:
  • ⁇ is the number of times the interest exploration is desired
  • k is the number of times the interest exploration has been performed on the non-registered user, which can be statistically obtained from the historical behavior information of the user.
  • k may exceed ⁇ , and the probability of exploration will be lower when k exceeds the expected number of interest explorations.
  • Content classification is a category for recommending access to content. Different application scenarios have different content classifications. Taking a webcast platform as an example, content classification may include games, entertainment, value, technology, etc., wherein the game can also be divided into computers. Small categories such as games, living room games and mobile games can be further subdivided. The content classification pushed to the user may be one of the above categories, subclasses, subdivisions, or a combination of several. In each content category, there may be multiple recommended access content, and each recommended access content can only be classified into one content category.
  • the interest exploration of the user is triggered.
  • the user's feedback information about the latest recommended access content is obtained.
  • the feedback information of the user is extracted from the in-memory database, and the feedback information is obtained from the client and stored in the in-memory database.
  • the feedback information includes the number of exposures of the recommended content and the number of clicks of the user in each content category.
  • the client used by the user forwards the user behavior log to the log recovery system in real time, and the log recovery system writes the user behavior log to the data stream, such as the Kafka distributed publish subscription message system.
  • the data stream such as the Kafka distributed publish subscription message system.
  • a streaming service is started to calculate the number of clicks and exposures of each user under their respective content categories.
  • the statistical results are written into the redis in-memory database in real time, and the feedback information can be rushed into the redis in-memory database in real time when offline.
  • the method for determining the content classification of the recommended access content is: sorting all the N 0 content categories that have not been accessed by each user according to the weight from large to small, and selecting the N 1 content categories ranked as the recommendation as the recommendation.
  • the content classification of the accessed content, where 1 ⁇ N 1 ⁇ N 0 , the weight of the content classification is obtained by statistically categorizing all the content that all users have visited, so that different content classifications can be explored differently to further improve The success rate of exploration.
  • the number of content categories pushed to each user may be the same or different.
  • the total number of content categories accessed by the user may be preset for each user, and the content of the recommended access content pushed to the user is classified into the difference between the total number of content categories and the number of content categories accessed by the user. Push multiple content categories for users each time to increase the efficiency of interest exploration.
  • the recommended access content may also be selected from the content categories that are less accessed by the user, and pushed to the user to further explore the user's interests or needs.
  • the user's feedback information includes user behavior information such as clicks, downloads, browsing, online installation, collection, and/or evaluation (eg, comments, ratings, likes, etc.) of the recommended access content.
  • user behavior information such as clicks, downloads, browsing, online installation, collection, and/or evaluation (eg, comments, ratings, likes, etc.) of the recommended access content.
  • the set threshold is selected according to the application scenario.
  • Step S3 specifically includes: calculating a score of the plurality of content categories, and classifying the content with the highest score as the content classification that continues to be pushed.
  • the scores for calculating multiple content categories may be one or a combination of two or more.
  • the following algorithms include an ⁇ -greedy algorithm, a sampling method algorithm, a Ranked Bandits algorithm, a Contextual Bandits algorithm, and a Reinforcement Learning algorithm.
  • the scores of all content categories are calculated according to the confidence interval upper bound algorithm, and the calculation method is:
  • I j is the score of the jth content classification
  • T j (n) is the cumulative interest exploration number for the jth content classification
  • n is the number of times the cumulative interest exploration is classified for all content
  • ⁇ is the weight of the jth content classification
  • c is the user's click on the jth content classification.
  • the number of times, v is the number of exposures for the jth content classification.
  • the user is determined whether to explore the interest according to the probability of the user's exploration, and avoids blindly exploring the interest of all users, thereby improving the success rate of interest exploration and the conversion rate of the user.
  • the embodiment of the invention can effectively solve the "surprise” problem in the recommendation system, and at the same time, can minimize the negative effects brought about by the interest exploration, can effectively reduce the loss of the user, and increase the viscosity of the user.
  • the embodiment of the invention further provides a storage medium on which a computer program is stored, and when the computer program is executed by the processor, the interest exploration method is implemented.
  • the storage medium includes a U disk, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk, or an optical disk, and the like, which can store program codes. medium.
  • an embodiment of the present invention further provides an electronic device, including a memory and a processor.
  • the memory stores a computer program running on the processor, and the method for exploring the interest is implemented when the processor executes the computer program.
  • the embodiment of the invention further provides an interest exploration system, which comprises a exploration probability calculation module, a feedback information acquisition module and a screening module.
  • the exploration probability calculation module is configured to: calculate an exploration probability of interest exploration for each user according to the historical behavior information of the user.
  • the feedback information obtaining module is configured to: when the user's exploration probability exceeds the set threshold, obtain feedback information of the user on the latest recommended access content, and the recommended access content belongs to multiple content categories.
  • the screening module is configured to: filter the content classification that is continuously pushed from the plurality of content categories according to the feedback information.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Accounting & Taxation (AREA)
  • General Engineering & Computer Science (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Computational Linguistics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本发明公开了一种兴趣探索方法、存储介质、设备及系统，涉及网络技术领域。该方法包括以下步骤：根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率；当用户的探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，推荐访问内容属于多个内容分类；根据反馈信息从多个内容分类中筛选出继续推送的内容分类。本发明根据用户的探索概率确定是否对该用户进行兴趣探索，避免盲目地向所有用户进行兴趣探索，从而提高兴趣探索的成功率和用户的转化率。

Description

兴趣探索方法、存储介质、电子设备及系统 技术领域
本发明涉及网络技术领域，具体来讲是一种兴趣探索方法、存储介质、设备及系统。
背景技术
在推荐领域中，个性化的推荐系统都是基于用户的历史行为，这样的推荐系统虽然可以做到个性化，但是基于这一模型的推荐很容易出现推荐的内容集中在用户经常关注的偏好内，久而久之，就会出现用户的兴趣偏好越来越窄，甚至会出现部分用户的活跃度下降或者流失。造成这一现象的原因主要是在信息爆炸的时代，用户本身对信息的筛选十分困难，需要借助工具。个性化的推荐系统虽然很好地帮助用户做到了这一点，但是却不能帮助用户发现新的兴趣点。在推荐领域中，这一问题被定义为惊喜度。由于推荐系统针对用户所探索的内容都是用户没有发生的行为，具有一定的盲目性，这可能会造成最终的转化率下降，因此大部分的推荐系统为了规避兴趣探索带来的负面效果，都没有使用这一功能。
发明内容
本发明的目的是为了克服上述背景技术的不足，提供一种兴趣探索方法、存储介质、设备及系统，根据用户的探索概率确定是否对该用户进行兴趣探索，提高兴趣探索的成功率和用户的转化率。
为达到以上目的，本发明采取的技术方案是：提供一种兴趣探索方法，该方法包括以下步骤：
根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率；
当用户的所述探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，所述推荐访问内容属于多个内容分类；
根据所述反馈信息从多个所述内容分类中筛选出继续推送的内容分类。
在上述技术方案的基础上，用户包括注册用户，注册用户的所述探索概率p m的计算方法为：
Figure PCTCN2018081315-appb-000001
其中，a为固定概率，0＜a＜1，m为该注册用户访问过的内容分类总数。
在上述技术方案的基础上，用户包括非注册用户，非注册用户的所述探索概率ρ k的计算方法为：
Figure PCTCN2018081315-appb-000002
其中，λ为期望进行兴趣探索的次数，k为对该非注册用户已经进行的兴趣探索的次数。
在上述技术方案的基础上，对每个用户未访问过的所有N 0个内容分类按照权重从大到小进行排序，并选取排在前面的N 1个内容分类作为所述推荐访问内容的内容分类，其中，1＜N 1≤N 0，所述内容分类的权重通过对所有用户访问过的所有内容分类进行统计得到。
在上述技术方案的基础上，从内存数据库中提取用户的反馈信息，所述反馈信息是从客户端获取后存入内存数据库的；
其中，所述反馈信息包括在每个所述内容分类中，所述推荐访问内容的曝光次数以及该用户的点击次数。
在上述技术方案的基础上，根据反馈信息从多个所述内容分类中筛选出继续推送的内容分类的具体流程包括：
计算多个所述内容分类的得分，将得分最高的所述内容分类作为继续推送的内容分类。
在上述技术方案的基础上，根据置信区间上界算法计算所有所述内容分类的得分，其计算方法为：
Figure PCTCN2018081315-appb-000003
其中，I j为第j个所述内容分类的得分，
Figure PCTCN2018081315-appb-000004
为第j个所述内容分类的平均分，
Figure PCTCN2018081315-appb-000005
T j(n)为对第j个所述内容分类累计兴趣探索次数，n为对所有所述内容分类累计兴趣探索的次数，β为第j个所述内容分类的权重，c为用户对第j个所述内容分类的点击次数，v为第j个所述内容分类的曝光次数。
本发明还提供一种存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现上述方法的步骤。
本发明还提供一种控件呈现设备，包括存储器、处理器及存储在存储器上并在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述方法的步骤。
本发明还提供一种兴趣探索系统，该系统包括探索概率计算模块、反馈信息获取模块和筛选模块；
所述探索概率计算模块用于：根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率；
所述反馈信息获取模块用于：当用户的所述探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，所述推荐访问内容属于多个内容分类；
所述筛选模块用于：根据所述反馈信息从多个所述内容分类中筛选出继续推送的内容分类。
与现有技术相比，本发明的优点在于：
(1)根据用户的探索概率确定是否对该用户进行兴趣探索，避免盲目地向所有用户进行兴趣探索，从而提高兴趣探索的成功率和用户的转化率。
(2)在每个用户未访问过的内容分类中，推送权重排序在前的内容分类，可以实现对不同的内容分类进行有差别地探索，以进一步提高探索的成功率。
(3)将用户分为注册用户和非注册用户，注册用户和非注册用户的探索概率的计算方法不同，以给予注册用户和非注册用户不同的探索概率，以进一步提高探索的成功率。
附图说明
图1为本发明实施例兴趣探索方法的流程图；
图2为本发明实施例中电子设备的结构示意图；
图3为本发明实施例中兴趣探索系统的结构框图。
具体实施方式
以下结合附图及实施例对本发明作进一步详细说明。
参见图1所示，本发明实施例提供一种兴趣探索方法，可以适用于用户在音乐、视频、新闻、游戏、网络直播或者网上购物等方面的线上实时探索，该方法包括以下步骤：
S1.根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率。
以网络直播平台为例，线上首先读取用户的历史行为信息(用户 画像)，确定用户的偏好；然后，根据配置计算用户的该次请求是否触发兴趣探索行为。
用户的历史行为信息包括用户的基础信息和行为信息。其中，基础信息包括用户注册时长、用户等级、用户邮箱认证状态，用户手机认证状态、来源类型和注册地等。行为信息包括观看信息、登录信息、充值信息、弹幕信息和交易信息等。从时间维度上，行为信息还包括历史行为信息指标、历史行为信息指标波动率和最近行为信息等。用户的历史行为信息还可以包括根据基础信息和行为信息中的一种、或者两种以上进行统计得到的用户数据。
在一种实施方式中，将用户分为注册用户和非注册用户，注册用户和非注册用户的探索概率的计算方法不同，以给予注册用户和非注册用户不同的探索概率，以进一步提高探索的成功率，具体的，倾向于给予注册用户较小的探索概率，并加大非注册用户的探索概率。注册用户的探索概率p m的计算方法为：
Figure PCTCN2018081315-appb-000006
其中，a为固定概率，a为根据具体的应用场景来选定的常数，0＜a＜1，m为该注册用户访问过的内容分类总数，可以从用户的历史行为信息中统计得到。当注册用户没有访问过任何内容分类时，对该注册用户的探索概率最大(为1)。当注册用户访问过的内容分类越多时，对该注册用户进行兴趣探索的探索概率越小。
非注册用户的探索概率ρ k的计算方法为：
Figure PCTCN2018081315-appb-000007
其中，λ为期望进行兴趣探索的次数，k为对该非注册用户已经进行的兴趣探索的次数，可以从用户的历史行为信息中统计得到。k 可能会超过λ，当k超过期望进行兴趣探索的次数越大时，探索概率会越低。
在实际应用中，由于几乎所有的注册用户都有自己的兴趣偏好的内容分类，即m都不等于0，而非注册用户普遍缺乏兴趣偏好的内容分类，m基本都为0，因此，如果采用公式(1)来计算非注册用户的探索概率，会导致所有非注册用户的探索概率都为1，没有考虑随着对非注册用户已经进行兴趣探索的次数增加，所造成的探索概率应当下降的因素。因此，与注册用户相比，对于非注册用户不但采用公式(2)来计算探索概率，而且倾向于加大非注册用户的探索概率。
内容分类是推荐访问内容的类别，不同的应用场景具有不同的内容分类，以网络直播平台为例，内容分类可以包括游戏、娱乐、颜值、科技等大类，其中，游戏还可以分为电脑游戏、客厅游戏和手机游戏等小类，各小类还可以进一步细分。推送给用户的内容分类可以是上述的大类、小类、细分中的一种，或者几种的组合。每一个内容分类中，可以有多个推荐访问内容，每个推荐访问内容只能归入一个内容分类中。
S2.当用户的探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，推荐访问内容属于多个内容分类。
当用户的探索概率超过设定的阈值时，则触发对该用户的兴趣探索，首先，获取该用户对最近的推荐访问内容的反馈信息。在一种实施方式中，从内存数据库中提取用户的反馈信息，反馈信息是从客户端获取后存入内存数据库的。其中，反馈信息包括在每个内容分类中，推荐访问内容的曝光次数以及该用户的点击次数。
具体的，在实际应用中，该用户所使用的客户端实时将用户行为日志传回日志回收系统，日志回收系统将用户行为日志写入数据流， 例如Kafka分布式发布订阅消息系统。同时启动一个流式服务实时统计各用户在各自内容分类下的点击次数和曝光次数。在每个统计周期内，实时将统计结果写入redis内存数据库，则该反馈信息可以在离线时实时流泻入redis内存数据库中。
进一步的，确定推荐访问内容的内容分类的方法为：对每个用户未访问过的所有N 0个内容分类按照权重从大到小进行排序，并选取排在前面的N 1个内容分类作为推荐访问内容的内容分类，其中，1＜N 1≤N 0，内容分类的权重通过对所有用户访问过的所有内容分类进行统计得到，可以实现对不同的内容分类进行有差别地探索，以进一步提高探索的成功率。
进一步的，推送给每个用户的内容分类数量可以相同，也可以不同。例如，可以为每个用户预先设定其访问的内容分类总数，则推送给该用户的推荐访问内容的内容分类为内容分类总数与该用户访问过的内容分类数量之差。每次为用户推送多个内容分类，以提高兴趣探索的效率。
在其他的实施方式中，也可以从用户较少访问的内容分类中选择推荐访问内容，并推送给用户，以进一步发掘用户的兴趣或者需求。
用户的反馈信息包括对推荐访问内容的点击、下载、浏览、在线安装、收藏和/或评价(例如评论、评分、点赞等)等用户行为信息。
可以理解的是，设定的阈值根据应用场景来选定。
S3.根据反馈信息从多个内容分类中筛选出继续推送的内容分类。
步骤S3具体包括：计算多个内容分类的得分，将得分最高的内容分类作为继续推送的内容分类。计算多个内容分类的得分可以采用以下算法中的一种或者两种以上的组合，以下算法包括ε-greedy算 法、抽样方法算法、Ranked Bandits算法、Contextual Bandits算法和Reinforcement Learning算法等。
在一种实施方式中，根据置信区间上界算法计算所有内容分类的得分，其计算方法为：
Figure PCTCN2018081315-appb-000008
其中，I j为第j个内容分类的得分，
Figure PCTCN2018081315-appb-000009
为第j个内容分类的平均分，
Figure PCTCN2018081315-appb-000010
T j(n)为对第j个内容分类累计兴趣探索次数，n为对所有内容分类累计兴趣探索的次数，β为第j个内容分类的权重，c为用户对第j个内容分类的点击次数，v为第j个内容分类的曝光次数。
每次出发兴趣探索，选取得分最大的内容分类进行探索。
根据用户的探索概率确定是否对该用户进行兴趣探索，避免盲目地向所有用户进行兴趣探索，从而提高兴趣探索的成功率和用户的转化率。
本发明实施例能够有效地解决推荐系统中的“惊喜度”问题，同时还能最大限度地规避兴趣探索带来的负面效果，能够有效地减少用户的流失，增加用户对产品的粘度。
本发明实施例还提供一种存储介质，存储介质上存储有计算机程序，计算机程序被处理器执行时实现上述兴趣探索方法。需要说明的是，存储介质包括U盘、移动硬盘、ROM(Read-Only Memory，只读存储器)、RAM(Random Access Memory，随机存取存储器)、磁碟或者光盘等各种可以存储程序代码的介质。
参见图2所示，本发明实施例还提供一种电子设备，包括存储器和处理器，存储器上储存有在处理器上运行的计算机程序，处理器执 行计算机程序时实现上述兴趣探索方法。
需要说明的是：本发明实施例提供的系统在进行模块间通信时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将系统的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。
本发明实施例还提供一种兴趣探索系统，该系统包括探索概率计算模块、反馈信息获取模块和筛选模块。
探索概率计算模块用于：根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率。
反馈信息获取模块用于：当用户的探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，推荐访问内容属于多个内容分类。
筛选模块用于：根据反馈信息从多个内容分类中筛选出继续推送的内容分类。
进一步，本发明不局限于上述实施方式，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也视为本发明的保护范围之内。本说明书中未作详细描述的内容属于本领域专业技术人员公知的现有技术。

Claims (10)

  1. 一种兴趣探索方法，其特征在于，该方法包括以下步骤：
    根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率；
    当用户的所述探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，所述推荐访问内容属于多个内容分类；
    根据所述反馈信息从多个所述内容分类中筛选出继续推送的内容分类。
  2. 如权利要求1所述的兴趣探索方法，其特征在于：用户包括注册用户，注册用户的所述探索概率p m的计算方法为：
    Figure PCTCN2018081315-appb-100001
    其中，a为固定概率，0＜a＜1，m为该注册用户访问过的内容分类总数。
  3. 如权利要求1所述的兴趣探索方法，其特征在于：用户包括非注册用户，非注册用户的所述探索概率ρ k的计算方法为：
    Figure PCTCN2018081315-appb-100002
    其中，λ为期望进行兴趣探索的次数，k为对该非注册用户已经进行的兴趣探索的次数。
  4. 如权利要求1所述的兴趣探索方法，其特征在于：对每个用户未访问过的所有N 0个内容分类按照权重从大到小进行排序，并选取排在前面的N 1个内容分类作为所述推荐访问内容的内容分类，其中，1＜N 1≤N 0，所述内容分类的权重通过对所有用户访问过的所有内容分类进行统计得到。
  5. 如权利要求1所述的兴趣探索方法，其特征在于：从内存数据库中提取用户的反馈信息，所述反馈信息是从客户端获取后存入内存数据库的；
    其中，所述反馈信息包括在每个所述内容分类中，所述推荐访问内容的曝光次数以及该用户的点击次数。
  6. 如权利要求1所述的兴趣探索方法，其特征在于，根据反馈信息从多个所述内容分类中筛选出继续推送的内容分类的具体流程包括：
    计算多个所述内容分类的得分，将得分最高的所述内容分类作为继续推送的内容分类。
  7. 如权利要求6所述的兴趣探索方法，其特征在于，根据置信区间上界算法计算所有所述内容分类的得分，其计算方法为：
    Figure PCTCN2018081315-appb-100003
    其中，I j为第j个所述内容分类的得分，
    Figure PCTCN2018081315-appb-100004
    为第j个所述内容分类的平均分，
    Figure PCTCN2018081315-appb-100005
    T j(n)为对第j个所述内容分类累计兴趣探索次数，n为对所有所述内容分类累计兴趣探索的次数，β为第j个所述内容分类的权重，c为用户对第j个所述内容分类的点击次数，v为第j个所述内容分类的曝光次数。
  8. 一种存储介质，该存储介质上存储有计算机程序，其特征在于：所述计算机程序被处理器执行时实现权利要求1至7任一项所述方法的步骤。
  9. 一种电子设备，包括存储器和处理器，存储器上储存有在处理器上运行的计算机程序，其特征在于：处理器执行计算机程序时实现权利要求1至7任一项所述方法的步骤。
  10. 一种兴趣探索系统，其特征在于：该系统包括探索概率计算模块、反馈信息获取模块和筛选模块；
    所述探索概率计算模块用于：根据用户的历史行为信息，计算对每个用户进行兴趣探索的探索概率；
    所述反馈信息获取模块用于：当用户的所述探索概率超过设定的阈值时，获取该用户对最近的推荐访问内容的反馈信息，所述推荐访问内容属于多个内容分类；
    所述筛选模块用于：根据所述反馈信息从多个所述内容分类中筛选出继续推送的内容分类。
PCT/CN2018/081315 2018-01-05 2018-03-30 兴趣探索方法、存储介质、电子设备及系统 WO2019134274A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810012144.7 2018-01-05
CN201810012144.7A CN108595461B (zh) 2018-01-05 2018-01-05 兴趣探索方法、存储介质、电子设备及系统

Publications (1)

Publication Number Publication Date
WO2019134274A1 true WO2019134274A1 (zh) 2019-07-11

Family

ID=63599810

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/081315 WO2019134274A1 (zh) 2018-01-05 2018-03-30 兴趣探索方法、存储介质、电子设备及系统

Country Status (2)

Country Link
CN (1) CN108595461B (zh)
WO (1) WO2019134274A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110704738A (zh) * 2019-09-29 2020-01-17 平安直通咨询有限公司上海分公司 基于法官画像的服务信息推送方法、装置、终端及存储介质
CN112445689A (zh) * 2020-11-27 2021-03-05 广州三七互娱科技有限公司 用户分类方法、装置及电子设备
CN112749332A (zh) * 2020-07-10 2021-05-04 腾讯科技（深圳）有限公司 数据处理方法、装置以及计算机可读介质
CN113344604A (zh) * 2021-04-16 2021-09-03 广州迅捷微风信息科技有限公司 一种基于用户行为数据及流计算的用户细分方法
CN113434660A (zh) * 2021-06-28 2021-09-24 平安银行股份有限公司 基于多领域分类的产品推荐方法、装置、设备及存储介质
CN114036403A (zh) * 2022-01-07 2022-02-11 智者四海（北京）技术有限公司 用户兴趣探测方法、装置和存储介质
CN115379004A (zh) * 2022-07-04 2022-11-22 南京推而广知网络科技有限公司 一种内容创意推送运营系统及其使用方法
CN117056611A (zh) * 2023-10-11 2023-11-14 北京小糖科技有限责任公司 新兴趣的试探方法、装置、电子设备及存储介质
CN117807293A (zh) * 2024-02-23 2024-04-02 中国电子科技集团公司第十研究所 一种证据信息按需组织与精准分发方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109522476A (zh) * 2018-10-30 2019-03-26 广东原昇信息科技有限公司 客户属性离散化指标转化率的静态预测方法
CN110225361B (zh) * 2019-06-03 2021-10-15 武汉瓯越网视有限公司 直播间推荐方法、存储介质、电子设备及系统
CN111026960B (zh) * 2019-12-02 2023-09-01 上海麦克风文化传媒有限公司 一种用户兴趣分类方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101866341A (zh) * 2009-04-17 2010-10-20 华为技术有限公司 一种信息推送方法、装置及系统
CN103607496A (zh) * 2013-11-15 2014-02-26 中国科学院深圳先进技术研究院 一种推断手机用户兴趣爱好的方法、装置及手机终端
CN103870453A (zh) * 2012-12-07 2014-06-18 盛乐信息技术（上海）有限公司 数据推荐方法及系统
WO2017128423A1 (zh) * 2016-01-31 2017-08-03 胡明祥 根据浏览记录推送广告的方法以及推送系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101266620B (zh) * 2008-04-07 2021-04-16 北京大学 向用户提供目标信息的方法及设备
KR102199590B1 (ko) * 2014-02-21 2021-01-07 삼성전자주식회사 관심정보에 따른 컨텐츠 추천장치 및 방법
CN108197331B (zh) * 2015-01-21 2020-02-04 北京字节跳动科技有限公司 用户兴趣的探索方法和装置
CN105912550A (zh) * 2015-12-15 2016-08-31 乐视网信息技术（北京）股份有限公司 一种移动终端的信息推荐方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101866341A (zh) * 2009-04-17 2010-10-20 华为技术有限公司 一种信息推送方法、装置及系统
CN103870453A (zh) * 2012-12-07 2014-06-18 盛乐信息技术（上海）有限公司 数据推荐方法及系统
CN103607496A (zh) * 2013-11-15 2014-02-26 中国科学院深圳先进技术研究院 一种推断手机用户兴趣爱好的方法、装置及手机终端
WO2017128423A1 (zh) * 2016-01-31 2017-08-03 胡明祥 根据浏览记录推送广告的方法以及推送系统

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110704738A (zh) * 2019-09-29 2020-01-17 平安直通咨询有限公司上海分公司 基于法官画像的服务信息推送方法、装置、终端及存储介质
CN110704738B (zh) * 2019-09-29 2023-11-07 平安直通咨询有限公司上海分公司 基于法官画像的服务信息推送方法、装置、终端及存储介质
CN112749332B (zh) * 2020-07-10 2023-12-15 腾讯科技（深圳）有限公司 数据处理方法、装置以及计算机可读介质
CN112749332A (zh) * 2020-07-10 2021-05-04 腾讯科技（深圳）有限公司 数据处理方法、装置以及计算机可读介质
CN112445689A (zh) * 2020-11-27 2021-03-05 广州三七互娱科技有限公司 用户分类方法、装置及电子设备
CN113344604A (zh) * 2021-04-16 2021-09-03 广州迅捷微风信息科技有限公司 一种基于用户行为数据及流计算的用户细分方法
CN113344604B (zh) * 2021-04-16 2023-05-16 广州迅捷微风信息科技有限公司 一种基于用户行为数据及流计算的用户细分方法
CN113434660A (zh) * 2021-06-28 2021-09-24 平安银行股份有限公司 基于多领域分类的产品推荐方法、装置、设备及存储介质
CN114036403A (zh) * 2022-01-07 2022-02-11 智者四海（北京）技术有限公司 用户兴趣探测方法、装置和存储介质
CN115379004A (zh) * 2022-07-04 2022-11-22 南京推而广知网络科技有限公司 一种内容创意推送运营系统及其使用方法
CN117056611A (zh) * 2023-10-11 2023-11-14 北京小糖科技有限责任公司 新兴趣的试探方法、装置、电子设备及存储介质
CN117056611B (zh) * 2023-10-11 2024-01-23 北京小糖科技有限责任公司 新兴趣的试探方法、装置、电子设备及存储介质
CN117807293A (zh) * 2024-02-23 2024-04-02 中国电子科技集团公司第十研究所 一种证据信息按需组织与精准分发方法
CN117807293B (zh) * 2024-02-23 2024-05-14 中国电子科技集团公司第十研究所 一种证据信息按需组织与精准分发方法

Also Published As

Publication number Publication date
CN108595461A (zh) 2018-09-28
CN108595461B (zh) 2021-03-16

Similar Documents

Publication Publication Date Title
WO2019134274A1 (zh) 兴趣探索方法、存储介质、电子设备及系统
CN110795624B (zh) 分析个性化框架
Zhang et al. A collective bayesian poisson factorization model for cold-start local event recommendation
KR101939554B1 (ko) 일시적 거래 한도 결정
CN112052387B (zh) 一种内容推荐方法、装置和计算机可读存储介质
US8966036B1 (en) Method and system for website user account management based on event transition matrixes
US20170236073A1 (en) Machine learned candidate selection on inverted indices
CN105868334B (zh) 一种基于特征递增型的电影个性化推荐方法及系统
US8832105B2 (en) System for incrementally clustering news stories
US9946775B2 (en) System and methods thereof for detection of user demographic information
CN102999588A (zh) 一种多媒体应用的推荐方法和系统
WO2018040069A1 (zh) 信息推荐系统及方法
US20070083343A1 (en) Statistical models for improving the performance of database operations
CN105022761A (zh) 群组查找方法和装置
WO2022095661A1 (zh) 推荐模型的更新方法、装置、计算机设备和存储介质
US10474670B1 (en) Category predictions with browse node probabilities
US10733244B2 (en) Data retrieval system
CN117216362A (zh) 内容推荐方法、装置、设备、介质和程序产品
CN114528480A (zh) 一种基于k均值聚类的时间感知自适应兴趣点推荐方法
Gisselbrecht et al. Whichstreams: A dynamic approach for focused data capture from large social media
Gu et al. CAMF: context aware matrix factorization for social recommendation
WO2017028791A1 (zh) 一种公众号推荐方法及系统
WO2023087933A1 (zh) 内容推荐方法、装置、设备、存储介质及程序产品
CN114329167A (zh) 超参数学习、智能推荐、关键词和多媒体推荐方法及装置
KR101568800B1 (ko) 실시간 이슈 검색어 선별 방법 및 시스템

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18898309

Country of ref document: EP

Kind code of ref document: A1