WO2018218403A1 - 一种内容推送方法及装置 - Google Patents
一种内容推送方法及装置 Download PDFInfo
- Publication number
- WO2018218403A1 WO2018218403A1 PCT/CN2017/086283 CN2017086283W WO2018218403A1 WO 2018218403 A1 WO2018218403 A1 WO 2018218403A1 CN 2017086283 W CN2017086283 W CN 2017086283W WO 2018218403 A1 WO2018218403 A1 WO 2018218403A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- content
- target
- user
- target user
- historical
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000007405 data analysis Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 5
- 238000007418 data mining Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L65/00—Network arrangements, protocols or services for supporting real-time applications in data packet communication
- H04L65/40—Support for services or applications
Definitions
- the invention belongs to the technical field of data analysis and processing, and in particular relates to a content pushing method and device.
- the recommendation system refers to an Internet website that provides users with item information or suggestions, allowing users to discover their potential interests and needs and help users select items.
- the item-based collaborative filtering algorithm is the most widely used recommendation algorithm in the industry. Whether it is Amazon.com or Netflix, Hulu, YouTube, etc., the basis of its recommendation algorithm is the algorithm.
- the advantages are as follows: 1. The calculation is simple; 2. The recommendation reason can be summarized according to the user's historical behavior; 3. The more the user behavior history, the higher the recommendation efficiency.
- the shortcomings are mainly 1. When the number of items is much larger than the user, the calculation of the co-occurrence matrix of the item is too expensive; 2. The cold start problem is serious; 3. When the seed item is selected for the calculation recommendation list, the time information is not considered.
- the technical problem to be solved by the embodiments of the present invention is to provide a content pushing method and device, which aim to solve the problem of inaccurate calculation of user interest in the prior art.
- a first aspect of the embodiments of the present invention provides a content pushing method, where the method includes:
- the content viewing history data of the user includes all historical content of the user and a viewing time point of each of the historical content, and the historical content is content viewed by the user;
- a second aspect of the embodiments of the present invention provides a content pushing apparatus, where the apparatus includes:
- An acquisition module configured to acquire content viewing history data of all users, where the user's content viewing history data includes all historical content of the user and a viewing time point of each of the historical content, where the historical content is content viewed by the user;
- a processing module configured to determine content associated with the historical content of the target user as the target content, calculate a similarity between the target content and the associated historical content of the target user, and acquire the target user pair and the target a user score of the historical content of the target user associated with the content, and calculating, according to a viewing time point of each of the historical content, a behavior time weight of the target user viewing the historical content of the target user associated with the target content;
- a calculating module configured to calculate, according to the similarity, the user score, and the behavior time weight, the degree of interest of the target user on the target content
- the pushing module is configured to select a preset number of the target content with the highest interest of the target user, and push the target content to the target user.
- the present invention determines the content associated with the historical content of the target user as the target content by acquiring the content viewing history data of all users, and calculates the similarity between the target content and the historical content of the associated target user. Obtaining a user rating of the historical content of the target user associated with the target content by the target user, and calculating, according to the viewing time point of each historical content, a behavior time weight of the target user viewing the historical content of the target user associated with the target content, according to the similarity User rating and behavior time weight, calculating the target user's interest in the target content, selecting the preset number of target content with the highest target user interest, and pushing it to the target user.
- the solution is to obtain the user push.
- the parameter of the behavior time weight of the user history content is introduced, so that the statistics of the user interest degree are more accurate, and the obtained user pushes the content more accurately.
- FIG. 1 is a schematic flowchart of an implementation process of a content pushing method according to a first embodiment of the present invention
- FIG. 2 is a schematic flowchart of an implementation process of a content pushing method according to a second embodiment of the present invention
- FIG. 3 is a schematic structural diagram of a content pushing apparatus according to a third embodiment of the present invention.
- FIG. 4 is a schematic structural diagram of a content pushing apparatus according to a fourth embodiment of the present invention.
- Figure 5 is a user behavior history matrix provided by a second embodiment of the present invention.
- FIG. 6 is a schematic diagram of target content interest degree calculation provided by a second embodiment of the present invention.
- FIG. 1 is a schematic flowchart of an implementation process of a content push method according to a first embodiment of the present invention, which may be applied to a terminal device. As shown in FIG. 1, the method mainly includes the following steps:
- the content viewing history data of the user includes all historical content of the user and a viewing time point of each historical content.
- the historical content is the content viewed by the user, that is, the content recorded by the terminal device and viewed by the terminal device.
- the historical content may include, but is not limited to, video, audio, news, or merchandise on the web. The way to view it includes clicking on the link to the historical content.
- S102 Determine content related to the historical content of the target user as the target content, calculate a similarity between the target content and the historical content of the associated target user, and obtain a user rating of the historical content of the target user associated with the target content by the target user. Calculating, according to the viewing time point of each historical content, a behavior time weight of the target user viewing the historical content of the target user associated with the target content;
- the content is considered to be the content associated with the historical content of the target user, and the content is determined as the target content.
- the user rating is 1 by default.
- S104 Select a preset number of target content with the highest interest of the target user, and push the target content to the target user.
- the preset number here can be set and changed as needed.
- the content pushing method determines the content associated with the historical content of the target user as the target content by acquiring the content viewing history data of all users, and calculates the similarity between the target content and the historical content of the associated target user. Obtaining a user rating of the historical content of the target user associated with the target content by the target user, and calculating the target user according to the viewing time point of each historical content Observing the behavior time weight of the historical content of the target user associated with the target content, calculating the interest degree of the target user to the target content according to the similarity, the user rating, and the behavior time weight, and selecting a preset number of target content with the highest target user interest degree And pushed to the target user, compared with the prior art, in the process of obtaining the user's push content, the scheme introduces the parameter of the behavior time weight of the user historical content when calculating the user's interest degree, so that the user interest degree is counted. More accurate, which in turn makes the user's push content more accurate.
- FIG. 2 is a schematic flowchart of an implementation process of a content push method according to a second embodiment of the present invention, which may be applied to a terminal device. As shown in FIG. 2, the method mainly includes the following steps:
- the content viewing history data of the user includes all historical content of the user and a viewing time point of each historical content.
- the historical content is the content viewed by the user, that is, the content recorded by the terminal device and viewed by the terminal device.
- the historical content may include, but is not limited to, video, audio, news, or merchandise on the web. The way to view it includes clicking on the link to the historical content.
- the content is considered to be the content associated with the historical content of the target user, and the content is determined as the target content.
- FIG. 5 is a user behavior history matrix established by the terminal device, where A, B, C, D, and E are users, and a, b, c, d, and e are historical contents.
- w ij is the similarity between the target content and the historical content of the associated target user
- N(i) is the number of users who have viewed the historical content i of the target user associated with the target content among all users
- N(j) is all users.
- the number of users who have viewed the target content j, N(i) ⁇ N(j) is the number of users who have viewed i and j at the same time.
- the user score is 1 by default.
- ⁇ is the interest attenuation factor, which can be adjusted as needed.
- t ui is the target user u to view the logical distance between the historical content i of the target user associated with the target content and the latest behavior of the target user.
- the latest behavior of the target user is that the target user views the historical content of the target user and the viewing time point is closest to the current time point.
- the behavior of the content The number of historical content of the target user between the target user u i see the point of view of time and the target user u view the latest behavior of the target user's point of view the more time a long time, the greater the value of t ui, t ui a non-negative integer.
- P uj is the degree of interest of the target user u to the target content j
- N(u) is the set of all historical contents of the target user u
- S(j, K) is the highest similarity to the target content j in the historical content of the target user u.
- the set of K historical contents, w ij is the similarity between the target content j and the historical content i of the target user u
- r ui is the target user u scores the user of the historical content i of the target user
- l ui is the target user u view
- the preset number here can be set and changed as needed.
- P uf User u's interest in f is P uf , and P uf is calculated as follows:
- the calculated interest levels are sorted in descending order, and the top TopN target content is recommended to the user. For example, if Top 5 target content is recommended to the user, the recommended list is [i, j, f, k, a].
- the content pushing method determines the content associated with the historical content of the target user as the target content by acquiring the content viewing history data of all users, and calculates the similarity between the target content and the historical content of the associated target user.
- the user score of the historical content of the user is calculated according to the viewing time point of each historical content, and the behavior time weight of the target user's historical content associated with the target content is calculated, and the target user is calculated according to the similarity, the user rating, and the behavior time weight.
- the preset number of target content with the highest target user interest is selected and pushed to the target user.
- the solution calculates the user's interest level in the process of obtaining the user's push content.
- the parameter of the behavior time weight of the user history content is introduced, so that the statistics of the user interest degree are more accurate, and the obtained user pushes the content more accurately.
- FIG. 3 is a schematic structural diagram of a content pushing apparatus according to a third embodiment of the present invention.
- the content pushing device illustrated in FIG. 3 may be an execution body of the content pushing method provided by the foregoing first embodiment, which may be a function module of a terminal device or a terminal device.
- the content pushing device illustrated in FIG. 3 mainly includes an obtaining module 301, a processing module 302, a calculating module 303, and a pushing module 304.
- Each function module is described in detail as follows:
- the obtaining module 301 is configured to obtain content viewing history data of all users.
- the content viewing history data of the user includes all historical content of the user and a viewing time point of each historical content, and the historical content is content viewed by the user.
- the processing module 302 is configured to determine the content associated with the historical content of the target user as the target content, calculate the similarity between the target content and the historical content of the associated target user, and obtain the history of the target user associated with the target content.
- the user rating of the content is calculated according to the viewing time point of each historical content, and the behavior time weight of the target user viewing the historical content of the target user associated with the target content is calculated.
- the calculating module 303 is configured to calculate, according to the similarity, the user score, and the behavior time weight, the target user's interest in the target content.
- the pushing module 304 is configured to select a preset number of target content with the highest degree of interest of the target user, and push the target content to the target user.
- the content pushing device determines the content associated with the historical content of the target user as the target content by acquiring the content viewing history data of all the users, and calculates the similarity between the target content and the historical content of the associated target user. Obtaining a user rating of the historical content of the target user associated with the target content by the target user, and calculating, according to the viewing time point of each historical content, a behavior time weight of the target user viewing the historical content of the target user associated with the target content, according to the similarity User rating and behavior time weight, calculating the target user's interest in the target content, selecting the preset number of target content with the highest target user interest, and pushing it to the target user.
- the solution is to obtain the user push.
- the parameter of the behavior time weight of the user history content is introduced, so that the statistics of the user interest degree are more accurate, and the obtained user pushes the content more accurately.
- FIG. 4 is a schematic structural diagram of a content pushing apparatus according to a fourth embodiment of the present invention.
- the content pushing device illustrated in FIG. 4 may be the execution body of the content pushing method provided by the foregoing second embodiment, which may be a function module of the terminal device or the terminal device.
- the content pushing device illustrated in FIG. 4 mainly includes an obtaining module 401, a processing module 402, a calculating module 403, and a pushing module 404.
- Each function module is described in detail as follows:
- the obtaining module 401 is configured to obtain content viewing history data of all users.
- the content viewing history data of the user includes all historical content of the user and a viewing time point of each historical content, and the historical content is content viewed by the user.
- the processing module 402 is configured to determine content associated with the historical content of the target user as the target content, and view the historical data according to the obtained content of all users, and establish a user behavior history matrix.
- the processing module 402 is further configured to: according to a user behavior history matrix and a formula Calculating the similarity between the target content and the historical content of the associated target user, where w ij is the similarity between the target content and the historical content of the associated target user, and N(i) is the historical content of the target user associated with the target content.
- the number of users of i, N(j) is the number of users who have viewed the target content j
- N(i) ⁇ N(j) is the number of users who have viewed i and j at the same time.
- the processing module 402 is further configured to obtain a user rating of the historical content of the target user associated with the target content by the target user, according to a formula Calculating a behavior time weight of the target user viewing the historical content of the target user associated with the target content, where ⁇ is an interest attenuation factor, and t ui is the target user u viewing the historical content of the target user associated with the target content i from the latest behavior of the target user Logical distance, the latest behavior of the target user is the behavior of the target user to view the content of the target user's historical content that is closest to the current time point.
- a calculation module 403 for using a formula according to Calculating the degree of interest of the target user to the target content, where P uj is the degree of interest of the target user u to the target content j, N(u) is the set of all historical content of the target user u, and S(j, K) is the target user u Among the historical contents, the set of K historical contents having the highest similarity with the target content j, w ij is the similarity between the target content j and the historical content i of the target user u, and r ui is the historical content of the target user u to the target user i The user rating, l ui is the target user u to view the behavior time weight of the historical content i of the target user.
- the pushing module 404 is configured to select a preset number of target content with the highest interest of the target user, and push the target content to the target user.
- the content pushing device determines the content associated with the historical content of the target user as the target content by acquiring the content viewing history data of all the users, and calculates the similarity between the target content and the historical content of the associated target user. Obtaining a user rating of the historical content of the target user associated with the target content by the target user, and calculating, according to the viewing time point of each historical content, a behavior time weight of the target user viewing the historical content of the target user associated with the target content, according to the similarity , user rating and behavior time weight, calculate the target user's interest in the target content, select the target user interest The highest preset number of target content is pushed to the target user.
- the scheme introduces the behavior time weight of the user historical content when calculating the user's interest degree in the process of obtaining the user's push content.
- a parameter makes the statistics of user interest more accurate, which makes the obtained user push the content more accurately.
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Information Transfer Between Computers (AREA)
Abstract
本发明适用于数据分析与处理技术领域，提供了一种内容推送方法及装置。该方法包括：获取全部用户的内容查看历史数据，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重，计算目标用户对目标内容的兴趣度，选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户。相较于现有技术，本方案在获取用户推送内容过程中，在计算用户的兴趣度时，引入了用户历史内容的行为时间权重这一参数，使用户兴趣度的统计更为准确，进而使获取的用户推送内容更为精确。
Description
本发明属于数据分析与处理技术领域，尤其涉及一种内容推送方法及装置。
随着逐渐步入信息时代，当今世界正处于信息大爆炸的环境下，同时面临着严峻的信息过载问题。仅在2011年，全球数据量就达到了1.8ZB，相当于全世界每人每年能产生200GB以上的数据，并且这个数字还在逐年增长，据保守预计，接下来几年中，数据产生量将始终保持每年50％的增长速度。现如今，在各大电商、视频播放平台、音频播放平台上，用户每天都产生海量的数据，因此如何有效地利用用户产生的数据是当今互联网企业亟需解决的问题。此时，个性化的推荐系统作为数据挖掘的手段便应运而生了。推荐系统是指互联网网站向用户提供物品信息或建议，让用户发现自己潜在的兴趣和需求并帮助用户选择物品。
基于物品的协同过滤(item-based collaborative filtering)算法是目前业界应用最多的推荐算法。无论是亚马逊网，还是Netflix、Hulu、YouTube等，其推荐算法的基础都是该算法。其优点有：1.计算简单；2.可以根据用户历史行为归纳推荐理由；3.用户行为历史越多推荐效率越高。其缺点主要有1.物品数量远大于用户时，计算物品共现矩阵代价太大；2.冷启动问题严重；3.选取种子物品进行计算推荐列表时，没有考虑时间信息。
发明内容
本发明实施例所要解决的技术问题在于提供一种内容推送方法及装置，旨在解决现有技术中用户兴趣度计算不精确的问题。
本发明实施例第一方面提供了一种内容推送方法，所述方法包括：
获取全部用户的内容查看历史数据，所述用户的内容查看历史数据包括用户的全部历史内容及各所述历史内容的查看时间点，所述历史内容为用户查看过的内容；
将与目标用户的历史内容相关联的内容确定为目标内容，计算所述目标内容与关联的所述目标用户的历史内容的相似度，获取所述目标用户对与所述目标内容关联的所述目标用户的历史内容的用户评分，根据各所述历史内容的查看时间点，计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重；
根据所述相似度、所述用户评分及所述行为时间权重，计算所述目标用户对所述目标内容的兴趣度；
选取所述目标用户兴趣度最高的预置数量个所述目标内容，推送给所述目标用户。
本发明实施例第二方面提供了一种内容推送装置，所述装置包括：
获取模块，用于获取全部用户的内容查看历史数据，所述用户的内容查看历史数据包括用户的全部历史内容及各所述历史内容的查看时间点，所述历史内容为用户查看过的内容；
处理模块，用于将与目标用户的历史内容相关联的内容确定为目标内容，计算所述目标内容与关联的所述目标用户的历史内容的相似度，获取所述目标用户对与所述目标内容关联的所述目标用户的历史内容的用户评分，根据各所述历史内容的查看时间点，计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重；
计算模块，用于根据所述相似度、所述用户评分及所述行为时间权重，计算所述目标用户对所述目标内容的兴趣度；
推送模块，用于选取所述目标用户兴趣度最高的预置数量个所述目标内容，推送给所述目标用户。
从上述本发明实施例可知，本发明通过获取全部用户的内容查看历史数据，将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重，根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度，选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户，相较于现有技术，本方案在获取用户推送内容过程中，在计算用户的兴趣度时，引入了用户历史内容的行为时间权重这一参数，使用户兴趣度的统计更为准确，进而使获取的用户推送内容更为精确。
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
附图1是本发明第一实施例提供的内容推送方法的实现流程示意图；
附图2是本发明第二实施例提供的内容推送方法的实现流程示意图；
附图3是本发明第三实施例提供的内容推送装置的结构示意图；
附图4是本发明第四实施例提供的内容推送装置的结构示意图；
附图5是本发明第二实施例提供的用户行为历史矩阵；
附图6是本发明第二实施例提供的目标内容兴趣度计算的示意图。
为使得本发明实施例的发明目的、特征、优点能够更加的明显和易懂，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完
整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而非全部实施例。基于本发明中的实施例，本领域技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
请参阅附图1，附图1为本发明第一实施例提供的内容推送方法的实现流程示意图，该方法可以应用于终端设备中。如附图1所示，该方法主要包括以下步骤：
S101、获取全部用户的内容查看历史数据；
其中，用户的内容查看历史数据包括用户的全部历史内容及各历史内容的查看时间点。进一步地，该历史内容为用户查看过的内容，即终端设备记录下的用户之前通过该终端设备查看过的内容。该历史内容可以但不限于包括：网络上的视频、音频、新闻或商品。查看的方式包括点击该历史内容的链接。
S102、将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重；
若存在用户既查看过某一内容，又目标用户的历史内容，则认为该内容为与目标用户的历史内容相关联的内容，将该内容确定为目标内容。
当目标用户对目标用户的历史内容无用户评分时，默认该用户评分为1。
S103、根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度；
S104、选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户。
可以理解的，此处的预置数量可以根据需要进行设置、更改。
本发明实施例提供的内容推送方法，通过获取全部用户的内容查看历史数据，将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查
看与目标内容关联的目标用户的历史内容的行为时间权重，根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度，选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户，相较于现有技术，本方案在获取用户推送内容过程中，在计算用户的兴趣度时，引入了用户历史内容的行为时间权重这一参数，使用户兴趣度的统计更为准确，进而使获取的用户推送内容更为精确。
请参阅附图2，附图2为本发明第二实施例提供的内容推送方法的实现流程示意图，该方法可以应用于终端设备中。如附图2所示，该方法主要包括以下步骤：
S201、获取全部用户的内容查看历史数据；
其中，用户的内容查看历史数据包括用户的全部历史内容及各历史内容的查看时间点。进一步地，该历史内容为用户查看过的内容，即终端设备记录下的用户之前通过该终端设备查看过的内容。该历史内容可以但不限于包括：网络上的视频、音频、新闻或商品。查看的方式包括点击该历史内容的链接。
S202、将与目标用户的历史内容相关联的内容确定为目标内容；
若存在用户既查看过某一内容，又目标用户的历史内容，则认为该内容为与目标用户的历史内容相关联的内容，将该内容确定为目标内容。
S203、根据获取的全部用户的内容查看历史数据，建立用户行为历史矩阵；
如图5所示，图5为终端设备建立的一个用户行为历史矩阵，其中A、B、C、D、E为用户，a、b、c、d、e为历史内容。
S204、根据用户行为历史矩阵及公式
计算目标内容与关联的目标用户的历史内容的相似度；
其中wij为目标内容与关联的目标用户的历史内容的相似度，N(i)为全部用户中查看过与目标内容关联的目标用户的历史内容i的用户数量，N(j)为全部用户中查看过目标内容j的用户数量，N(i)∩N(j)为同时查看过i和j
的用户数量。将用户行为历史矩阵中统计的数据带入公式
中，计算出目标内容与关联的目标用户的历史内容的相似度。以用户行为历史矩阵是图5为例，假设目标内容是a，则
S205、获取目标用户对与目标内容关联的目标用户的历史内容的用户评分；
其中，当目标用户对目标用户的历史内容无用户评分时，默认该用户评分为1。
S206、根据公式
计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重；
δ为兴趣衰减因子，该兴趣衰减因子可以根据需要进行调整。tui为目标用户u查看与目标内容关联的目标用户的历史内容i距离目标用户最新行为的逻辑距离，目标用户最新行为是目标用户查看目标用户的历史内容中查看时间点距当前时间点最近的内容的行为。目标用户u查看i的查看时间点与目标用户u查看目标用户最新行为的查看时间点之间的目标用户的历史内容的个数越多时，tui的值越大，tui为非负整数。
S207、根据公式
计算目标用户对目标内容的兴趣度；
Puj为目标用户u对目标内容j的兴趣度，N(u)为目标用户u的全部历史内容的集合，S(j，K)为目标用户u的历史内容中与目标内容j相似度最高的K个历史内容的集合，wij为目标内容j与目标用户u的历史内容i的相似度，rui为目标用户u对目标用户的历史内容i的用户评分，lui为目标用户u查看目标用户的历史内容i的行为时间权重。
S208、选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户。
可以理解的，此处的预置数量可以根据需要进行设置、更改。
以图6为例，假设目标用户u按查看时间从早到晚的顺序查看的全部历史内容分别为：A、B、C、D，此时D即为目标用户最新行为，则可以令tuD＝0、tuC＝1、tuB＝2、tuA＝3。用户对A、B、C、D的评分分别为0.7、0.6、0.5和0.8，这里取兴趣衰减因子δ＝10、取K＝3，与A相似度最高的3个视频分别是a、b、c，waA＝0.9、wbA＝0.8、wcA＝0.7；与B相似度最高的3个视频分别是d、e、f，wdB＝0.7、weB＝0.6、wfB＝0.5；与C相似度最高的3个视频分别是f、g、h，wfC＝0.6、wgC＝0.5、whC＝0.4；与D相似度最高的3个视频分别是i、j、k，wiD＝0.8、wjD＝0.7、wkD＝0.6，其中f既是与B相似度最高的3个视频中的一个，又是与C相似度最高的3个视频中的一个。
用户u对a的兴趣度为Pua，Pua计算如下：
用户u对f的兴趣度为Puf，Puf计算如下：
根据计算得，Pua＝0.467、Pub＝0.415、Puc＝0.363、Pud＝0.344、Pue＝0.295、Puf＝0.517、Pug＝0.226、Puh＝0.181、Pui＝0.64、Puj＝0.56、Puk＝0.48。
将计算得到的兴趣度降序排列，并取前TopN个目标内容推荐给用户。例如，如果取Top5个目标内容推荐给用户时，推荐列表为[i，j，f，k，a]。
本发明实施例提供的内容推送方法，通过获取全部用户的内容查看历史数据，将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标
用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重，根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度，选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户，相较于现有技术，本方案在获取用户推送内容过程中，在计算用户的兴趣度时，引入了用户历史内容的行为时间权重这一参数，使用户兴趣度的统计更为准确，进而使获取的用户推送内容更为精确。
请参阅附图3，附图3是本发明第三实施例提供的内容推送装置的结构示意图，为了便于说明，仅示出了与本发明实施例相关的部分。附图3示例的内容推送装置可以是前述第一实施例提供的内容推送方法的执行主体，其可以是终端设备或者终端设备中的一个功能模块。附图3示例的内容推送装置，主要包括：获取模块301、处理模块302、计算模块303及推送模块304。各功能模块详细说明如下：
获取模块301，用于获取全部用户的内容查看历史数据，用户的内容查看历史数据包括用户的全部历史内容及各历史内容的查看时间点，历史内容为用户查看过的内容。
处理模块302，用于将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重。
计算模块303，用于根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度。
推送模块304，用于选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户。
上述各功能模块实现各自功能的具体过程，可参考前述第一实施例提供的内容推送方法的相关内容，此处不再赘述。
本发明实施例提供的内容推送装置，通过获取全部用户的内容查看历史数据，将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重，根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度，选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户，相较于现有技术，本方案在获取用户推送内容过程中，在计算用户的兴趣度时，引入了用户历史内容的行为时间权重这一参数，使用户兴趣度的统计更为准确，进而使获取的用户推送内容更为精确。
请参阅附图4，附图4是本发明第四实施例提供的内容推送装置的结构示意图，为了便于说明，仅示出了与本发明实施例相关的部分。附图4示例的内容推送装置可以是前述第二实施例提供的内容推送方法的执行主体，其可以是终端设备或者终端设备中的一个功能模块。附图4示例的内容推送装置，主要包括：获取模块401、处理模块402、计算模块403及推送模块404。各功能模块详细说明如下：
获取模块401，用于获取全部用户的内容查看历史数据，用户的内容查看历史数据包括用户的全部历史内容及各历史内容的查看时间点，历史内容为用户查看过的内容。
处理模块402，用于将与目标用户的历史内容相关联的内容确定为目标内容，根据获取的全部用户的内容查看历史数据，建立用户行为历史矩阵。
处理模块402，还用于根据用户行为历史矩阵及公式
计算目标内容与关联的目标用户的历史内容的相似度，其中wij为目标内容与关联的目标用户的历史内容的相似度，N(i)为查看过与目标内容关联的目标用户的历史内容i的用户数量，N(j)为查看过目标内容j的用户数量，N(i)∩N(j)
为同时查看过i和j的用户数量。
处理模块402，还用于获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据公式
计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重，其中δ为兴趣衰减因子，tui为目标用户u查看与目标内容关联的目标用户的历史内容i距离目标用户最新行为的逻辑距离，目标用户最新行为是目标用户查看目标用户的历史内容中查看时间点距当前时间点最近的内容的行为。
计算模块403，用于根据公式
计算目标用户对目标内容的兴趣度，其中Puj为目标用户u对目标内容j的兴趣度，N(u)为目标用户u的全部历史内容的集合，S(j，K)为目标用户u的历史内容中与目标内容j相似度最高的K个历史内容的集合，wij为目标内容j与目标用户u的历史内容i的相似度，rui为目标用户u对目标用户的历史内容i的用户评分，lui为目标用户u查看目标用户的历史内容i的行为时间权重。
推送模块404，用于选取目标用户兴趣度最高的预置数量个目标内容，推送给目标用户。
上述各功能模块实现各自功能的具体过程，可参考前述第二实施例提供的内容推送方法的相关内容，此处不再赘述。
本发明实施例提供的内容推送装置，通过获取全部用户的内容查看历史数据，将与目标用户的历史内容相关联的内容确定为目标内容，计算目标内容与关联的目标用户的历史内容的相似度，获取目标用户对与目标内容关联的目标用户的历史内容的用户评分，根据各历史内容的查看时间点，计算目标用户查看与目标内容关联的目标用户的历史内容的行为时间权重，根据相似度、用户评分及行为时间权重，计算目标用户对目标内容的兴趣度，选取目标用户兴趣
度最高的预置数量个目标内容，推送给目标用户，相较于现有技术，本方案在获取用户推送内容过程中，在计算用户的兴趣度时，引入了用户历史内容的行为时间权重这一参数，使用户兴趣度的统计更为准确，进而使获取的用户推送内容更为精确。
需要说明的是，对于前述的各方法实施例，为了简便描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本发明并不受所描述的动作顺序的限制，因为依据本发明，某些步骤可以采用其它顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定都是本发明所必须的。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其它实施例的相关描述。
以上为对本发明所提供的内容推送方法、装置的描述，对于本领域的技术人员，依据本发明实施例的思想，在具体实施方式及应用范围上均会有改变之处，综上，本说明书内容不应理解为对本发明的限制。
Claims (10)
- 一种内容推送方法，其特征在于，所述方法包括：获取全部用户的内容查看历史数据，所述用户的内容查看历史数据包括用户的全部历史内容及各所述历史内容的查看时间点，所述历史内容为用户查看过的内容；将与目标用户的历史内容相关联的内容确定为目标内容，计算所述目标内容与关联的所述目标用户的历史内容的相似度，获取所述目标用户对与所述目标内容关联的所述目标用户的历史内容的用户评分，根据各所述历史内容的查看时间点，计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重；根据所述相似度、所述用户评分及所述行为时间权重，计算所述目标用户对所述目标内容的兴趣度；选取所述目标用户兴趣度最高的预置数量个所述目标内容，推送给所述目标用户。
- 如权利要求1所述的内容推送方法，其特征在于，所述根据所述相似度、所述用户评分及所述行为时间权重，计算所述目标用户对所述目标内容的兴趣度，包括：根据公式计算所述目标用户对所述目标内容的兴趣度，其中Puj为所述目标用户u对所述目标内容j的兴趣度，N(u)为所述目标用户u的全部历史内容的集合，S(j，K)为所述目标用户u的历史内容中与所述目标内容j相似度最高的K个历史内容的集合，wij为所述目标内容j与所述目标用户u的历史内容i的相似度，rui为所述目标用户u对所述目标用户的历史内容i的用户评分，lui为所述目标用户u查看所述目标用户的历史内容i 的行为时间权重。
- 如权利要求2所述的内容推送方法，其特征在于，所述计算所述目标内容与关联的所述目标用户的历史内容的相似度，包括：根据获取的全部用户的内容查看历史数据，建立用户行为历史矩阵；根据所述用户行为历史矩阵及公式计算所述目标内容与关联的所述目标用户的历史内容的相似度，其中wij为所述目标内容与关联的所述目标用户的历史内容的相似度，N(i)为查看过与所述目标内容关联的所述目标用户的历史内容i的用户数量，N(j)为查看过所述目标内容j的用户数量，N(i)∩N(j)为同时查看过i和j的用户数量。
- 如权利要求2所述的内容推送方法，其特征在于，所述根据各所述历史内容的查看时间点，计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重，包括：根据公式计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重，其中δ为兴趣衰减因子，tui为所述目标用户u查看与所述目标内容关联的所述目标用户的历史内容i距离目标用户最新行为的逻辑距离，所述目标用户最新行为是所述目标用户查看所述目标用户的历史内容中查看时间点距当前时间点最近的内容的行为。
- 如权利要求2至4任一项所述的内容推送方法，其特征在于，当所述目标用户u对所述目标用户的历史内容i无用户评分时，设定rui的值为1。
- 一种内容推送装置，其特征在于，所述装置包括：获取模块，用于获取全部用户的内容查看历史数据，所述用户的内容查看历史数据包括用户的全部历史内容及各所述历史内容的查看时间点，所述历史内容为用户查看过的内容；处理模块，用于将与目标用户的历史内容相关联的内容确定为目标内容，计算所述目标内容与关联的所述目标用户的历史内容的相似度，获取所述目标用户对与所述目标内容关联的所述目标用户的历史内容的用户评分，根据各所述历史内容的查看时间点，计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重；计算模块，用于根据所述相似度、所述用户评分及所述行为时间权重，计算所述目标用户对所述目标内容的兴趣度；推送模块，用于选取所述目标用户兴趣度最高的预置数量个所述目标内容，推送给所述目标用户。
- 如权利要求6所述的内容推送装置，其特征在于，所述计算模块，具体用于根据公式计算所述目标用户对所述目标内容的兴趣度，其中Puj为所述目标用户u对所述目标内容j的兴趣度，N(u)为所述目标用户u的全部历史内容的集合，S(j，K)为所述目标用户u的历史内容中与所述目标内容j相似度最高的K个历史内容的集合，wij为所述目标内容j与所述目标用户u的历史内容i的相似度，rui为所述目标用户u对所述目标用户的历史内容i的用户评分，lui为所述目标用户u查看所述目标用户的历史内容i的行为时间权重。
- 如权利要求7所述的内容推送装置，其特征在于，所述处理模块，还用于根据获取的全部用户的内容查看历史数据，建立用户行为历史矩阵；根据所述用户行为历史矩阵及公式计算所述目标内容与关联的所述目标用户的历史内容的相似度，其中wij为所述目标内容与关联的所述目标用户的历史内容的相似度，N(i)为查看过与所述目标内容关联的 所述目标用户的历史内容i的用户数量，N(j)为查看过所述目标内容j的用户数量，N(i)∩N(j)为同时查看过i和j的用户数量。
- 如权利要求7所述的内容推送装置，其特征在于，所述处理模块，还用于根据公式计算所述目标用户查看与所述目标内容关联的所述目标用户的历史内容的行为时间权重，其中δ为兴趣衰减因子，tui为所述目标用户u查看与所述目标内容关联的所述目标用户的历史内容i距离目标用户最新行为的逻辑距离，所述目标用户最新行为是所述目标用户查看所述目标用户的历史内容中查看时间点距当前时间点最近的内容的行为。
- 如权利要求7至9任一项所述的内容推送装置，其特征在于，当所述目标用户u对所述目标用户的历史内容i无用户评分时，设定rui的值为1。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2017/086283 WO2018218403A1 (zh) | 2017-05-27 | 2017-05-27 | 一种内容推送方法及装置 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2017/086283 WO2018218403A1 (zh) | 2017-05-27 | 2017-05-27 | 一种内容推送方法及装置 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2018218403A1 true WO2018218403A1 (zh) | 2018-12-06 |
Family
ID=64454329
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2017/086283 WO2018218403A1 (zh) | 2017-05-27 | 2017-05-27 | 一种内容推送方法及装置 |
Country Status (1)
| Country | Link |
|---|---|
| WO (1) | WO2018218403A1 (zh) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080398A (zh) * | 2019-11-19 | 2020-04-28 | 浙江大搜车软件技术有限公司 | 商品推荐方法、装置、计算机设备和存储介质 |
| CN111460281A (zh) * | 2020-02-27 | 2020-07-28 | 浙江口碑网络技术有限公司 | 信息推送的优化方法及装置、存储介质、终端 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103617540A (zh) * | 2013-10-17 | 2014-03-05 | 浙江大学 | 一种追踪用户兴趣变化的电子商务推荐方法 |
| CN104281956A (zh) * | 2014-10-27 | 2015-01-14 | 南京信息工程大学 | 基于时间信息的适应用户兴趣变化的动态推荐方法 |
| US9361583B1 (en) * | 2013-03-12 | 2016-06-07 | Trulia, Llc | Merged recommendations of real estate listings |
| CN106339502A (zh) * | 2016-09-18 | 2017-01-18 | 电子科技大学 | 一种基于用户行为数据分片聚类的建模推荐方法 |
| CN107277115A (zh) * | 2017-05-27 | 2017-10-20 | 深圳大学 | 一种内容推送方法及装置 |
-
2017
- 2017-05-27 WO PCT/CN2017/086283 patent/WO2018218403A1/zh unknown
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9361583B1 (en) * | 2013-03-12 | 2016-06-07 | Trulia, Llc | Merged recommendations of real estate listings |
| CN103617540A (zh) * | 2013-10-17 | 2014-03-05 | 浙江大学 | 一种追踪用户兴趣变化的电子商务推荐方法 |
| CN104281956A (zh) * | 2014-10-27 | 2015-01-14 | 南京信息工程大学 | 基于时间信息的适应用户兴趣变化的动态推荐方法 |
| CN106339502A (zh) * | 2016-09-18 | 2017-01-18 | 电子科技大学 | 一种基于用户行为数据分片聚类的建模推荐方法 |
| CN107277115A (zh) * | 2017-05-27 | 2017-10-20 | 深圳大学 | 一种内容推送方法及装置 |
Non-Patent Citations (1)
| Title |
|---|
| HE, LEI: "Intelligent tourism information pushing system based on cloud platform", CHINESE MASTER'S THESES FULL-TEXT DATABASE, 15 October 2014 (2014-10-15) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080398A (zh) * | 2019-11-19 | 2020-04-28 | 浙江大搜车软件技术有限公司 | 商品推荐方法、装置、计算机设备和存储介质 |
| CN111080398B (zh) * | 2019-11-19 | 2024-04-05 | 浙江大搜车软件技术有限公司 | 商品推荐方法、装置、计算机设备和存储介质 |
| CN111460281A (zh) * | 2020-02-27 | 2020-07-28 | 浙江口碑网络技术有限公司 | 信息推送的优化方法及装置、存储介质、终端 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102192863B1 (ko) | 정보 권고 방법 및 장치 | |
| US20190197416A1 (en) | Information recommendation method, apparatus, and server based on user data in an online forum | |
| KR101764696B1 (ko) | 사용자 영향력 및 시간 변화를 고려한 소셜 네트워크 핫 토픽 결정 방법 및 시스템 | |
| CN104462560B (zh) | 一种个性化推荐系统的推荐方法 | |
| AU2012294704B2 (en) | Filtering social search results | |
| RU2731654C1 (ru) | Способ и система для создания пуш-уведомлений, связанных с цифровыми новостями | |
| US10331749B2 (en) | Selective presentation of content types and sources in search | |
| JP5798022B2 (ja) | レコメンド装置、レコメンドシステム、レコメンド方法およびプログラム | |
| CN103577593B (zh) | 一种基于微博热门话题的视频聚合方法及系统 | |
| CN107277115A (zh) | 一种内容推送方法及装置 | |
| US9946799B2 (en) | Federated search page construction based on machine learning | |
| CN105224529A (zh) | 一种基于用户浏览行为的个性化推荐方法和装置 | |
| US9454750B2 (en) | Techniques for estimating distance between members of a social network service | |
| CN105608121B (zh) | 一种个性化推荐方法及装置 | |
| US10331734B2 (en) | Method and apparatus for recommending network service | |
| US9117250B2 (en) | Methods and systems for recommending social network connections | |
| TW201248435A (en) | Method and apparatus of providing suggested terms | |
| CN104902292B (zh) | 一种基于电视报道的舆情分析方法和系统 | |
| US10127322B2 (en) | Efficient retrieval of fresh internet content | |
| CN103218366A (zh) | 下载资源推荐方法及系统 | |
| WO2018218403A1 (zh) | 一种内容推送方法及装置 | |
| CN110992127B (zh) | 一种物品推荐方法及装置 | |
| US20150169794A1 (en) | Updating location relevant user behavior statistics from classification errors | |
| US8856112B2 (en) | Considering document endorsements when processing queries | |
| US8700628B1 (en) | Personalized aggregation of annotations |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17912178 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 09.03.2020) |