WO2020073526A1 - 基于信任网络的推送方法、装置、计算机设备及存储介质 - Google Patents

基于信任网络的推送方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2020073526A1
WO2020073526A1 PCT/CN2018/124954 CN2018124954W WO2020073526A1 WO 2020073526 A1 WO2020073526 A1 WO 2020073526A1 CN 2018124954 W CN2018124954 W CN 2018124954W WO 2020073526 A1 WO2020073526 A1 WO 2020073526A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
trust
value
trusted
row vector
Prior art date
Application number
PCT/CN2018/124954
Other languages
English (en)
French (fr)
Inventor
吴壮伟
Original Assignee
平安科技(深圳)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 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2020073526A1 publication Critical patent/WO2020073526A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0631Item recommendations
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/279Recognition of textual entities
    • G06F40/284Lexical analysis, e.g. tokenisation or collocates
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0609Buyer or seller confidence or verification

Definitions

  • This application relates to the field of information push technology, and in particular, to a push method, device, computer equipment, and storage medium based on a trusted network.
  • Embodiments of the present application provide a method, device, computer device, and storage medium for pushing based on a trusted network, and aim to solve that the recommendation algorithm based on collaborative filtering in the prior art needs to use a scoring matrix to calculate the similarity between users, It is easy to cause a cold start of the recommendation operation, thus affecting the recommendation process.
  • an embodiment of the present application provides a trust network-based push method, which includes:
  • the current trust value is obtained according to the review content corresponding to the newly added mutual review relationship between users
  • each The value indicates the trust value of the user corresponding to the row where the value is located to the user corresponding to the column where the value is located;
  • the commodity recommendation list is obtained from the commodity recommendation row vector, and the commodity recommendation list is pushed to the receiving end corresponding to the target user.
  • an embodiment of the present application provides a trust network-based push device, including:
  • the current trust value acquisition unit is used to determine if there is a new mutual review relationship between users if the product review information at the current time is compared with the product review information at the previous time, according to the new mutual review relationship between the users Comment content to obtain the current trust value;
  • the user trust matrix update unit is used to update the user trust matrix stored at the previous time according to the current trust value to obtain the user trust matrix stored at the current time; wherein, the user trust matrix stored at the previous time and the current stored at the current time In the user trust matrix, each value represents the trust value of the user corresponding to the row of the value to the user corresponding to the column of the value;
  • the trusted user cluster acquisition unit is used to acquire the target user corresponding to the row vector selected from the user trust matrix stored at the current time, and obtain the trust value according to the size of each trust value in the row vector corresponding to the target user at the preset ranking Trust users before the threshold to form a trust user cluster;
  • a recommendation line vector obtaining unit configured to obtain a trust user's trust score for each product according to the corresponding score line vector of each trusted user in the trusted user cluster in the user-scoring matrix to form a product recommendation line vector;
  • the pushing unit is configured to obtain a commodity recommendation list from the commodity recommendation row vector, and push the commodity recommendation list to a receiving end corresponding to the target user.
  • an embodiment of the present application further provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor executing the computer
  • the program implements the push method based on the trusted network described in the first aspect.
  • an embodiment of the present application further provides a storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor causes the processor to perform the above-mentioned first aspect The mentioned push method based on trust network.
  • FIG. 1 is a schematic flowchart of a push method based on a trusted network provided by an embodiment of the present application
  • FIG. 2 is a schematic diagram of a sub-process of a push method based on a trusted network provided by an embodiment of the present application;
  • FIG. 3 is a schematic diagram of another sub-process of a push method based on a trusted network provided by an embodiment of the present application;
  • FIG. 4 is a schematic diagram of another sub-process of a push method based on a trusted network provided by an embodiment of the present application;
  • FIG. 5 is a schematic block diagram of a trusted network-based pushing device provided by an embodiment of the present application.
  • FIG. 6 is a schematic block diagram of a subunit of a push device based on a trusted network provided by an embodiment of the present application;
  • FIG. 7 is a schematic block diagram of another subunit of a push device based on a trusted network according to an embodiment of the present application.
  • FIG. 8 is a schematic block diagram of another subunit of a push device based on a trusted network provided by an embodiment of the present application.
  • FIG. 9 is a schematic block diagram of a computer device provided by an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of a push method based on a trusted network provided by an embodiment of the present application.
  • the push method based on a trusted network is applied to a management server.
  • the method is executed by application software installed in the management server.
  • the management server is an enterprise terminal for pushing based on the trusted network.
  • the method includes steps S110-S150.
  • the current trust value is obtained according to the review content corresponding to the new mutual review relationship between users .
  • the management server can determine the trust relationship between users according to whether there is a review relationship between users and users, And obtain the current trust value according to the comment content corresponding to the newly added mutual comment relationship between users.
  • a user relationship is dominant. For example, in product reviews, user A affirms user B ’s comments, showing that user A trusts user B, or in the community space, user A affirms user B ’s content. Based on this explicit trust network, coupled with the existing historical data, the problem of cold start can be better solved.
  • step S110 includes:
  • the product review information by analyzing the product review information, it is mainly to determine whether there is a mutual review relationship between the users. If there is a review relationship between the two users, the positive, neutral and negative emotions of the review content are determined. .
  • the comment content is first segmented and then converted into a text vector, and the text vector is used as the input of the Naive Bayes model to obtain the emotion recognition result.
  • a decay coefficient which can also be understood as a time decay factor
  • Word2Vec is a model for learning semantic knowledge from a large number of text corpora in an unsupervised manner.
  • User 1 adds a comment below User 4's comment, and the comment is a neutral emotional comment.
  • the difference between user 1's additional comment time t and User 4's comment time t0 is 1 day, then user 1
  • user 1 adds a comment below user 3's comment time t0 is 1 day, then user 1 ’s current
  • step S111 includes:
  • S1111 Select candidate words from the comment content in order from left to right;
  • S1112 Query the probability value corresponding to each candidate word in the pre-stored dictionary, and record the left neighbor word of each candidate word;
  • the word segmentation is performed by a word segmentation method based on a probability statistical model.
  • C C1C2 ... Cm
  • C the Chinese character string to be segmented
  • W W1W2 ... Wn
  • W the result of segmentation
  • Wa, Wb, ... Wk are all possible segments of C Sub-program.
  • the segmentation model based on probability statistics is to find the target word string W, so that W satisfies: P (W
  • C) MAX (P (Wa
  • the word segmentation model, the word string W obtained by the above word segmentation model is the word string whose estimated probability is the largest.
  • each A value represents the trust value of the user corresponding to the row of the value to the user corresponding to the column of the value.
  • the user trust matrix represents the trust value between users. Both the horizontal axis and the vertical axis in the user trust matrix are user lists, and a ij in the user trust matrix An * n refers to user i vs. user j.
  • Trust value all values in the user trust matrix stored at the initial moment are 0, and then re-judge whether there is new comment content from user i to user j every time period T to determine whether to adjust the trust of user i to user j value.
  • the user trust matrix stored at the last moment is that the management server crawled the product review information corresponding to each web page in the preset URL address list at the last moment (such as Z1, Z2, X, Y, 2018). Whether there is a review relationship between users to determine the trust relationship between users, and obtain the trust value of the previous moment according to the review content corresponding to the mutual review relationship between users.
  • the time period T that is separated from the previous time corresponds to the current time (such as Z3, Z4, X, X, Y, 2018)
  • the current time such as Z3, Z4, X, X, Y, 2018
  • the current trust value of user i to user j is denoted as a ij
  • a ij + Sentiment ij * e - ⁇ (t-t0)
  • Self-increasing adjustments for example, user i's comments on user j will be commented every once in a while, then each comment will affect the current trust value
  • the user trust matrix stored at the current time is as follows:
  • the row vector selected by the user trust matrix stored at the current moment is the row vector of the first row
  • the target user corresponding to the row vector of the first row is user 1
  • each value in the row vector of the first row represents user 1 and other users Trust value between.
  • the preset ranking threshold is 3, the trust value size before the third place (that is, the trust value of the top 2) is 3 and 1, respectively corresponding to user 2 and user 4, and the trust user cluster includes user 2 and User 4.
  • the user-rating matrix represents the user's rating of the item (item can be understood as a specific product), the horizontal axis of the user-rating matrix is the item, and the vertical axis is the user, and the value in it is user i to item j 'S rating.
  • the user-score matrix S is a 4 ⁇ 6 matrix, such as:
  • the row vector of the first row in the user-scoring matrix S represents the score of user 1 for items 1 to 5 respectively
  • the row vector of the second row represents the score of user 2 for items 1 to 5 respectively
  • the row vector of the third row Represents user 3's ratings for items 1 to 5 respectively.
  • the user-scoring matrix correspondingly obtains the rating row vectors of user 2 and user 4, that is, the row vector of the second row and the row vector of the fourth row.
  • the product recommendation row vector can be calculated.
  • step S140 includes:
  • S141 Obtain a scoring matrix composed of corresponding scoring row vectors in the user-scoring matrix of each trusted user in the trusted user cluster;
  • the user-scoring matrix correspondingly obtains the rating row vectors of user 2 and user 4, that is, the row vector sum of the second row In the row vector of the fourth row, the rating matrix composed by users 2 and 4 for each item (that is, each commodity) is as follows:
  • step S141 specifically includes: obtaining the corresponding scoring row vector of each trusted user in the trusted user cluster in the user-scoring matrix, and according to the row of the scoring row vector corresponding to each trusted user in the user-scoring matrix
  • the serial numbers are arranged in order to obtain a scoring matrix. According to the order in which each user appears in each row of the user-score matrix, the corresponding score row vector of each trusted user in the user-score matrix is obtained in sequence. The above method can accurately obtain the score row vector corresponding to each trusted user.
  • the scoring matrix facilitates the subsequent calculation of the user's trust score for each product.
  • the trusted user line vector composed of each trusted value in the line vector corresponding to the target user by the trusted user cluster is [3] 1.
  • step S142 specifically includes: acquiring a trusted user whose trust value is before a preset ranking threshold, and placing the trusted user whose trust value is before the ranking threshold according to the sequence number in the row vector corresponding to the target user
  • the order is arranged in order to get the trusted user line vector. According to the order in which the above users appear in the columns of the row vector corresponding to the target user, the trust values in the row vector corresponding to the target user are obtained sequentially, and the trust user row vector is composed. The above method can accurately obtain trust The user row vector is convenient for subsequent calculation of the user's trust score for each product.
  • the product recommendation line vector is [2,13,8,13,7], and the product corresponding to the higher-ranking rating in this product recommendation line vector can be regarded as the product recommended by the trust group to the target user.
  • the product recommendation row vector is obtained by multiplying the trust user row vector and the scoring matrix, which can effectively refer to the comprehensive score of each product by the trust user as a reference index for product recommendation to the target user.
  • the comprehensive score of each product in the trusted user cluster for each product can be known.
  • the ranking of the ratings corresponding to the product recommendation row vector can be selected at the preset The score before the ranking value (such as the preset ranking value of 4), and then obtain the product information corresponding to the score whose ranking is before the preset ranking value in each score, compose the product information into a product recommendation list and push it to the target user's corresponding Receiving end.
  • Obtaining the product recommendation list by calculating the product recommendation row vector can effectively improve the accuracy of the recommendation.
  • the top 3 products ranked by descending order are product 2, product 3, and product 4, respectively.
  • the method adopts intelligent recommendation technology to realize the determination of the trust user cluster through the mutual comment relationship between users, and accurately recommend the target user according to the products recommended by the trust user.
  • An embodiment of the present application further provides a trust network-based push device, which is used to perform any embodiment of the foregoing trust network-based push method.
  • FIG. 5 is a schematic block diagram of a trust network-based push device provided by an embodiment of the present application.
  • the push device 100 based on the trusted network may be configured in the management server.
  • the trust network-based push device 100 includes a current trust value acquisition unit 110, a user trust matrix update unit 120, a trusted user cluster acquisition unit 130, a recommended row vector acquisition unit 140, and a push unit 150.
  • the current trust value acquisition unit 110 is used to determine whether there is a newly added mutual review relationship between users if the product review information at the current time is compared with the product review information at the previous time, according to the newly added mutual review relationship between the users 'S comment content gets the current trust value.
  • the current trust value acquisition unit 110 includes:
  • the word segmentation unit 111 is configured to obtain comment content corresponding to the newly added mutual comment relationship between users, and perform word segmentation on the comment content to obtain a word segmentation result;
  • the text vector acquiring unit 112 is used to acquire the word vector corresponding to each review keyword in the word segmentation result through the Word2Vec model, and the text vector corresponding to each review keyword in the word segmentation result corresponds to acquire the text vector;
  • the emotion recognition unit 113 is used to input the text vector as the input of the pre-trained Naive Bayes model to obtain the emotion recognition result corresponding to the comment content; wherein, if the emotion recognition result is a positive emotion result, the emotion recognition result takes the value 1. If the emotion recognition result is a negative emotion result, the emotion recognition result takes a value of -1, and if the emotion recognition result is a neutral emotion result, the emotion recognition result takes a value of 0;
  • the trust value calculation unit 114 is used to multiply the emotion recognition result and the attenuation coefficient to obtain the current trust value; where the attenuation coefficient is e- ⁇ (t-t0) , ⁇ is the preset adjustment parameter and the value range is (0 , 1), t-t0 is the comment time interval corresponding to the mutual comment relationship between users.
  • the word segmentation unit 111 includes:
  • the candidate word selection unit 1111 is used to extract candidate words from the comment content in order from left to right;
  • the initial left neighbor acquisition unit 1112 is used to query the probability value corresponding to each candidate word in the pre-stored dictionary, and record the left neighbor word of each candidate word;
  • the best left neighbor acquisition unit 1113 is used to calculate the cumulative probability of acquiring each candidate word, and acquire the respective cumulative probability of each left neighbor word corresponding to each candidate word, if multiple left neighbor words of each candidate word There are left neighbor words whose cumulative probability is the maximum value among the cumulative probabilities of multiple left neighbor words, and the left neighbor word with the largest cumulative probability is taken as the best left neighbor word corresponding to the candidate word;
  • the word segmentation result output unit 1114 is configured to output the best left-neighbor words corresponding to each candidate word in order from the end word of the review content as a starting point to obtain a word segmentation result.
  • the user trust matrix update unit 120 is configured to update the user trust matrix stored at the previous time according to the current trust value to obtain the user trust matrix stored at the current time; wherein, the user trust matrix stored at the previous time and the current time are stored In the user trust matrix of, each value represents the trust value of the user in the row corresponding to the value in the user in the column corresponding to the value.
  • the trusted user cluster obtaining unit 130 is used to obtain the target user corresponding to the row vector selected from the user trust matrix stored at the current moment, and according to the size of each trust value in the row vector corresponding to the target user, the obtained trust value is located at a preset Trust users before the ranking threshold form a trust user cluster.
  • the recommendation row vector obtaining unit 140 is configured to obtain a trust user's trust score value for each product according to the corresponding rating row vector of each trusted user in the trusted user cluster in the user-score matrix to form a product recommendation row vector.
  • the recommended row vector obtaining unit 140 includes:
  • the scoring matrix obtaining unit 141 is used to obtain a scoring matrix composed of corresponding scoring row vectors in the user-scoring matrix of each trusted user in the trusted user cluster;
  • the trusted user line vector obtaining unit 142 is configured to obtain a trusted user line vector composed of each trust value of the trusted user cluster in the line vector corresponding to the target user;
  • the matrix calculation unit 143 is configured to multiply the trusted user row vector and the scoring matrix to obtain a product recommendation row vector.
  • the scoring matrix obtaining unit 141 is further used to: obtain the scoring row vector corresponding to each trusted user in the user-scoring matrix of the trusted user cluster, according to the corresponding score of each trusted user in the user-scoring matrix
  • the row numbers in the row vector are arranged in order, to obtain a scoring matrix.
  • the corresponding score row vector of each trusted user in the user-score matrix is obtained in sequence. The above method can accurately obtain the score row vector corresponding to each trusted user.
  • the scoring matrix facilitates the subsequent calculation of the user's trust score for each product.
  • the pushing unit 150 is configured to obtain a commodity recommendation list from the commodity recommendation row vector, and push the commodity recommendation list to a receiving end corresponding to the target user.
  • the aforementioned push device based on the trusted network may be implemented in the form of a computer program, and the computer program may run on the computer device shown in FIG. 9.
  • FIG. 9 is a schematic block diagram of a computer device provided by an embodiment of the present application.
  • the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, where the memory may include a non-volatile storage medium 503 and an internal memory 504.
  • the non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032.
  • the computer program 5032 When executed, it may cause the processor 502 to execute a push method based on a trusted network.
  • the processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.
  • the internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503.
  • the processor 502 can execute the push method based on the trusted network.
  • the network interface 505 is used for network communication, such as the transmission of data information.
  • the processor 502 is used to run the computer program 5032 stored in the memory, so as to implement the push method based on the trusted network according to the embodiment of the present application.
  • the embodiment of the computer device shown in FIG. 9 does not constitute a limitation on the specific configuration of the computer device.
  • the computer device may include more or fewer components than shown in the figure. Or combine certain components, or arrange different components.
  • the computer device may only include a memory and a processor. In such an embodiment, the structures and functions of the memory and the processor are consistent with the embodiment shown in FIG. 9 and will not be repeated here.
  • the processor 502 may be a central processing unit (Central Processing Unit, CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), Application specific integrated circuit (Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may be any conventional processor.
  • a computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • the computer-readable storage medium stores a computer program, where when the computer program is executed by the processor, the push method based on the trusted network of the embodiment of the present application is implemented.
  • the storage medium may be an internal storage unit of the aforementioned device, such as a hard disk or a memory of the device.
  • the storage medium may also be an external storage device of the device, such as a plug-in hard disk equipped on the device, a smart memory card (Smart) Card (SMC), a secure digital (SD) card, or a flash memory card (Flash Card) etc.
  • the storage medium may also include both an internal storage unit of the device and an external storage device.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请公开了基于信任网络的推送方法、装置、计算机设备及存储介质。该方法解析商品评论信息判定用户之间存在相互评论关系,以获取当前时刻存储的用户信任矩阵,获取从当前信任矩阵选中的行向量所对应的目标用户,根据行向量中各信任值大小,获取信任值大小位于排名阈值之前的信任用户以组成信任用户簇;根据信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;由商品推荐行向量得到商品推荐列表,将商品推荐列表推送至目标用户对应的接收端。

Description

基于信任网络的推送方法、装置、计算机设备及存储介质
本申请要求于2018年10月12日提交中国专利局、申请号为201811191704.6、申请名称为“基于信任网络的推送方法、装置、计算机设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及信息推送技术领域,尤其涉及一种基于信任网络的推送方法、装置、计算机设备及存储介质。
背景技术
目前,在基于互联网的在线商城上进行网络购物已越来越频繁,这些在线商城对用户进行商品推荐时,通常使用的是基于协同过滤的推荐算法(协同过滤算法,原理是用户喜欢那些具有相似兴趣的用户喜欢过的商品,比如你的朋友喜欢电影哈利波特I,那么就会推荐给你,这是最简单的基于用户的协同过滤算法)。在使用基于协同过滤的推荐算法时,采用的数据集主要是基于用户-商品的评分矩阵,进而测算出用户之间的相似性,属于隐形的信任网络。但是采用评分矩阵测算出用户之间的相似性存在冷启动的问题,影响到推荐过程。
发明内容
本申请实施例提供了一种基于信任网络的推送方法、装置、计算机设备及存储介质,旨在解决现有技术中采用基于协同过滤的推荐算法需采用评分矩阵测算出用户之间的相似性,易导致推荐运算的冷启动,从而影响到推荐过程的问题。
第一方面,本申请实施例提供了一种基于信任网络的推送方法,其包括:
若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值;
根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存 储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值;
获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇;
根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;以及
由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
第二方面,本申请实施例提供了一种基于信任网络的推送装置,其包括:
当前信任值获取单元,用于若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值;
用户信任矩阵更新单元,用于根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值;
信任用户簇获取单元,用于获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇;
推荐行向量获取单元,用于根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;
推送单元,用于由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
第三方面,本申请实施例又提供了一种计算机设备,其包括存储器、处理器及存储在所述存储器上并可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第一方面所述的基于信任网络的推送方法。
第四方面,本申请实施例还提供了一种存储介质,其中所述计算机可读存储介质存储有计算机程序,所述计算机程序当被处理器执行时使所述处理器执 行上述第一方面所述的基于信任网络的推送方法。
附图说明
为了更清楚地说明本申请实施例技术方案,下面将对实施例描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的基于信任网络的推送方法的流程示意图;
图2为本申请实施例提供的基于信任网络的推送方法的子流程示意图;
图3为本申请实施例提供的基于信任网络的推送方法的另一子流程示意图;
图4为本申请实施例提供的基于信任网络的推送方法的另一子流程示意图;
图5为本申请实施例提供的基于信任网络的推送装置的示意性框图;
图6为本申请实施例提供的基于信任网络的推送装置的子单元示意性框图;
图7为本申请实施例提供的基于信任网络的推送装置的另一子单元示意性框图;
图8为本申请实施例提供的基于信任网络的推送装置的另一子单元示意性框图;
图9为本申请实施例提供的计算机设备的示意性框图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通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,都属于本申请保护的范围。
应当理解,当在本说明书和所附权利要求书中使用时,术语“包括”和“包含”指示所描述特征、整体、步骤、操作、元素和/或组件的存在,但并不排除一个或多个其它特征、整体、步骤、操作、元素、组件和/或其集合的存在或添加。
还应当理解,在此本申请说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样,除非上下文清楚地指明其它情况,否则单数形式的“一”、“一个”及“该” 意在包括复数形式。
还应当进一步理解,在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合,并且包括这些组合。
请参阅图1,图1是本申请实施例提供的基于信任网络的推送方法的流程示意图,该基于信任网络的推送方法应用于管理服务器中,该方法通过安装于管理服务器中的应用软件进行执行,管理服务器即是用于进行基于信任网络的推送的企业终端。
如图1所示,该方法包括步骤S110~S150。
S110、若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值。
在本实施例中,当管理服务器在爬取了预设URL地址列表中每一网页对应的商品评论信息后,即可根据用户与用户之间是否存在评论关系以判断用户之间的信任关系,并根据用户之间新增的相互评论关系对应的评论内容获取当前信任值。这样的用户关系是显性的,如在商品评论上,A用户肯定了B的用户的评论,显示了A用户信任B用户,或者在社区空间中,A用户肯定了B用户的内容。基于这种显式的信任网络,加上既有的历史数据,可以更好地解决冷启动的问题。
在一实施例中,如图2所示,步骤S110包括;
S111、获取用户之间新增的的相互评论关系对应的评论内容,将所述评论内容进行分词得到分词结果;
S112、通过Word2Vec模型获取分词结果中各评论关键词对应的词向量,由分词结果中各评论关键词对应的词向量对应获取文本向量;
S113、将文本向量作为预先训练得到的朴素贝叶斯模型的输入,得到与评论内容对应的情感识别结果;其中,若情感识别结果为正面情感结果时情感识别结果取值为1,若情感识别结果为负面情感结果时情感识别结果取值为-1,若情感识别结果为中性情感结果时情感识别结果取值为0;
S114、将情感识别结果与衰减系数相乘,得到当前信任值;其中衰减系数为e -λ(t-t0),λ为预设的调节参数且取值范围为(0,1),t-t0为用户之间的相互 评论关系对应的评论时间间隔。
在本实施例中,通过对商品评论信息进行解析,主要是判断用户之间是否存在相互评论关系,若两用户之间存在评论关系,对评论内容进行正面情感,中性情感和负面情感的判定。在对评论内容进行情感识别时,先将评论内容进行分词再转化为文本向量,将文本向量作为朴素贝叶斯模型的输入,得到情感识别结果。但为了考虑到用户与用户之间评论关系受到时间衰减的影响,则需要将情感识别结果乘以一个衰减系数(也可以理解为时间衰退因子)得到更为客观准确的当前信任值。其中,Word2Vec是从大量文本语料中以无监督的方式学习语义知识的一种模型。
若将用户i对用户j的当前信任值记为a ij,则a ij+=Sentiment ij*e -λ(t-t0),也即当前时刻的信任值是在上一时刻的信任值基础上进行自增的调整(例如i用户对j用户的评论每隔一段时间都发表一次评论,那么每次发表的评论都会对当前信任值产生影响),其中Sentiment ij表示i用户对j用户的评论而发表的评论内容进行情感识别得到的情感识别结果,若情感识别结果为正面情感结果时Sentiment ij=1,若情感识别结果为负面情感结果时Sentiment ij=-1,若情感识别结果为中性情感结果时Sentiment ij=0。
例如,若预设的调节参数λ=0.5,用户1在用户2的评论下方追加评论,且该评论为正面情感的评论,用户1的追加评论时间t与用户2的评论时间t0之差为1天,则用户1对用户2的当前信任值记为a 12=0+1*e -0.5*(1-0)=0.607;用户1在用户3的评论下方追加评论,且该评论为负面情感的评论,用户1的追加评论时间t与用户3的评论时间t0之差为1天,则用户1对用户3的当前信任值记为a 13=0-1*e -0.5*(1-0)=-0.607;用户1在用户4的评论下方追加评论,且该评论为中性情感的评论,用户1的追加评论时间t与用户4的评论时间t0之差为1天,则用户1对用户4的当前信任值记为a 15=0-0*e -0.5*(1-0)=0。
以用户之间的评论内容对应的情感识别结果,作为用户之间信任值的计算基础,是一种显式而非隐式的信任网络搭建方式,可以解决推荐过程中冷启动的问题。
在一实施例中,如图4所示,步骤S111包括:
S1111、按从左至右的顺序从评论内容中取出候选词;
S1112、在预先存储的词典中查询与每一候选词对应的概率值,并记录每一 候选词的左邻词;
S1113、计算获取每一候选词的累积概率,并获取每一候选词对应的多个左邻词各自的累积概率,若每一候选词的多个左邻词中存在累积概率为多个左邻词的累积概率中最大值的左邻词,将累积概率中最大值的左邻词作为与候选词对应的最佳左邻词;
S1114、以评论内容的终点词为起点,从右至左依次输出与每一候选词对应的最佳左邻词,得到分词结果。
在本实施例中,对评论内容进行分词时,是通过基于概率统计模型的分词方法进行分词。例如,令C=C1C2...Cm,C是待切分的汉字串,令W=W1W2...Wn,W是切分的结果,Wa,Wb,….Wk是C的所有可能的切分方案。那么,基于概率统计的切分模型就是能够找到目的词串W,使得W满足:P(W|C)=MAX(P(Wa|C),P(Wb|C)...P(Wk|C))的分词模型,上述分词模型得到的词串W即估计概率为最大之词串。
即对一个待分词的子串S,按照从左到右的顺序取出全部候选词w1、w2、…、wi、…、wn;在词典中查出每个候选词的概率值P(wi),并记录每个候选词的全部左邻词;计算每个候选词的累积概率,同时比较得到每个候选词的最佳左邻词;如果当前词wn是字串S的尾词,且累积概率P(wn)最大,则wn就是S的终点词;从wn开始,按照从右到左顺序,依次将每个词的最佳左邻词输出,即S的分词结果。
S120、根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值。
在本实施例中,用户信任矩阵表示用户之间的信任值,用户信任矩阵中的横轴和纵轴均为用户列表,用户信任矩阵A n*n中a ij指的是用户i对用户j的信任值,在初始时刻存储的用户信任矩阵中所有值均为0,之后每隔时间周期T重新判断用户i对用户j是否有新增的评论内容以确定是否调整用户i对用户j的信任值。上一时刻存储的用户信任矩阵是管理服务器在上一时刻(如2018年X月Y日Z1时Z2分)爬取了预设URL地址列表中每一网页对应的商品评论信息后,即可根据用户与用户之间是否存在评论关系以判断用户之间的信任关系,并根 据用户之间的相互评论关系对应的评论内容获取上一时刻的信任值。
若与上一时刻相隔时间周期T对应的是当前时刻(如2018年X月Y日Z3时Z4分)爬取了预设URL地址列表中每一网页对应的商品评论信息后,即可根据用户与用户之间是否存在新增的评论关系以判断用户之间的信任关系是否发生改变,并根据用户之间的相互评论关系对应的评论内容获取当前时刻的信任值。即用户i对用户j的当前信任值记为a ij,则a ij+=Sentiment ij*e -λ(t-t0),也即当前时刻的信任值是在上一时刻的信任值基础上进行自增的调整(例如i用户对j用户的评论每隔一段时间都发表一次评论,那么每次发表的评论都会对当前信任值产生影响)。
S130、获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇。
在本实施例中,例如当前时刻存储的用户信任矩阵如下:
Figure PCTCN2018124954-appb-000001
且在当前时刻存储的用户信任矩阵选中的行向量为第一行的行向量,第一行的行向量对应的目标用户是用户1,第一行的行向量中各值表示用户1与其他用户之间的信任值。若预设排名阈值为3,则信任值大小位于第3名之前(即排前2名的信任值)的为3和1,分别对应用户2和用户4,则信任用户簇中包括用户2和用户4。
S140、根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量。
在本实施例中,用户-评分矩阵表示用户对项目(项目可以理解为具体的商品)的评分,用户-评分矩阵的横轴为项目,纵轴为用户,当中的值是用户i对项目j的评分。例如用户-评分矩阵S为4×6的矩阵,如:
Figure PCTCN2018124954-appb-000002
用户-评分矩阵S中第一行的行向量表示用户1分别针对项目1-项目5的评分,第二行的行向量表示用户2分别针对项目1-项目5的评分,第三行的行向量表示用户3分别针对项目1-项目5的评分。
例如,确定了信任用户簇中包括用户2和用户4,则在用户-评分矩阵中对应获取用户2和用户4的评分行向量,也即第二行的行向量和第四行的行向量。在综合考虑目标用户与信任用户簇中各用户之间的信任值,以及信任用户对各商品的评分值,即可运算得到商品推荐行向量。
在一实施例中,如图3所示,步骤S140包括:
S141、获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵;
S142、获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量;
S143、将所述信任用户行向量与所述评分矩阵相乘,以得到商品推荐行向量。
在本实施中,例如选定的确定了信任用户簇中包括用户2和用户4,则在用户-评分矩阵中对应获取用户2和用户4的评分行向量,也即第二行的行向量和第四行的行向量,则用户2和用户4针对各项目(也即各商品)之间评分组成的评分矩阵如下:
Figure PCTCN2018124954-appb-000003
在一实施例中,步骤S141中具体包括:获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,按照各信任用户在用户-评分矩阵中对应的评分行向量中行序号的先后顺序进行排列,得到评分矩阵。按照上述各用户在用户-评分矩阵中各行出现的先后顺序,来依序获取各信任用户在用户-评分矩阵中对应的评分行向量,上述方式能精确获取各信任用户对应的评分行向量组成的评分矩阵,便于后续计算用户对各商品的信任评分值。
由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量为[3 1]。
在一实施例中,步骤S142中具体包括:获取信任值大小位于预设排名阈值之前的信任用户,将信任值大小位于所述排名阈值之前的信任用户按照在目标 用户对应的行向量中列序号的先后顺序进行排列,得到信任用户行向量。按照上述各用户在目标用户对应的行向量中各列出现的先后顺序,来依序获取在目标用户对应的行向量中各信任值大小,以组成的信任用户行向量,上述方式能精确获取信任用户行向量,便于后续计算用户对各商品的信任评分值。
将信任用户行向量与评分矩阵相乘,即
Figure PCTCN2018124954-appb-000004
即商品推荐行向量为[2 13 8 13 7],这一商品推荐行向量中评分排名较高的评分所对应商品即可作为信任群体对目标用户所推荐的商品。通过将信任用户行向量与评分矩阵相乘的方式获取商品推荐行向量,能有效参考信任用户所对各商品的综合评分,以作为对目标用户进行商品推荐的参考指标。
S150、由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
在本实施例中,在获取了商品推荐行向量后,即可获知信任用户簇中各信任用户对各商品的综合评分,此时可选择商品推荐行向量对应的各评分中评分排名位于预设排名值(如预设的排名值为4)之前的评分,再获取各评分中评分排名位于预设排名值之前的评分对应的商品信息,将商品信息组成商品推荐列表并推送至目标用户对应的接收端。采用计算商品推荐行向量的方式来获取商品推荐列表,能有效提高推荐的准确度。
例如,上述信任群体(用户2和用户4)向用户1推荐的5件商品中,评分按降序排序后排名位于前3名的商品分别是商品2、商品3和商品4,此时将上述3件商品作为商品推荐行向量。
该方法采用智能推荐技术实现了通过用户之间存在相互评论关系确定信任用户簇,根据信任用户所推荐的商品以对目标用户进行精准推荐。
本申请实施例还提供一种基于信任网络的推送装置,该基于信任网络的推送装置用于执行前述基于信任网络的推送方法的任一实施例。具体地,请参阅图5,图5是本申请实施例提供的基于信任网络的推送装置的示意性框图。该基于信任网络的推送装置100可以配置于管理服务器中。
如图5所示,基于信任网络的推送装置100包括当前信任值获取单元110、用户信任矩阵更新单元120、信任用户簇获取单元130、推荐行向量获取单元140、 和推送单元150。
当前信任值获取单元110,用于若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值。
在一实施例中,如图6所示,当前信任值获取单元110包括:
分词单元111,用于获取用户之间新增的的相互评论关系对应的评论内容,将所述评论内容进行分词得到分词结果;
文本向量获取单元112,用于通过Word2Vec模型获取分词结果中各评论关键词对应的词向量,由分词结果中各评论关键词对应的词向量对应获取文本向量;
情感识别单元113,用于将文本向量作为预先训练得到的朴素贝叶斯模型的输入,得到与评论内容对应的情感识别结果;其中,若情感识别结果为正面情感结果时情感识别结果取值为1,若情感识别结果为负面情感结果时情感识别结果取值为-1,若情感识别结果为中性情感结果时情感识别结果取值为0;
信任值计算单元114,用于将情感识别结果与衰减系数相乘,得到当前信任值;其中衰减系数为e -λ(t-t0),λ为预设的调节参数且取值范围为(0,1),t-t0为用户之间的相互评论关系对应的评论时间间隔。
在一实施例中,如图8所示,分词单元111包括:
候选词选取单元1111,用于按从左至右的顺序从评论内容中取出候选词;
初始左邻词获取单元1112,用于在预先存储的词典中查询与每一候选词对应的概率值,并记录每一候选词的左邻词;
最佳左邻词获取单元1113,用于计算获取每一候选词的累积概率,并获取每一候选词对应的多个左邻词各自的累积概率,若每一候选词的多个左邻词中存在累积概率为多个左邻词的累积概率中最大值的左邻词,将累积概率中最大值的左邻词作为与候选词对应的最佳左邻词;
分词结果输出单元1114,用于以评论内容的终点词为起点,从右至左依次输出与每一候选词对应的最佳左邻词,得到分词结果。
用户信任矩阵更新单元120,用于根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在 行对应的用户对取值所在列对应的用户的信任值。
信任用户簇获取单元130,用于获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇。
推荐行向量获取单元140,用于根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量。
在一实施例中,如图7所示,推荐行向量获取单元140包括:
评分矩阵获取单元141,用于获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵;
信任用户行向量获取单元142,用于获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量;
矩阵计算单元143,用于将所述信任用户行向量与所述评分矩阵相乘,以得到商品推荐行向量。
在一实施例中,评分矩阵获取单元141还用于:获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,按照各信任用户在用户-评分矩阵中对应的评分行向量中行序号的先后顺序进行排列,得到评分矩阵。按照上述各用户在用户-评分矩阵中各行出现的先后顺序,来依序获取各信任用户在用户-评分矩阵中对应的评分行向量,上述方式能精确获取各信任用户对应的评分行向量组成的评分矩阵,便于后续计算用户对各商品的信任评分值。
推送单元150,用于由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
上述基于信任网络的推送装置可以实现为计算机程序的形式,该计算机程序可以在如图9所示的计算机设备上运行。请参阅图9,图9是本申请实施例提供的计算机设备的示意性框图。
参阅图9,该计算机设备500包括通过系统总线501连接的处理器502、存储器和网络接口505,其中,存储器可以包括非易失性存储介质503和内存储器504。该非易失性存储介质503可存储操作系统5031和计算机程序5032。该计算机程序5032被执行时,可使得处理器502执行基于信任网络的推送方法。该处理器502用于提供计算和控制能力,支撑整个计算机设备500的运行。该内 存储器504为非易失性存储介质503中的计算机程序5032的运行提供环境,该计算机程序5032被处理器502执行时,可使得处理器502执行基于信任网络的推送方法。该网络接口505用于进行网络通信,如提供数据信息的传输等。本领域技术人员可以理解,图9中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备500的限定,具体的计算机设备500可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
其中,所述处理器502用于运行存储在存储器中的计算机程序5032,以实现本申请实施例的基于信任网络的推送方法。
本领域技术人员可以理解,图9中示出的计算机设备的实施例并不构成对计算机设备具体构成的限定,在其他实施例中,计算机设备可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。例如,在一些实施例中,计算机设备可以仅包括存储器及处理器,在这样的实施例中,存储器及处理器的结构及功能与图9所示实施例一致,在此不再赘述。
应当理解,在本申请实施例中,处理器502可以是中央处理单元(Central Processing Unit,CPU),该处理器502还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中,通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
在本申请的另一实施例中提供计算机可读存储介质。该计算机可读存储介质可以为非易失性的计算机可读存储介质。该计算机可读存储介质存储有计算机程序,其中计算机程序被处理器执行时实现本申请实施例的基于信任网络的推送方法。
所述存储介质可以是前述设备的内部存储单元,例如设备的硬盘或内存。所述存储介质也可以是所述设备的外部存储设备,例如所述设备上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,所述存储介质还可以既包括所述设备的内部存储单元也包括外部存储设备。
所属领域的技术人员可以清楚地了解到,为了描述的方便和简洁,上述描 述的设备、装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到各种等效的修改或替换,这些修改或替换都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以权利要求的保护范围为准。

Claims (20)

  1. 一种基于信任网络的推送方法,包括:
    若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值;
    根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值;
    获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇;
    根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;以及
    由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
  2. 根据权利要求1所述的基于信任网络的推送方法,其中,所述根据用户之间新增的相互评论关系对应的评论内容获取当前信任值,包括:
    获取用户之间新增的的相互评论关系对应的评论内容,将所述评论内容进行分词得到分词结果;
    通过Word2Vec模型获取分词结果中各评论关键词对应的词向量,由分词结果中各评论关键词对应的词向量对应获取文本向量;
    将文本向量作为预先训练得到的朴素贝叶斯模型的输入,得到与评论内容对应的情感识别结果;其中,若情感识别结果为正面情感结果时情感识别结果取值为1,若情感识别结果为负面情感结果时情感识别结果取值为-1,若情感识别结果为中性情感结果时情感识别结果取值为0;
    将情感识别结果与衰减系数相乘,得到当前信任值;其中衰减系数为e -λ(t-t0),λ为预设的调节参数且取值范围为(0,1),t-t0为用户之间的相互评论关系对应的评论时间间隔。
  3. 根据权利要求1所述的基于信任网络的推送方法,其中,所述根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量,包括:
    获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵;
    获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量;
    将所述信任用户行向量与所述评分矩阵相乘,以得到商品推荐行向量。
  4. 根据权利要求2所述的基于信任网络的推送方法,其中,其特征在于,所述将所述评论内容进行分词得到分词结果,包括:
    按从左至右的顺序从评论内容中取出候选词;
    在预先存储的词典中查询与每一候选词对应的概率值,并记录每一候选词的左邻词;
    计算获取每一候选词的累积概率,并获取每一候选词对应的多个左邻词各自的累积概率,若每一候选词的多个左邻词中存在累积概率为多个左邻词的累积概率中最大值的左邻词,将累积概率中最大值的左邻词作为与候选词对应的最佳左邻词;
    以评论内容的终点词为起点,从右至左依次输出与每一候选词对应的最佳左邻词,得到分词结果。
  5. 根据权利要求3所述的基于信任网络的推送方法,其中,所述获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵,包括:
    获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,按照各信任用户在用户-评分矩阵中对应的评分行向量中行序号的先后顺序进行排列,得到评分矩阵。
  6. 根据权利要求3所述的基于信任网络的推送方法,其中,所述获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量,包括:
    获取信任值大小位于预设排名阈值之前的信任用户,将信任值大小位于所述排名阈值之前的信任用户按照在目标用户对应的行向量中列序号的先后顺序 进行排列,得到信任用户行向量。
  7. 一种基于信任网络的推送装置,包括:
    当前信任值获取单元,用于若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值;
    用户信任矩阵更新单元,用于根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值;
    信任用户簇获取单元,用于获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇;
    推荐行向量获取单元,用于根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;
    推送单元,用于由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
  8. 根据权利要求7所述的基于信任网络的推送装置,其中,所述当前信任值获取单元,包括:
    分词单元,用于获取用户之间新增的的相互评论关系对应的评论内容,将所述评论内容进行分词得到分词结果;
    文本向量获取单元,用于通过Word2Vec模型获取分词结果中各评论关键词对应的词向量,由分词结果中各评论关键词对应的词向量对应获取文本向量;
    情感识别单元,用于将文本向量作为预先训练得到的朴素贝叶斯模型的输入,得到与评论内容对应的情感识别结果;其中,若情感识别结果为正面情感结果时情感识别结果取值为1,若情感识别结果为负面情感结果时情感识别结果取值为-1,若情感识别结果为中性情感结果时情感识别结果取值为0;
    信任值计算单元,用于将情感识别结果与衰减系数相乘,得到当前信任值;其中衰减系数为e -λ(t-t0),λ为预设的调节参数且取值范围为(0,1),t-t0为用户之间的相互评论关系对应的评论时间间隔。
  9. 根据权利要求8所述的基于信任网络的推送装置,其中,所述分词单元,包括:
    候选词选取单元,用于按从左至右的顺序从评论内容中取出候选词;
    初始左邻词获取单元,用于在预先存储的词典中查询与每一候选词对应的概率值,并记录每一候选词的左邻词;
    最佳左邻词获取单元,用于计算获取每一候选词的累积概率,并获取每一候选词对应的多个左邻词各自的累积概率,若每一候选词的多个左邻词中存在累积概率为多个左邻词的累积概率中最大值的左邻词,将累积概率中最大值的左邻词作为与候选词对应的最佳左邻词;
    分词结果输出单元,用于以评论内容的终点词为起点,从右至左依次输出与每一候选词对应的最佳左邻词,得到分词结果。
  10. 根据权利要求7所述的基于信任网络的推送装置,其中,所述推荐行向量获取单元,包括:
    评分矩阵获取单元,用于获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵;
    信任用户行向量获取单元,用于获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量;
    矩阵计算单元,用于将所述信任用户行向量与所述评分矩阵相乘,以得到商品推荐行向量。
  11. 一种计算机设备,包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序,其中,所述处理器执行所述计算机程序时实现以下步骤:
    若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值;
    根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值;
    获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户, 根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇;
    根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;以及
    由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
  12. 根据权利要求11所述的计算机设备,其中,所述根据用户之间新增的相互评论关系对应的评论内容获取当前信任值,包括:
    获取用户之间新增的的相互评论关系对应的评论内容,将所述评论内容进行分词得到分词结果;
    通过Word2Vec模型获取分词结果中各评论关键词对应的词向量,由分词结果中各评论关键词对应的词向量对应获取文本向量;
    将文本向量作为预先训练得到的朴素贝叶斯模型的输入,得到与评论内容对应的情感识别结果;其中,若情感识别结果为正面情感结果时情感识别结果取值为1,若情感识别结果为负面情感结果时情感识别结果取值为-1,若情感识别结果为中性情感结果时情感识别结果取值为0;
    将情感识别结果与衰减系数相乘,得到当前信任值;其中衰减系数为e -λ(t-t0),λ为预设的调节参数且取值范围为(0,1),t-t0为用户之间的相互评论关系对应的评论时间间隔。
  13. 根据权利要求11所述的计算机设备,其中,所述根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量,包括:
    获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵;
    获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量;
    将所述信任用户行向量与所述评分矩阵相乘,以得到商品推荐行向量。
  14. 根据权利要求12所述的计算机设备,其中,所述将所述评论内容进行分词得到分词结果,包括:
    按从左至右的顺序从评论内容中取出候选词;
    在预先存储的词典中查询与每一候选词对应的概率值,并记录每一候选词的左邻词;
    计算获取每一候选词的累积概率,并获取每一候选词对应的多个左邻词各自的累积概率,若每一候选词的多个左邻词中存在累积概率为多个左邻词的累积概率中最大值的左邻词,将累积概率中最大值的左邻词作为与候选词对应的最佳左邻词;
    以评论内容的终点词为起点,从右至左依次输出与每一候选词对应的最佳左邻词,得到分词结果。
  15. 根据权利要求13所述的计算机设备,其中,所述获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵,包括:
    获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,按照各信任用户在用户-评分矩阵中对应的评分行向量中行序号的先后顺序进行排列,得到评分矩阵。
  16. 根据权利要求13所述的计算机设备,其中,所述获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量,包括:
    获取信任值大小位于预设排名阈值之前的信任用户,将信任值大小位于所述排名阈值之前的信任用户按照在目标用户对应的行向量中列序号的先后顺序进行排列,得到信任用户行向量。
  17. 一种计算机可读存储介质,其中,所述计算机可读存储介质存储有计算机程序,所述计算机程序当被处理器执行时使所述处理器执行以下操作:
    若当前时刻的商品评论信息与上一时刻的商品评论信息比对以判定用户之间存在新增的相互评论关系,根据用户之间新增的相互评论关系对应的评论内容获取当前信任值;
    根据所述当前信任值对上一时刻存储的用户信任矩阵进行更新,得到当前时刻存储的用户信任矩阵;其中,上一时刻存储的用户信任矩阵和当前时刻存储的用户信任矩阵中,每一取值表示取值所在行对应的用户对取值所在列对应的用户的信任值;
    获取从所述当前时刻存储的用户信任矩阵选中的行向量所对应的目标用户,根据目标用户对应的行向量中各信任值大小,获取信任值大小位于预设排名阈值之前的信任用户以组成信任用户簇;
    根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量;以及
    由商品推荐行向量得到商品推荐列表,将所述商品推荐列表推送至目标用户对应的接收端。
  18. 根据权利要求17所述的存储介质,其中,所述根据用户之间新增的相互评论关系对应的评论内容获取当前信任值,包括:
    获取用户之间新增的的相互评论关系对应的评论内容,将所述评论内容进行分词得到分词结果;
    通过Word2Vec模型获取分词结果中各评论关键词对应的词向量,由分词结果中各评论关键词对应的词向量对应获取文本向量;
    将文本向量作为预先训练得到的朴素贝叶斯模型的输入,得到与评论内容对应的情感识别结果;其中,若情感识别结果为正面情感结果时情感识别结果取值为1,若情感识别结果为负面情感结果时情感识别结果取值为-1,若情感识别结果为中性情感结果时情感识别结果取值为0;
    将情感识别结果与衰减系数相乘,得到当前信任值;其中衰减系数为e -λ(t-t0),λ为预设的调节参数且取值范围为(0,1),t-t0为用户之间的相互评论关系对应的评论时间间隔。
  19. 根据权利要求17所述的存储介质,其中,所述根据所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量,获取信任用户对各商品的信任评分值以组成商品推荐行向量,包括:
    获取所述信任用户簇中各信任用户在用户-评分矩阵中对应的评分行向量组成的评分矩阵;
    获取由所述信任用户簇在目标用户对应的行向量中各信任值大小组成的信任用户行向量;
    将所述信任用户行向量与所述评分矩阵相乘,以得到商品推荐行向量。
  20. 根据权利要求18所述的存储介质,其中,所述将所述评论内容进行分词得到分词结果,包括:
    按从左至右的顺序从评论内容中取出候选词;
    在预先存储的词典中查询与每一候选词对应的概率值,并记录每一候选词的左邻词;
    计算获取每一候选词的累积概率,并获取每一候选词对应的多个左邻词各自的累积概率,若每一候选词的多个左邻词中存在累积概率为多个左邻词的累积概率中最大值的左邻词,将累积概率中最大值的左邻词作为与候选词对应的最佳左邻词;
    以评论内容的终点词为起点,从右至左依次输出与每一候选词对应的最佳左邻词,得到分词结果。
PCT/CN2018/124954 2018-10-12 2018-12-28 基于信任网络的推送方法、装置、计算机设备及存储介质 WO2020073526A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811191704.6 2018-10-12
CN201811191704.6A CN109360058A (zh) 2018-10-12 2018-10-12 基于信任网络的推送方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020073526A1 true WO2020073526A1 (zh) 2020-04-16

Family

ID=65349036

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/124954 WO2020073526A1 (zh) 2018-10-12 2018-12-28 基于信任网络的推送方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN109360058A (zh)
WO (1) WO2020073526A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111597220B (zh) * 2019-02-21 2024-03-05 北京沃东天骏信息技术有限公司 数据挖掘方法和装置
CN111460819B (zh) * 2020-03-31 2023-06-20 湖南大学 基于细粒度情感分析的个性化评论文本推荐系统及推荐方法
CN113495942B (zh) * 2020-04-01 2022-07-05 百度在线网络技术(北京)有限公司 推送信息的方法和装置
CN113837846B (zh) * 2021-10-27 2023-09-22 武汉卓尔数字传媒科技有限公司 商品推荐方法、装置、计算机设备和存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559623A (zh) * 2013-09-24 2014-02-05 浙江大学 一种基于联合非负矩阵分解的个性化产品推荐方法
CN106682114A (zh) * 2016-12-07 2017-05-17 广东工业大学 一种融合用户信任关系和评论信息的个性化推荐方法
CN107273438A (zh) * 2017-05-24 2017-10-20 深圳大学 一种推荐方法、装置、设备及存储介质
CN107967641A (zh) * 2017-10-18 2018-04-27 美的智慧家居科技有限公司 商品推荐方法、装置及计算机可读存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107025606B (zh) * 2017-03-29 2021-04-16 西安电子科技大学 一种社交网络中结合评分数据和信任关系的项目推荐方法
CN107506480B (zh) * 2017-09-13 2020-05-05 浙江工业大学 一种基于评论挖掘与密度聚类的双层图结构推荐方法
CN108228867A (zh) * 2018-01-15 2018-06-29 武汉大学 一种基于观点增强的主题协同过滤推荐方法
CN108573411B (zh) * 2018-04-17 2021-09-21 重庆理工大学 基于用户评论的深度情感分析和多源推荐视图融合的混合推荐方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559623A (zh) * 2013-09-24 2014-02-05 浙江大学 一种基于联合非负矩阵分解的个性化产品推荐方法
CN106682114A (zh) * 2016-12-07 2017-05-17 广东工业大学 一种融合用户信任关系和评论信息的个性化推荐方法
CN107273438A (zh) * 2017-05-24 2017-10-20 深圳大学 一种推荐方法、装置、设备及存储介质
CN107967641A (zh) * 2017-10-18 2018-04-27 美的智慧家居科技有限公司 商品推荐方法、装置及计算机可读存储介质

Also Published As

Publication number Publication date
CN109360058A (zh) 2019-02-19

Similar Documents

Publication Publication Date Title
WO2020073534A1 (zh) 基于重聚类的推送方法、装置、计算机设备及存储介质
US10853360B2 (en) Searchable index
US20210209182A1 (en) Systems and methods for improved web searching
WO2020073526A1 (zh) 基于信任网络的推送方法、装置、计算机设备及存储介质
RU2694001C2 (ru) Способ и система создания параметра качества прогноза для прогностической модели, выполняемой в алгоритме машинного обучения
WO2020119063A1 (zh) 专家知识推荐方法、装置、计算机设备及存储介质
US11580168B2 (en) Method and system for providing context based query suggestions
WO2020037930A1 (zh) 信息推荐方法、装置、计算机设备及存储介质
TWI582619B (zh) Method and apparatus for providing referral words
WO2018157625A1 (zh) 基于强化学习的排序学习方法及服务器
US10437894B2 (en) Method and system for app search engine leveraging user reviews
CN109325182B (zh) 基于会话的信息推送方法、装置、计算机设备及存储介质
WO2013019688A2 (en) Social network powered query suggestions
CN109241243B (zh) 候选文档排序方法及装置
CN112100396B (zh) 一种数据处理方法和装置
CN111104511A (zh) 一种提取热点话题的方法、装置及存储介质
CN109933708A (zh) 信息检索方法、装置、存储介质及计算机设备
WO2013107031A1 (zh) 基于评论信息确定视频质量参数的方法、装置和系统
CN112632261A (zh) 智能问答方法、装置、设备及存储介质
CN113254795B (zh) 一种推荐模型的训练方法和装置
CN115374362A (zh) 多路召回模型训练方法、多路召回方法、装置及电子设备
CN106095941B (zh) 基于大数据知识库的解决方案推荐方法和系统
WO2021000400A1 (zh) 导诊相似问题对生成方法、系统及计算机设备
US10318594B2 (en) System and method for enabling related searches for live events in data streams
CN110162535B (zh) 用于执行个性化的搜索方法、装置、设备以及存储介质

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: 18936816

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18936816

Country of ref document: EP

Kind code of ref document: A1