CN111461841A - Article recommendation method, device, server and storage medium - Google Patents
Article recommendation method, device, server and storage medium Download PDFInfo
- Publication number
- CN111461841A CN111461841A CN202010264243.1A CN202010264243A CN111461841A CN 111461841 A CN111461841 A CN 111461841A CN 202010264243 A CN202010264243 A CN 202010264243A CN 111461841 A CN111461841 A CN 111461841A
- Authority
- CN
- China
- Prior art keywords
- user
- sequence
- behavior
- item
- recommendation
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Images
Classifications
-
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- 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
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0203—Market surveys; Market polls
Abstract
The application provides an article recommendation method, an article recommendation device, a server and a storage medium, and belongs to the technical field of internet. The method comprises the following steps: performing relevance extraction on at least two behavior sequence characteristics of a user to obtain behavior sequence relevant characteristics, wherein the behavior sequence characteristics are used for expressing the characteristics of behaviors executed by the user according to a time sequence; fusing user characteristics of a user, article characteristics of an article to be recommended and related characteristics of the behavior sequence to obtain fused characteristics, outputting recommendation probability based on the fused characteristics, wherein the recommendation probability is used for indicating the probability of recommending the article to be recommended to the user; and recommending the item to be recommended to the user in response to the recommendation probability being greater than the target recommendation probability. According to the technical scheme, the recommendation probability is determined based on the fusion characteristics, so that the recommendation probability is related to the characteristics of the user and the to-be-recommended articles and is also related to the characteristics of the behaviors executed by the user according to the time sequence, and the recommendation precision is high.
Description
Technical Field
The present application relates to the field of internet technologies, and in particular, to a method, an apparatus, a server, and a storage medium for recommending an item.
Background
With the rise of the online consumption mode and the rapid development of the e-commerce platform, the competitive strength of each e-commerce platform can be reflected only by more accurately knowing the personalized requirements of the user and providing personalized recommendation service for the user. Therefore, how to provide the user with the articles required or possibly preferred by the user based on the behavior habits and characteristics of the user has important significance for preempting the market opportunity of each e-commerce platform.
In the related art, a recommendation method based on artificial statistics may be adopted to implement a personalized recommendation function. The recommendation method based on the artificial statistics is to determine the degree of correlation between the current user and other users according to the basic information of the current user, so that the articles which are associated with the current user to a higher degree and are favored by other users are recommended to the current user.
In the technical scheme, the recommendation mode based on the artificial statistics is rough, and is only suitable for simple recommendation, the recommendation precision is low, and the recommendation effect is poor.
Disclosure of Invention
The embodiment of the application provides an article recommendation method, an article recommendation device, a server and a storage medium, wherein recommendation probability is determined based on fused fusion characteristics, so that the recommendation probability is related to characteristics of a user and an article to be recommended and behavior characteristics of other articles purchased by the user, and recommendation precision is high. The technical scheme is as follows:
in one aspect, an item recommendation method is provided, the method including:
performing relevance extraction on at least two behavior sequence characteristics of a user to obtain behavior sequence relevant characteristics, wherein the behavior sequence characteristics are used for representing characteristics of behaviors executed by the user according to a time sequence;
fusing the user characteristics of the user, the article characteristics of the articles to be recommended and the behavior sequence related characteristics to obtain fused characteristics, and determining recommendation probability based on the fused characteristics, wherein the recommendation probability is used for indicating the probability of recommending the articles to be recommended to the user;
and recommending the item to be recommended to the user in response to the recommendation probability being greater than a target recommendation probability.
In another aspect, an article recommendation apparatus is provided, the apparatus comprising:
the characteristic extraction module is used for carrying out correlation extraction on at least two behavior sequence characteristics of a user to obtain behavior sequence correlation characteristics, and the behavior sequence characteristics are used for expressing the characteristics of behaviors executed by the user according to a time sequence;
the characteristic fusion module is used for fusing the user characteristics of the user, the article characteristics of the articles to be recommended and the behavior sequence related characteristics to obtain fusion characteristics, and determining recommendation probability based on the fusion characteristics, wherein the recommendation probability is used for indicating the probability of recommending the articles to be recommended to the user;
and the article recommending module is used for recommending the article to be recommended to the user in response to the recommending probability being greater than the target recommending probability.
In an optional implementation manner, the apparatus further includes a model training apparatus, configured to obtain a preferred item set of a sample user, where the preferred item set is used as a training label of a model to be trained; obtaining at least two sample behavior sequence characteristics based on at least two sample user behavior sequences of the sample user, wherein the sample behavior sequence characteristics are used for representing behavior characteristics of the sample user when the sample user purchases the items in the preference item set; inputting the sample user characteristics of the sample object, the sample object characteristics of at least two preference objects in the preference object set and the at least two sample behavior sequence characteristics into a model to be trained, and adjusting model parameters according to an output result; and when the training completion condition is met, taking the trained model as the item recommendation model.
In another aspect, a server is provided, where the server includes a processor and a memory, where the memory is used to store at least one program code, and the at least one program code is loaded and executed by the processor to implement the operations performed in the item recommendation method in the embodiments of the present application.
In another aspect, a storage medium is provided, where at least one program code is stored, where the at least one program code is used to execute the item recommendation method in the embodiment of the present application.
The technical scheme provided by the embodiment of the application has the following beneficial effects:
in the embodiment of the application, the behavior sequence related features are obtained by extracting the related information in at least two behavior sequence features, the behavior sequence related features of a user can be represented, the behavior sequence related features are fused with the user features and the article features, and the recommendation probability is determined based on the fused feature, so that the recommendation probability is related to the features of the user and the article to be recommended and is also related to the features of behaviors executed by the user according to the time sequence, and the recommendation precision is high.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a block diagram of an item recommendation system 100 provided in accordance with an embodiment of the present application;
FIG. 2 is a flow chart of an item recommendation method provided according to an embodiment of the application;
FIG. 3 is a flow chart of another item recommendation method provided in accordance with an embodiment of the present application;
FIG. 4 is a schematic structural diagram of a Transformer provided in an embodiment of the present application;
FIG. 5 is a training flow diagram for training an item recommendation model according to an embodiment of the present application;
FIG. 6 is a schematic structural diagram of an item recommendation model provided in an embodiment of the present application;
FIG. 7 is a flow chart of another training process for training an item recommendation model according to an embodiment of the present application;
FIG. 8 is a block diagram of an item recommendation device provided in accordance with an embodiment of the present application;
fig. 9 is a schematic structural diagram of a server provided according to an embodiment of the present application.
Detailed Description
To make the objects, technical solutions and advantages of the present application more clear, embodiments of the present application will be described in further detail below with reference to the accompanying drawings.
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.
The embodiment of the application provides an article recommendation method which can be applied to a scene of recommending articles for a user. For example, a shopping website recommends a commodity for a user, a service website recommends an value-added service for the user, a video website recommends a video for the user, and the like. The article can be an entity commodity, a multimedia resource, a service in real life, and even a virtual resource such as stocks, bonds, digital currency and the like.
The main steps of the item recommendation method provided by the embodiment of the present application are briefly described below. Firstly, the server may perform correlation extraction on at least two behavior sequence features of the user to obtain a behavior sequence correlation feature, where the behavior sequence feature is used to represent a feature of behaviors performed by the user in a time sequence. Then, the server can fuse the user characteristics of the user, the article characteristics of the article to be recommended and the behavior sequence related characteristics to obtain fusion characteristics, and determine recommendation probability based on the fusion characteristics, wherein the recommendation probability is used for indicating the probability of recommending the article to be recommended to the user. Finally, the server responds to the fact that the recommendation probability is larger than the target recommendation probability, and the article to be recommended can be recommended to the user. In the item recommendation method, the recommendation probability is related to the characteristics of the user and the item to be recommended and also related to the characteristics of the behaviors executed by the user according to the time sequence, so that the recommendation precision is high and the recommendation effect is good.
Fig. 1 is a block diagram of an item recommendation system 100 according to an embodiment of the present application. The item recommendation system 100 includes: a terminal 110 and an item recommendation platform 120.
The terminal 110 is connected to the item recommendation platform 120 through a wireless network or a wired network. The terminal 110 may be at least one of a smartphone, a game console, a desktop computer, a tablet computer, an e-book reader, an MP3 player, an MP4 player, and a laptop portable computer. The terminal 110 is installed and operated with an application program supporting browsing of an article. The application may be a shopping-like application, a video-like application, or a social-like application, among others. Illustratively, the terminal 110 is a terminal used by a user, and an application running in the terminal 110 has a user account logged therein.
The item recommendation platform 120 includes at least one of a server, a plurality of servers, a cloud computing platform, and a virtualization center. The item recommendation platform 120 is used to provide recommendation services for applications that support browsing items. Optionally, the item recommendation platform 120 undertakes primary recommendation work, and the terminal 110 undertakes secondary recommendation work; or, the item recommendation platform 120 undertakes the secondary recommendation work, and the terminal 110 undertakes the primary recommendation work; alternatively, the item recommendation platform 120 or the terminal 110 may undertake the recommendation separately.
Optionally, the item recommendation platform 120 comprises: the system comprises an access server, an item recommendation server and a database. The access server is used for providing the terminal 110 with access service. The item recommendation server is used for providing item recommendation service. The article recommendation server can be one or more. When the item recommendation server is multiple, there are at least two item recommendation servers for providing different services, and/or there are at least two item recommendation servers for providing the same service, for example, providing the same service in a load balancing manner, which is not limited in the embodiment of the present application. The item recommendation server can be provided with a recommendation model. In the embodiment of the application, the recommendation model is constructed based on a combination of sequence pattern mining and a Transformer structure.
The terminal 110 may be generally referred to as one of a plurality of terminals, and the embodiment is only illustrated by the terminal 110.
Those skilled in the art will appreciate that the number of terminals described above may be greater or fewer. For example, the number of the terminals may be only one, or several tens or hundreds, or more, and in this case, the item recommendation system further includes other terminals. The number of terminals and the type of the device are not limited in the embodiments of the present application.
Fig. 2 is a flowchart of an item recommendation method according to an embodiment of the present application, and as shown in fig. 2, the method is described in the embodiment of the present application by taking an application to a server as an example. The item recommendation method comprises the following steps:
201. the server performs relevance extraction on at least two behavior sequence characteristics of the user to obtain behavior sequence relevant characteristics, wherein the behavior sequence characteristics are used for expressing the characteristics of behaviors executed by the user according to a time sequence.
In this embodiment of the application, the server may determine the behavior sequence characteristics of the user through a user behavior sequence of the user, where the user behavior sequence is used to indicate user behaviors based on a time sequence, and may be a record of behaviors performed by the user in a time sequence within a period of time, and the behaviors performed by the user may be consumption behaviors, browsing behaviors, collection behaviors, and the like. Accordingly, the server may obtain characteristics of at least two behaviors that the user continuously performs within a period of time, and obtain a behavior sequence characteristic of the user. The server can extract and embed the correlation of the behavior sequence characteristics of the user based on an attention mechanism so as to capture the incidence relation contained in the behavior sequence characteristics to obtain the behavior sequence incidence characteristics, and the behavior sequence incidence characteristics are associated with the correlation among the user behaviors. For example, the user 1 browses the good opinion of the product a, and then the user 1 makes a desire to purchase the product a and adds the product a to the shopping cart, even purchasing the product a. The series of behaviors of the user 1 actually reflects that the sequence of behaviors of the user has not only a precedence relationship but also a correlation, such as a correlation between browsing the good-rated content and purchasing. Because the sequence behaviors of the user often have a certain correlation or causal relationship, the recommendation effect can be improved based on the extracted characteristics by performing correlation extraction on the behavior sequence characteristics.
202. The server fuses the user characteristics of the user, the article characteristics of the article to be recommended and the behavior sequence related characteristics to obtain fusion characteristics, and determines recommendation probability based on the fusion characteristics, wherein the recommendation probability is used for indicating the probability of recommending the article to be recommended to the user.
In the embodiment of the present application, the user characteristics may include user basic attribute characteristics, such as age, gender, school calendar, city rating, and the like; user consumption characteristics such as total number of payments, total amount, number of payments over a period of time (24 hours, week, month, half year), amount of payments, and average amount per payment, etc. may also be included; the method and the device can also comprise user behavior characteristics such as browsing duration, page click times and the like, and the user characteristics are not limited in the embodiment of the application. The item features may include item base attribute features such as item category, item price, item brand, item purchase score, and item review sentiment; the method can further comprise item consumption characteristics, such as the number of times that the item is purchased, the number of times that the item is clicked and browsed, the number of times that a shopping cart is added, the number of times that the like item is purchased, and the like, and the characteristics of the item are not limited in the embodiments of the present application. The server can fuse the user features and the article features based on the article recommendation model to obtain user article combination features, and fuse the user article combination features and the behavior sequence features based on at least one feed-forward neural network to obtain fusion features. The server can predict the probability of recommending the item to be recommended to the user based on the fusion characteristics, so as to determine the recommendation probability.
203. And the server recommends the item to be recommended to the user in response to the recommendation probability being greater than the target recommendation probability.
The server can determine whether to recommend the item to be recommended to the user according to the relationship between the recommendation probability and the target recommendation probability, if the recommendation probability is greater than the target recommendation probability, the server can recommend the item to be recommended to the user, and if the recommendation probability is not greater than the target recommendation probability, the server can not recommend the item to be recommended to the user. The target recommendation probability can be set according to an application scenario, which is not limited in the embodiment of the present application.
In the embodiment of the application, the behavior sequence related features are obtained by extracting the related information in at least two behavior sequence features based on an article recommendation model, the behavior sequence related features of a user can be represented, then the behavior sequence related features are fused with the user features and the article features, and the recommendation probability is determined based on the fused features, so that the recommendation probability is related to the features of the user and the article to be recommended and is also related to the features of the behaviors executed by the user according to the time sequence, and the recommendation precision is high.
Fig. 3 is a flowchart of another item recommendation method provided in an embodiment of the present application, and as shown in fig. 3, an application to a server is taken as an example in the embodiment of the present application for description. The item recommendation method is realized by a server based on an item recommendation model, and comprises the following steps:
301. the server acquires the user characteristics of the user and the article characteristics of the article to be recommended.
It should be noted that the item to be recommended is an item that may be of interest to the user. The server can mine the sequence mode of the user according to the user behavior sequence of the user, find similar users of the user according to the sequence mode of the user, and determine the articles which are possibly interested by the user according to the articles purchased by the similar users.
In an alternative implementation manner, the step of mining the sequence pattern of the user by the server according to the user behavior sequence of the user may be: the server may determine a minimum support threshold according to the number of sequences of at least two user behavior sequences of the user and a minimum support rate, where the minimum support rate is positively correlated with the number of sequences. The server may determine at least one sequence pattern satisfying the minimum support threshold from the at least two sequences of user behavior through a sequence pattern mining algorithm. The minimum support threshold is higher as the number of the sequences is larger, so that the accuracy of the sequence pattern mining can be ensured by mining with higher support, and the recall ratio is improved by multi-round iterative mining. And with the change of the number of the iteration sequences in each round, the minimum support threshold is changed, namely, a plurality of minimum support thresholds are adopted during iteration, so that redundant sequence modes are reduced, and the mining efficiency is improved.
The minimum support threshold can be calculated by formula (1):
min_sup＝p×n (1)；
wherein min _ sup represents the minimum support degree, p represents the minimum support rate, and n represents the number of sequences.
The main steps of the sequence pattern mining algorithm are briefly introduced as follows: for any item, multiple users may purchase the item in a different sequence of user actions. (1) The server acquires a prefix of a user behavior sequence with the unit length of 1 and a projection data set corresponding to the prefix from the user characteristic sequences of a plurality of users. (2) The server counts the frequency of occurrence of each behavior sequence prefix, and adds the prefixes with the support degree higher than the minimum support degree to the data set to obtain a sequence mode of one prefix, which can also be called as a frequent one-item set time sequence mode. (3) The server conducts prefix recursive mining on the basis of the projection data set with the prefix of 1, namely conducting recursive mining on the prefix with the length of i and meeting the minimum support degree, wherein i is a positive integer. When recursive mining is carried out, a projection data set with the length of i as a prefix is mined, and if the projection data set is empty, recursion is returned; if the projection data set is not empty, the support degree of each item in the projection data set is counted, a single item higher than the minimum support degree is merged with the prefix with the length of i, and the item not higher than the minimum support degree is recursively returned. If i is equal to i +1, the prefix with the length of i +1 is each new prefix merged in the step (3), and the step (3) is executed recursively. (4) And returning the sequence modes of the plurality of users. Because the user behavior sequences have a certain sequence, the sequence patterns contained in the user behavior sequences can be mined based on a sequence pattern mining algorithm, and the sequence patterns are used for expressing the behavior habits shared by different users.
For example, taking the sequence information left when the user clicks browsing as an example, the user 1 clicks the X button on the X page to enter the Y page, and then the user 1 clicks the Y button again to enter the Z page after browsing for a period of time, so the browsing sequence of the user 1 may be represented as xxyz. The user 2 clicks the X button on the X page to enter the Y page, and then clicks the z button to return to the X page after browsing for a period of time, and then the browsing sequence of the user 2 can be represented as xyzx. Assuming that the minimum support threshold is 0.5, a prefix and the corresponding projection data set that satisfy the minimum support threshold can be seen in table 1.
TABLE 1
Accordingly, when the prefix length is 2, the two prefixes and the corresponding projection data set that satisfy the minimum support degree threshold can be referred to table 2.
TABLE 2
Accordingly, when the prefix length is 3, the three prefixes and the corresponding projection data sets that satisfy the minimum support threshold can be referred to in table 3.
TABLE 3
Then the user behavior sequence patterns obtained by mining the sequence patterns of the user 1 and the user 2 are X, Y, Xx, xY, XxY.
In an optional implementation manner, the server may further represent a series of behavior tracks of the user by using a behavior tag, and construct a user behavior sequence of the user according to a behavior code corresponding to the behavior tag. As shown in Table 4, the behavior tags illustratively include a purchase behavior, an add shopping cart behavior, a favorite behavior, a comment behavior, a search behavior, a login behavior, a registration behavior, and a browse behavior. Each behavior tag corresponds to a different behavior code. Of course, the behavior tag may also include other behaviors, which are not described in detail. The user behavior sequence is constructed through the behavior code corresponding to the user behavior tag, and the behavior category of the user in the actual application scene can be clearly and finely represented.
TABLE 4
| Behavior tag | Behavior coding |
| Purchasing behavior | h |
| Add shopping cart behavior | g |
| Collecting behavior | f |
| Commenting behaviors | e |
| Search behavior | d |
| Login behavior | c |
| Registration behavior | b |
| Browsing behavior | a |
For example, taking an e-commerce website as an example, the user 1 registers and logs in after entering the e-commerce website, enters a detail page of an item after browsing the page for a period of time, clicks a collection button to collect the item, then clicks an add shopping cart button to add the item to a shopping cart, and finally purchases the item, and then the user behavior sequence of the user 1 may be represented as bcafgh. The user 2 enters the e-commerce website to register and log in, clicks the search button to search for a specific item after the page is browsed for a period of time, clicks the add shopping cart button to add the specific item to the add shopping cart after browsing the specific item, then purchases the specific item, and finally clicks the collection button to collect the specific item, so that the user behavior sequence of the user 2 can be represented as bcdaghf. Assuming that the minimum support threshold is 0.5, based on the sequence pattern mining algorithm, a prefix and a corresponding projection data set meeting the minimum support threshold are mined, as shown in table 5.
TABLE 5
Accordingly, when the prefix length is 2, the two prefixes and the corresponding projection data set that satisfy the minimum support threshold can be seen from table 6.
TABLE 6
Accordingly, when the prefix length is 3, the three prefixes and the corresponding projection data sets satisfying the minimum support threshold can be seen from table 7.
TABLE 7
Accordingly, when the prefix length is 4, the four prefixes and the corresponding projection data sets satisfying the minimum support threshold can be seen in table 8.
TABLE 8
Accordingly, when the prefix length is 5, the five prefixes and the corresponding projection data sets satisfying the minimum support threshold can be seen in table 9.
TABLE 9
| Prefix of five items | Projection data set |
| bcagh | f |
It should be noted that, the above exemplary example is described by taking each behavior sequence as an example, where the length of each behavior sequence is 1, and in an actual scenario, the behavior sequence of the user may include multiple items. For example, if the browsing sequence of user 1 is represented by xyyz, and the user behavior sequence of user 1 encoded based on the corresponding behavior tag is represented by bcafgh, where a represents browsing behavior, the user behavior sequence of user 1 may be replaced by bc (xyyz) fgh, where the length of xyz is 5.
In an alternative implementation manner, the step of determining, by the server, the item to be recommended according to the similar user may be: the server may determine at least one similar user of the users, the similar user being a user satisfying a similar condition with the user. And the server responds to that any item in the at least one item purchased by the at least one similar user is not purchased by the user, and takes the item not purchased by the user as the item to be recommended. Wherein the similarity condition may be that the length of a common sequence pattern with the user is greater than a target length threshold corresponding to the user, the common sequence pattern being used to indicate the same portion of the sequence patterns of the two users. The size of the target length threshold is not limited in the embodiment of the present application, and may be 5, 7, or 10, for example. Since similar users are determined by the length of the common sequence pattern so that the similar users have the same characteristics in hobbies and consumption habits as the user, the similar users have potential associations with the user on items of interest.
For example, when a certain item is purchased, the tag of the behavior sequence of user 1 is denoted by bcafgh, the tag of the behavior sequence of user 2 is denoted by bcdaghf, and the common sequence pattern of user 1 and user 2 is bcagh, which denotes the behavior trace common to both users. Assuming that the target length threshold corresponding to user 1 is 5, since the length of bcagh is equal to 5, user 2 is a similar user to user 1. The server may select, from the items purchased by user 2, items not purchased by user 1, i.e. items that may be of interest to user 1.
In an optional implementation manner, as the complexity of the behavior trajectory of the user increases, the length of the behavior sequence of the user continuously increases, and the server may determine the target length threshold corresponding to the user according to the sequence mode of the user and the sequence modes of other users. Correspondingly, the step of determining the target length threshold corresponding to the user by the server may be: in response to a length of a common sequence pattern of a user and at least one other user not being less than a target length, the server may obtain a shortest length, a longest length, and a pattern accumulation ratio of the common sequence pattern. The server may determine a product of the difference of the longest length and the shortest length and the mode accumulation ratio. The server may use the sum of the shortest length and the product as the target length threshold. The value range of the mode accumulation ratio is 0 to 1, and the value of the mode accumulation ratio is not limited in the application. The embodiment of the application does not limit the size of the target length. When the user behavior sequence length is continuously increased, the target length threshold corresponding to the user is correspondingly adjusted, so that the problem that different users perform a large number of same behaviors and are easily judged as similar users by mistake due to the fact that content paths are deep when different users access different contents of the same website is solved, and the determined similar users can better meet the requirements.
The target length threshold may be calculated by equation (2):
L＝Lmin+(Lmax-Lmin)×R (2)；
where L denotes the target length threshold, LminMinimum length representing common sequence pattern, LmaxDenotes the longest length of the common sequence pattern and R denotes the pattern accumulation ratio.
302. The server inputs the user characteristics of the user, the item characteristics of the item to be recommended and at least two behavior sequence characteristics of the user into an item recommendation model, wherein the behavior sequence characteristics are used for representing the characteristics of behaviors performed by the user according to the time sequence.
In this embodiment of the present application, the server may determine the behavior sequence characteristics of the user through the behavior sequence of the user, and correspondingly, the step of determining at least two behavior sequence characteristics of the user by the server may be: the server acquires at least two user behavior sequences of a user, and for any user behavior sequence, the server can encode at least one sequence tag included in the user behavior sequence and perform weighting based on the support degree of the at least one sequence tag. The vector obtained by coding the sequence in the user behavior sequence can be used as the input of the model, has the position sequence attribute, and is weighted by the support degree of the sequence label, so that the weight of the same behavior can be increased, and the recommendation effect can be improved.
For example, the server obtains a user behavior sequence in which the user purchases 10 items, where if one sequence tag appears 9 times, the support degree of the sequence tag is 0.9, and if one sequence tag appears 3 times, the support degree of the sequence tag is 0.3.
303. And the server performs correlation extraction on the at least two behavior sequence characteristics based on the article recommendation model to obtain the behavior sequence correlation characteristics.
It should be noted that the server may adopt a Transformer structure to perform correlation extraction on the behavior sequence features. The Transformer can acquire global information compared with CNN (Convolutional Neural Networks), and improves the disadvantage of slow training of RNN (Recurrent Neural Networks), and uses self-attention mechanism to realize fast parallelization, and the structure of the Transformer can be shown in fig. 4. In FIG. 4, the transform structure includes a sum and normalization module, a feed forward module, and a multi-headed attention module. The Multi-Head Attention (Multi-Head Self Attention) module is composed of Attention modules with the same multilayer structure and different weight matrixes, so that the phenomenon that a Transformer only pays Attention to a part of features can be prevented, each Head pays Attention to different features through Multi-Head involvement, the Transformer can pay Attention to more features as a whole and learn the correlation among behavior sequence features, and the Transformer can learn different information in multiple aspects from different angles, so that the recommendation effect is improved.
304. The server fuses the user characteristics, the article characteristics and the behavior sequence related characteristics based on the article recommendation model to obtain fusion characteristics, and outputs recommendation probability based on the fusion characteristics, wherein the recommendation probability is used for indicating the probability of recommending the article to be recommended to the user.
It should be noted that the server may further fuse the user item combination feature and the behavior sequence related feature obtained in step 303 based on a GRU (Gate recovery Unit, an RNN) to obtain a fusion feature, and input the fusion feature into the feed-forward neural network, where the GRU is a model that can process sequence information better than L STM (L ong Short-Term Memory, long-Short Term Memory network) parameters, and may perform deep feature extraction.
305. And the server recommends the item to be recommended to the user in response to the recommendation probability being greater than the target recommendation probability.
It should be noted that, the above steps 301 to 305 are optional implementations of the item recommendation method provided by the embodiment of the present disclosure, and there are other optional implementations of the item recommendation method, for example, standing at an angle of an item to select a user: the server inputs the user characteristics of the user to be recommended, the item characteristics of the item and at least two behavior sequence characteristics associated with the item into an item recommendation model, the at least two behavior sequence characteristics associated with the item are used for representing the behavior characteristics of at least one other user when the item is purchased, and then the recommendation probability of recommending the item to the user to be recommended is input based on the model. The recommendation probability may be 0 and 100%, where a recommendation probability of 0 indicates that the item is not recommended to the user to be recommended, and a recommendation probability of 100% indicates that the item is recommended to the user to be recommended.
In the embodiment of the application, the behavior sequence related features are obtained by extracting the related information in at least two behavior sequence features based on an article recommendation model, the behavior sequence related features of a user can be represented, then the behavior sequence related features are fused with the user features and the article features, and the recommendation probability is determined based on the fused features, so that the recommendation probability is related to the features of the user and the article to be recommended and is also related to the features of the behaviors executed by the user according to the time sequence, and the recommendation precision is high.
Fig. 3 above is a flowchart illustrating an item recommendation method, where a training flow of an item recommendation model used in the item recommendation method may be shown in fig. 5, and fig. 5 is a training flow chart for training an item recommendation model according to an embodiment of the present application, and as shown in fig. 5, the method includes the following steps:
501. the server constructs sample user characteristics and sample item characteristics.
In the embodiment of the application, the server can acquire a large amount of authorized user information and article information, extract user characteristics from the user information according to the user information, and extract article characteristics from the article information. The content included in the user feature and the article feature can be referred to the above step 201, and will not be described herein again.
It should be noted that the server may respectively perform splicing and combining on the obtained multiple user features and multiple article features to construct a combined feature in the form of < user, article >, and then perform data preprocessing on the combined feature.
In an alternative implementation manner, the step of the server performing data preprocessing on the combined feature may be: for any combination feature, the server may determine a data missing value of the combination feature, and filter the combination feature if the data missing value of the combination feature exceeds a missing value filtering threshold; if the data missing value of the combined feature exceeds the missing value filtering threshold, the server may fill the missing value, such as continuously filling the feature with a mean value and filling the discrete feature with a constant as a separate category. The server may also filter single value features. The server may further process the abnormal value, for example, discard a value that is too large or within a target distribution range, which may be one ten thousandth or one thousandth, according to the characteristic distribution, which is not limited in this embodiment of the present application. The server can also perform discretization processing on the continuity characteristics in boxes, and perform encoding processing on the discrete characteristics, such as encoding through one-hot. The size of the missing value filtering threshold is determined by a product of the sample data size and a filtering coefficient, and a value range of the filtering coefficient is between 0 and 1, which is not limited in the embodiment of the present application. Due to the fact that the data are preprocessed, the data which do not meet the requirements can be cleaned, and therefore good sample data can be obtained.
It should be noted that the server may further determine at least one user with a target characteristic according to the user characteristics, and the target characteristic may be at least one of higher consumption, higher purchase frequency, and member service purchase. The server may construct a target characteristic user sample library using the target characteristic users as positive samples, so that potential target characteristic users may be determined from the new users based on the target characteristic sample library.
502. The server obtains a set of preferred items for a sample user.
In the embodiment of the present application, the preferred item set of the sample user is a set of items that the sample user has purchased. For example, if a sample user 1 purchases 10 items, the sample user 1's set of preferred items includes the 10 items.
It should be noted that the server may rank the items in the preferred item set according to the user's preference scores for the items. The step of the server determining the preference score of the user for the item may be: for any item, the server takes the product of the time when the user recently purchased the item, the number of times the user purchased the item within the target time period, and the amount of money the user purchased the item within the target time period as the preference score of the user for the item.
It should be noted that the server may also determine a similar user set based on the sequence pattern of the users, and the users in the similar user set have a common behavior pattern. The server may determine a set of items of potential interest to other users in the set of similar users based on the preferred item set of each user in the set of similar users. The server can take the items in the potential interest item set as the items to be recommended for the other users. The step of determining the sequence mode of the user may refer to step 302, which is not described herein again.
503. The server constructs sample behavior sequence features.
See step 201, which is not described herein.
504. And the server takes the preference item set of the sample user as a label, and trains an initial model based on the sample user characteristics, the sample item characteristics and the sample behavior sequence characteristics.
In this embodiment, the step of the server training the initial model may be: the server may obtain a set of preferred items of a sample user as training labels for a model to be trained. The server can obtain at least two sample behavior sequence characteristics based on at least two sample user behavior sequences of the sample user, wherein the sample behavior sequence characteristics are used for representing behavior characteristics of the sample user when the sample user purchases the items in the preference item set. The server can input the sample user characteristics of the sample object, the sample object characteristics of at least two preference objects in the preference object set and the at least two sample behavior sequence characteristics into the model to be trained, and adjust the model parameters according to the output result. The processing procedure of the model on the sample user characteristic, the sample article characteristic, and the sample behavior sequence characteristic may refer to step 303 and step 304, which are not described herein again.
It should be noted that, because the server uses the preferred item set as the label, the output of the initial model is the recommendation probability of each item in the preferred item set of the prediction sample user, and the ranking is performed according to the recommendation probability. Thus, the server may employ the activation function as an output layer of the model, with the loss function to compute the model loss. The activation function may be a sigmoid function, a tanh function, a Relu function, and the like, which is not limited in this embodiment of the application. The loss function may be a cross entropy loss function, an average absolute error function, a root mean square error function, and the like, which is not limited in this application.
As for the structure of the item recommendation model, see fig. 6, fig. 6 is a schematic structural diagram of an item recommendation model provided in an embodiment of the present application, in fig. 6, a server inputs n user features and m item features into the item recommendation model, and fuses the user features and the item features based on the item recommendation model to obtain user item combination features, where n and m are positive integers. The server inputs N user behavior sequence characteristics into a Transformer module in an item recommendation model, and the Transformer module outputs behavior sequence related characteristics, wherein N is a positive integer larger than 2. And then fusing the combination characteristics of the user articles and the related characteristics of the behavior sequences based on the GRU layer to obtain fusion characteristics, inputting the fusion characteristics into a feedforward neural network, and outputting recommendation probability based on an activation function.
505. And when the server responds to the condition that the training is finished, taking the model obtained by training as an article recommendation model.
In this embodiment of the present application, the training completion condition may be that a target training number is reached or a model converges, and this is not limited in this embodiment of the present application.
It should be noted that, when the initial model is trained, the training may be performed from the perspective of the user, that is, the user recommends the item, or the training may be performed from the perspective of the item, that is, the user selects the recommended item. The two training modes have the same thinking, and the difference is that the action sequence of purchasing the preferred articles by the user is input during the training from the user perspective, and the preferred article set of the user is used as a label; and when training from the perspective of an item, the input is a sequence of actions of a plurality of users to purchase the item, and at least one user who prefers the item is used as a label. The embodiment of the application is exemplarily described by taking training from the perspective of a user as an example.
It should be noted that, the above steps 501 to 505 are optional implementations of the item recommendation model training process, and correspondingly, the server may also train the item recommendation model in other optional manners, for example, as shown in fig. 7, fig. 7 is a training flowchart of another training item recommendation model provided in this embodiment of the present application. The method comprises the steps of 701, building user characteristics and article characteristics by a server, 702, mining user behavior sequences by the server based on a sequence mode, 703, mining potential articles of interest of the user based on behavior sequence matching by the server, 704, building a Transformer structure by the server to capture behavior correlation information, and 705, building an article recommendation model by the server to carry out personalized article recommendation.
In the embodiment of the application, the recommendation model is built by combining the sequence mode of the user and the Transformer structure, the behavior sequence characteristics of the user are considered, the attention mechanism is adopted to capture the correlation information of the behavior sequence of the user based on the Transformer structure, and the recommendation effect of the recommendation model is improved.
Fig. 8 is a block diagram of an item recommendation device according to an embodiment of the present application. The apparatus is used for executing the steps executed by the item recommendation method, and referring to fig. 8, the apparatus includes: a feature extraction module 801, a feature fusion module 802, and an item recommendation module 803.
A feature extraction module 801, configured to perform correlation extraction on at least two behavior sequence features of a user to obtain a behavior sequence correlation feature, where the behavior sequence feature is used to represent a feature of a behavior executed by the user according to a time sequence;
a feature fusion module 802, configured to fuse a user feature of the user, an article feature of the to-be-recommended article, and a behavior sequence related feature to obtain a fusion feature, and determine a recommendation probability based on the fusion feature, where the recommendation probability is used to indicate a probability of recommending the to-be-recommended article to the user;
and the article recommending module is used for recommending the article to be recommended to the user in response to the recommending probability being greater than the target recommending probability.
In an optional implementation, the apparatus further includes: the article determining module is used for determining at least one similar user of the user, wherein the similar user is a user meeting similar conditions with the user; and in response to any one of the at least one item purchased by the at least one similar user, the user does not purchase the item, and the unpurchased item is taken as the item to be recommended.
In an alternative implementation, the similar condition is:
the length of the common sequence mode of the user is not less than a target length threshold corresponding to the user, the common sequence mode is used for indicating the same part of the sequence modes of the two users, and the target length threshold is used for indicating the shortest length of the common sequence mode of the user.
In an optional implementation, the apparatus further includes: a target length threshold determination module, configured to, in response to that a length of a common sequence pattern of the user and at least one other user is not less than a target length, obtain a shortest length, a longest length, and a mode accumulation ratio of the common sequence pattern; determining a product of a difference of the longest length and the shortest length and the mode accumulation ratio; and taking the sum of the shortest length and the product as the target length threshold.
In an optional implementation, the apparatus further includes:
a minimum support degree determining module, configured to determine a minimum support degree threshold according to the sequence number and a minimum support rate of at least two user behavior sequences of the user, where the minimum support rate is positively correlated to the sequence number;
and the sequence pattern determining module is used for determining at least one sequence pattern meeting the minimum support degree threshold value from the at least two user behavior sequences through a sequence pattern mining algorithm.
In an optional implementation, the apparatus further includes: the behavior sequence feature construction module is used for acquiring at least two user behavior sequences of the user; for any user behavior sequence, at least one sequence label included in the user behavior sequence is coded, and weighting is carried out based on the support degree of the at least one sequence label.
In an optional implementation manner, the apparatus further includes a model training apparatus, configured to obtain a preferred item set of the sample user, where the preferred item set is used as a training label of the model to be trained; acquiring at least two sample behavior sequence characteristics based on at least two sample user behavior sequences of the sample user, wherein the sample behavior sequence characteristics are used for representing behavior characteristics of the sample user when purchasing the articles in the preference article set; inputting the sample user characteristics of the sample object, the sample object characteristics of at least two preference objects in the preference object set and the at least two sample behavior sequence characteristics into a model to be trained, and adjusting model parameters according to an output result; and in response to the training completion condition being reached, taking the trained model as the item recommendation model.
In the embodiment of the application, the behavior sequence related features are obtained by extracting the related information in at least two behavior sequence features, the behavior sequence related features of a user can be represented, the behavior sequence related features are fused with the user features and the article features, and the recommendation probability is determined based on the fused feature, so that the recommendation probability is related to the features of the user and the article to be recommended and is also related to the features of behaviors executed by the user according to the time sequence, and the recommendation precision is high.
It should be noted that: in practical applications, the function distribution may be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the article recommendation device and the article recommendation method provided by the above embodiments belong to the same concept, and specific implementation processes thereof are detailed in the method embodiments and are not described herein again.
Fig. 9 is a schematic structural diagram of a server according to an embodiment of the present application, where the server 900 may generate a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 901 and one or more memories 902, where the memory 902 stores at least one instruction, and the at least one instruction is loaded and executed by the processors 901 to implement the methods provided by the foregoing method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface, so as to perform input/output, and the server may also include other components for implementing the functions of the device, which are not described herein again.
The embodiment of the present application also provides a computer-readable storage medium, which is applied to a server, and the computer-readable storage medium stores at least one program code, and the at least one program code is used for being executed by a processor and implementing the operations performed by the server in the item recommendation method in the embodiment of the present application.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The above description is only exemplary of the present application and should not be taken as limiting, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims (15)
1. An item recommendation method, characterized in that the method comprises:
performing relevance extraction on at least two behavior sequence characteristics of a user to obtain behavior sequence relevant characteristics, wherein the behavior sequence characteristics are used for representing characteristics of behaviors executed by the user according to a time sequence;
fusing the user characteristics of the user, the article characteristics of the articles to be recommended and the behavior sequence related characteristics to obtain fused characteristics, and determining recommendation probability based on the fused characteristics, wherein the recommendation probability is used for indicating the probability of recommending the articles to be recommended to the user;
and recommending the item to be recommended to the user in response to the recommendation probability being greater than a target recommendation probability.
2. The method of claim 1, wherein the item to be recommended is determined by:
determining at least one similar user of the users, wherein the similar user is a user meeting similar conditions with the users;
and in response to the fact that any one of at least one item purchased by the at least one similar user is not purchased by the user, taking the unpurchased item as the item to be recommended.
3. The method according to claim 2, wherein the similar condition is:
the length of the common sequence mode of the user is not less than a target length threshold corresponding to the user, the common sequence mode is used for indicating the same part of the sequence modes of the two users, and the target length threshold is used for indicating the shortest length of the common sequence mode of the users.
4. The method of claim 3, wherein the target length threshold corresponding to the user is determined by:
in response to that the length of a common sequence mode of the user and at least one other user is not less than a target length, obtaining the shortest length, the longest length and a mode accumulation ratio of the common sequence mode;
determining a product of a difference of the longest length and the shortest length and the mode accumulation ratio;
taking the sum of the shortest length and the product as the target length threshold.
5. The method according to claim 3, wherein before performing the correlation extraction on the at least two behavior sequence features of the user to obtain the behavior sequence correlation feature, the method further comprises:
determining a minimum support threshold according to the sequence number and the minimum support rate of at least two user behavior sequences of the user, wherein the minimum support rate is positively correlated with the sequence number;
determining, by a sequence pattern mining algorithm, at least one sequence pattern from the at least two sequences of user behavior that satisfies the minimum support threshold.
6. The method according to claim 1, characterized in that the at least two behavior sequence features are obtained by:
acquiring at least two user behavior sequences of the user;
for any user behavior sequence, at least one sequence label included in the user behavior sequence is coded, and weighting is carried out based on the support degree of the at least one sequence label.
7. The method of claim 1, wherein the item recommendation method is implemented based on an item recommendation model trained by:
acquiring a preference item set of a sample user, wherein the preference item set is used as a training label of a model to be trained;
obtaining at least two sample behavior sequence characteristics based on at least two sample user behavior sequences of the sample user, wherein the sample behavior sequence characteristics are used for representing behavior characteristics of the sample user when the sample user purchases the items in the preference item set;
inputting the sample user characteristics of the sample object, the sample object characteristics of at least two preference objects in the preference object set and the at least two sample behavior sequence characteristics into a model to be trained, and adjusting model parameters according to an output result;
and when the training completion condition is met, taking the trained model as the item recommendation model.
8. An item recommendation device, the device comprising:
the characteristic extraction module is used for carrying out correlation extraction on at least two behavior sequence characteristics of a user to obtain behavior sequence correlation characteristics, and the behavior sequence characteristics are used for expressing the characteristics of behaviors executed by the user according to a time sequence;
the characteristic fusion module is used for fusing the user characteristics of the user, the article characteristics of the articles to be recommended and the behavior sequence related characteristics to obtain fusion characteristics, and determining recommendation probability based on the fusion characteristics, wherein the recommendation probability is used for indicating the probability of recommending the articles to be recommended to the user;
and the article recommending module is used for recommending the article to be recommended to the user in response to the recommending probability being greater than the target recommending probability.
9. The apparatus of claim 8, further comprising: the article determining module is used for determining at least one similar user of the users, wherein the similar user is a user meeting similar conditions with the user; and in response to the fact that any one of at least one item purchased by the at least one similar user is not purchased by the user, taking the unpurchased item as the item to be recommended.
10. The apparatus of claim 9, wherein the similar condition is:
the length of the common sequence mode of the user is not less than a target length threshold corresponding to the user, the common sequence mode is used for indicating the same part of the sequence modes of the two users, and the target length threshold is used for indicating the shortest length of the common sequence mode of the users.
11. The apparatus of claim 10, further comprising: a target length threshold determination module, configured to, in response to that a length of a common sequence pattern of the user and at least one other user is not less than a target length, obtain a shortest length, a longest length, and a pattern accumulation ratio of the common sequence pattern; determining a product of a difference of the longest length and the shortest length and the mode accumulation ratio; taking the sum of the shortest length and the product as the target length threshold.
12. The apparatus of claim 10, further comprising:
a minimum support degree determining module, configured to determine a minimum support degree threshold according to a sequence number and a minimum support rate of at least two user behavior sequences of the user, where the minimum support rate is positively correlated to the sequence number;
and the sequence pattern determining module is used for determining at least one sequence pattern meeting the minimum support degree threshold value from the at least two user behavior sequences through a sequence pattern mining algorithm.
13. The apparatus of claim 8, further comprising: the behavior sequence feature construction module is used for acquiring at least two user behavior sequences of the user; for any user behavior sequence, at least one sequence label included in the user behavior sequence is coded, and weighting is carried out based on the support degree of the at least one sequence label.
14. A server, characterized in that the server comprises a processor and a memory for storing at least one piece of program code, which is loaded by the processor and which performs the item recommendation method according to any one of claims 1 to 7.
15. A storage medium for storing at least one program code for performing the item recommendation method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010264243.1A CN111461841B (en) | 2020-04-07 | 2020-04-07 | Article recommendation method, device, server and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010264243.1A CN111461841B (en) | 2020-04-07 | 2020-04-07 | Article recommendation method, device, server and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111461841A true CN111461841A (en) | 2020-07-28 |
| CN111461841B CN111461841B (en) | 2023-04-07 |
Family
ID=71684490
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010264243.1A Active CN111461841B (en) | 2020-04-07 | 2020-04-07 | Article recommendation method, device, server and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461841B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112163963A (en) * | 2020-09-27 | 2021-01-01 | 中国平安财产保险股份有限公司 | Service recommendation method and device, computer equipment and storage medium |
| CN112381623A (en) * | 2020-12-04 | 2021-02-19 | 南京浪脆电子商务有限公司 | Electronic commerce platform commodity intelligent recommendation method based on big data analysis |
| CN112381615A (en) * | 2020-11-27 | 2021-02-19 | 华中科技大学 | Short sequence recommendation method based on user repeated behavior pattern mining |
| CN112507234A (en) * | 2020-12-21 | 2021-03-16 | 北京明略软件系统有限公司 | Material pushing method and device, storage medium and electronic equipment |
| CN112712418A (en) * | 2021-03-25 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Method and device for determining recommended commodity information, storage medium and electronic equipment |
| CN112785390A (en) * | 2021-02-02 | 2021-05-11 | 微民保险代理有限公司 | Recommendation processing method and device, terminal device and storage medium |
| CN113095884A (en) * | 2021-04-21 | 2021-07-09 | 山东大学 | Television member user recommendation method and system based on user feedback |
| CN113379482A (en) * | 2021-05-28 | 2021-09-10 | 车智互联（北京）科技有限公司 | Item recommendation method, computing device and storage medium |
| CN117196640A (en) * | 2023-11-06 | 2023-12-08 | 青岛巨商汇网络科技有限公司 | Full-flow visual management system and method based on service experience |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007049282A2 (en) * | 2005-10-26 | 2007-05-03 | Cortica Ltd. | A computing device, a system and a method for parallel processing of data streams |
| CN103400286A (en) * | 2013-08-02 | 2013-11-20 | 世纪禾光科技发展（北京）有限公司 | Recommendation system and method for user-behavior-based article characteristic marking |
| CN107423442A (en) * | 2017-08-07 | 2017-12-01 | 火烈鸟网络(广州)股份有限公司 | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis |
| CN108665333A (en) * | 2017-03-31 | 2018-10-16 | 北京京东尚科信息技术有限公司 | Method of Commodity Recommendation, device, electronic equipment and storage medium |
| CN109064283A (en) * | 2018-07-27 | 2018-12-21 | 广州视源电子科技股份有限公司 | Method of Commodity Recommendation and device, computer readable storage medium |
| CN109685631A (en) * | 2019-01-10 | 2019-04-26 | 博拉网络股份有限公司 | A kind of personalized recommendation method based on big data user behavior analysis |
| CN109919699A (en) * | 2017-12-12 | 2019-06-21 | 北京京东尚科信息技术有限公司 | Item recommendation method, item recommendation system and computer-readable medium |
| CN110012356A (en) * | 2019-04-16 | 2019-07-12 | 腾讯科技（深圳）有限公司 | Video recommendation method, device and equipment and computer storage medium |
-
2020
- 2020-04-07 CN CN202010264243.1A patent/CN111461841B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007049282A2 (en) * | 2005-10-26 | 2007-05-03 | Cortica Ltd. | A computing device, a system and a method for parallel processing of data streams |
| CN103400286A (en) * | 2013-08-02 | 2013-11-20 | 世纪禾光科技发展（北京）有限公司 | Recommendation system and method for user-behavior-based article characteristic marking |
| CN108665333A (en) * | 2017-03-31 | 2018-10-16 | 北京京东尚科信息技术有限公司 | Method of Commodity Recommendation, device, electronic equipment and storage medium |
| CN107423442A (en) * | 2017-08-07 | 2017-12-01 | 火烈鸟网络(广州)股份有限公司 | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis |
| CN109919699A (en) * | 2017-12-12 | 2019-06-21 | 北京京东尚科信息技术有限公司 | Item recommendation method, item recommendation system and computer-readable medium |
| CN109064283A (en) * | 2018-07-27 | 2018-12-21 | 广州视源电子科技股份有限公司 | Method of Commodity Recommendation and device, computer readable storage medium |
| CN109685631A (en) * | 2019-01-10 | 2019-04-26 | 博拉网络股份有限公司 | A kind of personalized recommendation method based on big data user behavior analysis |
| CN110012356A (en) * | 2019-04-16 | 2019-07-12 | 腾讯科技（深圳）有限公司 | Video recommendation method, device and equipment and computer storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 张钰琦: "基于行为模型的个性化推荐" * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112163963A (en) * | 2020-09-27 | 2021-01-01 | 中国平安财产保险股份有限公司 | Service recommendation method and device, computer equipment and storage medium |
| CN112163963B (en) * | 2020-09-27 | 2024-04-02 | 中国平安财产保险股份有限公司 | Service recommendation method, device, computer equipment and storage medium |
| CN112381615A (en) * | 2020-11-27 | 2021-02-19 | 华中科技大学 | Short sequence recommendation method based on user repeated behavior pattern mining |
| CN112381623A (en) * | 2020-12-04 | 2021-02-19 | 南京浪脆电子商务有限公司 | Electronic commerce platform commodity intelligent recommendation method based on big data analysis |
| CN112381623B (en) * | 2020-12-04 | 2021-07-06 | 深圳市快云科技有限公司 | Electronic commerce platform commodity intelligent recommendation method based on big data analysis |
| CN112507234A (en) * | 2020-12-21 | 2021-03-16 | 北京明略软件系统有限公司 | Material pushing method and device, storage medium and electronic equipment |
| CN112507234B (en) * | 2020-12-21 | 2024-03-01 | 北京明略软件系统有限公司 | Material pushing method and device, storage medium and electronic equipment |
| CN112785390B (en) * | 2021-02-02 | 2024-02-09 | 微民保险代理有限公司 | Recommendation processing method, device, terminal equipment and storage medium |
| CN112785390A (en) * | 2021-02-02 | 2021-05-11 | 微民保险代理有限公司 | Recommendation processing method and device, terminal device and storage medium |
| CN112712418A (en) * | 2021-03-25 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Method and device for determining recommended commodity information, storage medium and electronic equipment |
| CN112712418B (en) * | 2021-03-25 | 2021-07-09 | 腾讯科技（深圳）有限公司 | Method and device for determining recommended commodity information, storage medium and electronic equipment |
| CN113095884A (en) * | 2021-04-21 | 2021-07-09 | 山东大学 | Television member user recommendation method and system based on user feedback |
| CN113095884B (en) * | 2021-04-21 | 2022-12-06 | 山东大学 | Television member user recommendation method and system based on user feedback |
| CN113379482B (en) * | 2021-05-28 | 2023-12-01 | 车智互联（北京）科技有限公司 | Article recommendation method, computing device and storage medium |
| CN113379482A (en) * | 2021-05-28 | 2021-09-10 | 车智互联（北京）科技有限公司 | Item recommendation method, computing device and storage medium |
| CN117196640A (en) * | 2023-11-06 | 2023-12-08 | 青岛巨商汇网络科技有限公司 | Full-flow visual management system and method based on service experience |
| CN117196640B (en) * | 2023-11-06 | 2024-02-02 | 青岛巨商汇网络科技有限公司 | Full-flow visual management system and method based on service experience |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111461841B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111461841B (en) | Article recommendation method, device, server and storage medium | |
| CN108629665B (en) | Personalized commodity recommendation method and system | |
| CN110941740B (en) | Video recommendation method and computer-readable storage medium | |
| CN110532479A (en) | A kind of information recommendation method, device and equipment | |
| CN111506820B (en) | Recommendation model, recommendation method, recommendation device, recommendation equipment and recommendation storage medium | |
| CN110717098A (en) | Meta-path-based context-aware user modeling method and sequence recommendation method | |
| CN111709810A (en) | Object recommendation method and device based on recommendation model | |
| CN110909182A (en) | Multimedia resource searching method and device, computer equipment and storage medium | |
| EP2960849A1 (en) | Method and system for recommending an item to a user | |
| KR20200088134A (en) | An apparatus for improving way of product recommendation and a method thereof | |
| CN113191838A (en) | Shopping recommendation method and system based on heterogeneous graph neural network | |
| Eide et al. | Deep neural network marketplace recommenders in online experiments | |
| CN111429161A (en) | Feature extraction method, feature extraction device, storage medium, and electronic apparatus | |
| CN116823410B (en) | Data processing method, object processing method, recommending method and computing device | |
| CN113220974A (en) | Click rate prediction model training and search recall method, device, equipment and medium | |
| CN113032676A (en) | Recommendation method and system based on micro-feedback | |
| US20170147721A1 (en) | Generating a model based on input | |
| Zheng et al. | Incorporating price into recommendation with graph convolutional networks | |
| CN115687690A (en) | Video recommendation method and device, electronic equipment and storage medium | |
| CN111460300B (en) | Network content pushing method, device and storage medium | |
| CN114429384A (en) | Intelligent product recommendation method and system based on e-commerce platform | |
| CN114693409A (en) | Product matching method, device, computer equipment, storage medium and program product | |
| CN114090848A (en) | Data recommendation and classification method, feature fusion model and electronic equipment | |
| CN113761349A (en) | Mining method and device for recommendation information | |
| Hu et al. | Utilizing users' tipping points in E-commerce Recommender systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |