CN113962753B - Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment - Google Patents

Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment Download PDF

Info

Publication number
CN113962753B
CN113962753B CN202111577701.8A CN202111577701A CN113962753B CN 113962753 B CN113962753 B CN 113962753B CN 202111577701 A CN202111577701 A CN 202111577701A CN 113962753 B CN113962753 B CN 113962753B
Authority
CN
China
Prior art keywords
candidate
commodity
historical
operation data
target
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.)
Active
Application number
CN202111577701.8A
Other languages
Chinese (zh)
Other versions
CN113962753A (en
Inventor
陶文武
高理强
黄健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Koubei Network Technology Co Ltd
Original Assignee
Zhejiang Koubei Network Technology Co Ltd
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 Zhejiang Koubei Network Technology Co Ltd filed Critical Zhejiang Koubei Network Technology Co Ltd
Priority to CN202111577701.8A priority Critical patent/CN113962753B/en
Publication of CN113962753A publication Critical patent/CN113962753A/en
Application granted granted Critical
Publication of CN113962753B publication Critical patent/CN113962753B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • 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
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Databases & Information Systems (AREA)
  • Finance (AREA)
  • Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the application provides a candidate commodity sorting method, a candidate commodity display device, electronic equipment and a computer storage medium, wherein the sorting method obtains historical effective operation data in advance, and the historical effective operation data comprises historical time information and historical space information when a target user effectively operates commodities; and then acquiring behavior data of the target user aiming at the target candidate commodity based on the historical effective operation data and the attention weight of the target candidate commodity, wherein the acquired behavior data also integrates historical time information and historical space information. Therefore, the obtained ranking information of the candidate commodities based on the characteristic information of the target candidate commodity and the behavior data of the target user, which is fused with the historical time information and the historical space information, for the target candidate commodity can embody the preference of the target user under the condition of the time information and the space information matched with the historical time information and the historical space information, and further is matched with the target user.

Description

Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment
Technical Field
The present application relates to the field of computer technologies, and in particular, to a candidate commodity sorting method, a candidate commodity display method, an apparatus, an electronic device, and a computer storage medium.
Background
With the rapid development of science and technology, the living material level is continuously improved, and meanwhile, the online shopping mode is more popular. When a user purchases in an online shopping mode, a plurality of commodities are recommended to the user on a page applied by a user side. Generally, when a plurality of products are recommended to a user, the plurality of products are sorted according to the degree of correlation between the products and the user.
Specifically, when recommending commodities to a user on a page applied by a user side, the order of the commodities may be determined according to the correlation between the commodities and the user, and after the order of the commodities is determined, all the commodities in the commodities may be displayed in a commodity floor manner according to the order, or some of the commodities in the commodities may be displayed in a commodity floor manner according to the order. The commodity floor may be a floor-by-floor display of a plurality of commodities, but generally, commodities ranked in the top may be ranked on an upper commodity floor and displayed preferentially to the user. Therefore, how to determine the ranking of a plurality of products when recommending the products to the user is a technical problem which needs to be solved at present.
Disclosure of Invention
The embodiment of the application provides a candidate commodity ordering method which is used for determining the ordering of a plurality of commodities matched with a user when the commodities are recommended to the user. The embodiment of the application also provides a candidate commodity display method, a candidate commodity sequencing device, a candidate commodity display device, electronic equipment and a computer storage medium.
The embodiment of the application provides a candidate commodity ordering method, which comprises the following steps: recalling a plurality of candidate items for the target user; obtaining historical effective operation data of a target user for a commodity, wherein the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity; acquiring behavior data of the target user for any target candidate commodity in the plurality of candidate commodities according to the historical effective operation data and the attention weight of the target user for the target candidate commodity; and obtaining the ranking information of the candidate commodities according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity.
Optionally, the obtaining of the historical valid operation data of the target user for the commodity includes: and obtaining historical effective operation data of the target user for the commodity within the preset time and for preset times.
Optionally, the method further includes: and obtaining the attention weight of the target candidate commodity according to the historical effective operation data and the characteristic information of the target candidate commodity.
Optionally, the obtaining the attention weight of the target candidate product according to the historical valid operation data and the feature information of the target candidate product includes: obtaining each historical effective operation data vector used for carrying out vector representation on each time of historical effective operation data in the historical effective operation data of preset times; obtaining a characteristic information vector of the target candidate commodity for vector representation of the characteristic information of the target candidate commodity; and obtaining the attention weight of the target candidate commodity aiming at the historical effective operation data of each time according to each historical effective operation data vector and the characteristic information vector of the target candidate commodity.
Optionally, the history valid operation data further includes: the target user effectively operates the characteristic information of the operated commodity; the obtaining of each history valid operation data vector for vector representation of each history valid operation data in the history valid operation data of the preset times comprises: aiming at the historical effective operation data of each time, obtaining a characteristic information vector of the operated commodity, wherein the characteristic information vector is used for carrying out vector representation on the characteristic information of the operated commodity in the historical effective operation data of each time; obtaining a historical time information vector used for carrying out vector representation on the historical time information in the historical effective operation data each time; obtaining a historical spatial information vector used for carrying out vector representation on the historical spatial information in the historical effective operation data each time; and obtaining each history effective operation data vector used for carrying out vector representation on each history effective operation data in the history effective operation data of preset times based on the characteristic information vector, the history time information vector and the history space information vector of the operated commodity.
Optionally, the obtaining, according to each historical valid operation data vector and the feature information vector of the target candidate product, the attention weight of the target candidate product for each historical valid operation data includes: and performing attention operation on each historical effective operation data vector and the characteristic information vector of the target candidate commodity respectively to obtain the attention weight of the target candidate commodity aiming at each historical effective operation data.
Optionally, the following formula is adopted to perform attention operation on each historical valid operation data vector and the feature information vector of the target candidate commodity:
Figure 429331DEST_PATH_IMAGE002
wherein,Qfor a history of valid operation data vectors,K、Vis the feature information vector of the target candidate commodity,d k is composed ofKThe dimensions of the vector.
Optionally, the obtaining, according to the historical valid operation data and the attention weight of the target user for any target candidate good in the plurality of candidate goods, behavior data of the target user for the target candidate good includes: and performing corresponding weighted summation operation on each historical effective operation data vector and the attention weight of the target candidate commodity aiming at the historical effective operation data of each time to obtain a behavior data vector for performing vector representation on the behavior data of the target user aiming at the target candidate commodity. Optionally, the obtaining, according to the behavior data of the target user for the target candidate product and the feature information of the target candidate product, the ranking information of the plurality of candidate products includes: acquiring attention data of the target user for the target candidate commodity according to the behavior data of the target user for the target candidate commodity and the characteristic information of the target candidate commodity; obtaining ranking information of the candidate commodities according to the attention data of the target user for each candidate commodity in the candidate commodities.
Optionally, the obtaining, according to the behavior data of the target user for the target candidate product and the feature information of the target candidate product, the attention data of the target user for the target candidate product includes: and obtaining the attention data of the target user for the target candidate commodity by taking the behavior data vector and the feature information vector of the target candidate commodity as input data of an attention data obtaining model, wherein the attention data obtaining model is used for obtaining the attention data of the user for the candidate commodity according to the behavior data vector of the user for the candidate commodity and the feature information vector of the candidate commodity.
Optionally, the obtaining the attention data of the target user for the target candidate commodity by using the behavior data vector and the feature information vector of the target candidate commodity as input data of an attention data obtaining model includes: splicing the behavior data vector with the characteristic information vector of the target candidate commodity to obtain a spliced vector; and taking the spliced vector as input data of an attention data acquisition model to acquire the attention data of the target user for the target candidate commodity.
Optionally, the attention data obtaining model is obtained by: obtaining historical operation sample data of a user for a commodity, an attention weight sample of the user for a candidate commodity, a feature information sample of the candidate commodity and an attention sample data of the user for the candidate commodity; wherein the historical operation sample data, the attention weight sample of the user for the candidate commodity, the characteristic information sample of the candidate commodity and the attention sample data of the user for the candidate commodity correspond to each other; training a deep convolutional neural network based on the historical operation sample data, the attention weight sample of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity to obtain the attention data obtaining model.
Optionally, the training the deep convolutional neural network based on the historical operation sample data, the sample of the attention weight of the user for the candidate commodity, the sample of the feature information of the candidate commodity, and the sample of the attention of the user for the candidate commodity includes: acquiring behavior sample data of the user for the candidate commodity according to the historical operation sample data and the attention weight sample of the user for the candidate commodity; training the deep convolutional neural network based on the behavior sample data of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity.
Optionally, the training the deep convolutional neural network based on the behavior sample data of the user for the candidate commodity, the feature information sample of the candidate commodity, and the attention sample data of the user for the candidate commodity includes: obtaining a behavior sample data vector for vector representation of the behavior sample data of the candidate commodity for the user; obtaining a characteristic information sample vector of the candidate commodity for carrying out vector representation on the characteristic information sample of the candidate commodity; training the deep convolutional neural network based on the behavior sample data vector, the characteristic information sample vector of the candidate commodity and the attention sample data of the user for the candidate commodity.
Optionally, the historical operation sample data includes historical valid operation sample data and historical invalid operation sample data; when the deep convolutional neural network is trained, the historical valid operation sample data is used as a positive sample of training data, and the historical invalid operation sample data is used as a negative sample of the training data.
Optionally, the method further includes: acquiring a first request message sent by a user side and used for requesting to acquire page information; and sending the sequencing information of the candidate commodities to the user terminal aiming at the first request message.
The embodiment of the application further provides a candidate commodity display method, which comprises the following steps: obtaining the sequencing information of a plurality of candidate commodities sent by a server; the ranking information of the candidate commodities is obtained according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity; the behavior data of the target user for the target candidate commodity is obtained according to historical effective operation data of the target user for the commodity and attention weight of the target user for any one target candidate commodity in the candidate commodities, and the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity; and displaying recommended candidate commodities recommended to the target user based on the ranking information of the candidate commodities.
Optionally, the target candidate commodity is any one of the candidate commodities; the attention weight of the target candidate commodity is obtained according to the historical valid operation data and the characteristic information of the target candidate commodity.
Optionally, the method further includes: sending a first request message for requesting to acquire page information to a server; the obtaining of the ranking information of the plurality of candidate commodities sent by the server includes: and obtaining the ranking information of the candidate commodities sent by the server aiming at the first request message.
Optionally, the displaying, based on the ranking information of the plurality of candidate commodities, a recommended candidate commodity for recommendation to a target user includes: determining recommended candidate commodities for recommendation to a target user based on the ranking information of the candidate commodities; and displaying the determined recommended candidate commodity recommended to the target user on a page of the user side.
Optionally, the method further includes: acquiring a second request message which is sent by the target user through a page of a user side and used for requesting to display the to-be-recommended commodity; the presenting of the recommended candidate goods for recommendation to the target user includes: and displaying recommended candidate commodities recommended to the target user on the page based on the second request message.
Correspondingly, the embodiment of the application provides a candidate commodity sequencing device, which comprises: a recall unit that recalls a plurality of candidate commodities for a target user; a history effective operation data obtaining unit, configured to obtain history effective operation data of a target user for a commodity, where the history effective operation data includes history time information and history space information when the target user performs an effective operation on the commodity; a behavior data obtaining unit, configured to obtain behavior data of the target user for any target candidate commodity in the plurality of candidate commodities according to the historical valid operation data and an attention weight of the target user for the target candidate commodity; and the ranking information obtaining unit is used for obtaining ranking information of the candidate commodities according to the behavior data of the target user for the target candidate commodity and the characteristic information of the target candidate commodity.
Correspondingly, the embodiment of the present application further provides a candidate product display device, including: the system comprises a ranking information obtaining unit, a ranking information obtaining unit and a ranking information obtaining unit, wherein the ranking information obtaining unit is used for obtaining ranking information of a plurality of candidate commodities sent by a server; the ranking information of the candidate commodities is obtained according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity; the behavior data of the target user for the target candidate commodity is obtained according to historical effective operation data of the target user for the commodity and attention weight of the target user for any one target candidate commodity in the candidate commodities, and the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity; and the display unit is used for displaying recommended candidate commodities recommended to the target user based on the ranking information of the candidate commodities.
Correspondingly, an embodiment of the present application provides an electronic device, including: a processor; the memory is used for storing a computer program, and the computer program is executed by the processor to execute the candidate commodity ordering method and the candidate commodity displaying method of the embodiment.
Correspondingly, the embodiment of the present application provides a computer storage medium, where a computer program is stored, and the computer program is run by a processor to execute the candidate product sorting method and the candidate product display method of the above embodiments.
Compared with the prior art, the embodiment of the application has the following advantages:
the embodiment of the application provides a candidate commodity sorting method, which is characterized in that before sorting information of a plurality of candidate commodities is obtained, a plurality of candidate commodities are recalled in advance for a target user, and historical effective operation data of the target user for the commodities are obtained, wherein the historical effective operation data comprise historical time information and historical space information when the target user effectively operates the commodities; and then, acquiring behavior data of the target user for the target candidate commodity based on the historical effective operation data and the attention weight of the target user for any one target candidate commodity in the plurality of candidate commodities, wherein the historical effective operation data comprises historical time information and historical space information when the target user effectively operates the commodity, and the subsequently acquired behavior data of the target user for the target candidate commodity is fused with the historical time information and the historical space information. Therefore, the ranking information of the candidate commodities, which is obtained based on the characteristic information of the candidate commodity and the behavior data of the target user fusing the historical time information and the historical space information and aiming at the candidate commodity, can embody the preference of the target user under the condition of the time information and the space information matched with the historical time information and the historical space information, and the obtained ranking information of the candidate commodities is better matched with the target user.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art according to the drawings.
Fig. 1 is a schematic view of an application scenario of the candidate commodity ranking method provided by the present application.
FIG. 1A is a schematic diagram of the zoom click attention provided in the present application.
Fig. 2 is a flowchart of a candidate commodity ranking method according to a first embodiment of the present application.
Fig. 3 is a flowchart of a candidate product display method according to a second embodiment of the present application.
Fig. 4 is a schematic diagram of a candidate product sorting apparatus according to a third embodiment of the present application.
Fig. 5 is a schematic view of a candidate merchandise display device according to a fourth embodiment of the present application.
Fig. 6 is a schematic view of an electronic device according to a fifth embodiment of the present application.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. This application is capable of implementation in many different ways than those herein set forth and of similar import by those skilled in the art without departing from the spirit of this application and is therefore not limited to the specific implementations disclosed below.
The application provides a candidate commodity ranking method, and some embodiments of the candidate commodity ranking method provided by the application can be applied to scenes of recommending some candidate commodities to target users. For example, when a user purchases goods online, some goods are displayed to the user on a home page of a user-side application for purchasing goods, and the goods displayed on the page are displayed based on pre-obtained ranking information of a plurality of candidate goods. By adopting the candidate commodity sorting method of the embodiment, sorting information of a plurality of candidate commodities can be obtained in advance. Of course, when ranking a plurality of candidate products, the ranking is performed according to the degree of correlation or matching between the candidate products and the user.
Since different users may have different shopping habits, the candidate merchandise displayed on the home page of the user-side application may be different for different users. Or, for different users, even if the candidate commodities displayed on the home page applied at the user side are the same, the displayed candidate commodities are in different orders. This is mainly because different users have different shopping habits, and thus the degree of correlation or matching between the candidate product and different users is also different.
When obtaining ranking information of a plurality of candidate products for a certain target user, time information and space information related to the candidate products are particularly important. The time information may refer to information about a certain time period of a day, and may specifically be morning, noon, afternoon, evening, or the like. It may also refer to a work day or a rest day, etc. The spatial information may refer to, for example, a distance between a location of a candidate product to be recommended currently and a location of the target user, or may also refer to whether the candidate product to be recommended can be purchased in a certain market or a marketplace under the online condition, a distance between the location of the market or the marketplace and the location of the user, and the like.
The candidate commodity ranking method is mainly used for obtaining ranking information of a plurality of candidate commodities under certain time information and space information conditions. The above-mentioned time information and space information are considered to be particularly important with respect to obtaining ranking information of a plurality of candidate products, and mainly, the correlation or matching degree between the candidate products and the target user can be represented under the condition that the time information corresponds to the space information through the time information and the space information. For example, a user may only be interested in dishes that are relatively close in proximity; or, a certain user likes to drink milk tea at the afternoon tea time, a certain user likes to go out to a store for dining on weekends, and the like. These items can be regarded as some preference information of the user with respect to time information or space information related to the product, and the obtained time information and space information are obtained based on the historical behavior data of the user. In short, the candidate commodity ranking method of the embodiment can obtain the preference degree of the target user for the candidate commodity under the specific time information and space information conditions.
In a conventional method for obtaining ranking information of a plurality of candidate commodities, the above-mentioned time information and spatial information are simply encoded and then input to a neural network model to obtain ranking information of the candidate commodities. However, the sorting method cannot well reflect the preference degree of a certain user for candidate goods under the condition of specific time information and space information.
In this embodiment, a target (target attention) attention mechanism is adopted, and time information and space information are fused in the historical behavior data of the target user, so that the preference degree of the target user for the candidate goods under the specific conditions of the time information and the space information can be more accurately reflected.
Specifically, in this embodiment, a candidate commodity ranking method may be set forth with reference to fig. 1, where fig. 1 is an application scenario diagram of the candidate commodity ranking method provided in this application.
Firstly, a request message sent by a target user for requesting to display a to-be-recommended commodity is obtained, and in order to distinguish other subsequent request messages, the request message sent by the target user for requesting to display the to-be-recommended commodity is used as a second request message. In practice, the second request message may be sent by the target user through a page on the user side. The user-side page may specifically refer to a page of a user-side application for online shopping. In a scenario of obtaining ranking information for a plurality of candidate items in a certain promotional activity, the second request message may be a request message for entering a meeting place page issued by a target user.
After obtaining the second request message, the user terminal may send the second request message to the server terminal, so that the server terminal recalls the candidate commodities based on the second request message.
Taking the example of obtaining the ranking information for a plurality of candidate commodities in a certain promotion activity as an example, after the server obtains the second request message, the server backtracks the behavior log of the target user for the commodities, where the behavior log is a recording mode of historical behavior data, such as commodity history records, paid commodity history records, collected commodity history records, browsed commodity history records, and the like that the target user has made an order.
The behavior log of the backtracked target user for the product includes historical time information and historical space information when the user performs some operation on the product. For example, when a user clicks on a certain product, the behavior log for the product includes the characteristic information of the clicked product, the historical time information and the historical space information of the clicked product.
Meanwhile, the server recalls a plurality of candidate commodities, and various modes can be adopted when the plurality of candidate commodities are recalled. For example, the target user may recall a product near the target user, or a product personalized for the target user may be recalled, and of course, a plurality of candidate products may be recalled in other manners.
After obtaining the behavior log of the target user for the commodity and recalling a plurality of candidate commodities, the server obtains historical valid operation data of the target user for the commodity based on the behavior log of the target user for the commodity, in this embodiment, the valid operation may refer to operations such as clicking, collecting, purchasing, placing an order, paying, and the like, and may be other operations, in this embodiment, the operation of deleting the target user for the commodity is taken as an invalid operation. Of course, the above cases of valid operation and invalid operation are merely applicable cases in the present embodiment.
Specifically, a plurality of systems as shown in fig. 1 are provided at the server, and each system correspondingly executes a different data processing procedure. The server comprises a commodity release system, a distributed column-oriented database (HBase) and a model online scoring system as shown in FIG. 1.
Each system performs the following data processing procedures:
after obtaining the behavior log of the target user for the commodity, the commodity release system filters historical effective operation data of the target user for the commodity, and as an example of the historical effective operation data, the historical effective operation data may be historical click data of the target user for clicking on the commodity.
After obtaining the historical effective operation data and the recalled candidate commodities, the commodity putting system obtains behavior data of the target user for the target candidate commodity according to the historical effective operation data and the attention weight of the target user for any target candidate commodity in the candidate commodities.
Specifically, the behavior data of the target user for the target candidate commodity may refer to a behavior sequence of the target user for the target candidate commodity, which includes time information and space information. The time information and the space information are matched with the historical time information and the historical space information. For example, the target user is used to shop at saturday night, and saturday night is the time information; for another example, if the target user is used to purchase a commodity with a distance of one kilometer, the distance of one kilometer is the spatial information. In this embodiment, the spatial information may refer to encoded information that is encoded in a geographic location by two or more ways, such as: the coded information may be in a GeoHash coded string format of 5 bits and 6 bits, respectively.
Since the commodity issuing system obtains a plurality of recalled candidate commodities, the commodity issuing system can also obtain feature information of any one target candidate commodity among the plurality of candidate commodities.
And then, the commodity delivery system can provide the behavior data of the target candidate commodity and the characteristic information of the target candidate commodity for the target user to a distributed column-oriented development database, on one hand, the development database can obtain sample data of a training model based on the behavior data and the characteristic information of the target candidate commodity, and provide the sample data of the training model to a deep convolutional neural network for model training, on the other hand, real-time synchronous characteristics can be provided to a model online scoring system for the model online scoring system to score according to the synchronous characteristics.
The features of real-time synchronization include: and the target user aims at the behavior data of the target candidate commodity and the characteristic information of the target candidate commodity.
The model online scoring system comprises an attention data obtaining model, wherein the attention data obtaining model is a trained model, and can obtain attention data of a target user for a target candidate commodity based on behavior data of the target user for the target candidate commodity and characteristic information of the target candidate commodity, and as an example of the attention data, the attention data obtaining model can refer to a preference score of the target user for the target candidate commodity.
When obtaining behavior data of a target user for a target candidate commodity, the attention weight of the target user for any one of a plurality of candidate commodities is used, and therefore the attention weight of the target user for any one of the plurality of candidate commodities needs to be obtained in advance.
The above manner of adopting the target attention mechanism is mainly used for obtaining the attention weight of the target user for any target candidate product in the plurality of candidate products. For example, assuming that there are three candidate items recalled in advance and the history valid operation data is history valid operation data of twenty target users for an item, it is necessary to obtain an attention weight of each of the twenty history valid operation data for the target candidate item. Then, the behavior data of the target user for the target candidate commodity is that: and carrying out weighted summation on the historical effective operation data and the corresponding weight each time to obtain data.
Finally, the behavior data of the target user for the target candidate commodity and the feature information of the target candidate commodity can be input into the attention data obtaining model, and the attention data of the target user for the target candidate commodity is obtained. Namely: a preference score for the target user for the target candidate good.
In this scenario, since the historical time information and the historical space information are fused in the behavior data of the target user for the target candidate commodity, the acquired attention data of the target user for the target candidate commodity can reflect the preference of the target user under the condition of the time information and the space information matched with the historical time information and the historical space information, and the ranking information of a plurality of candidate commodities acquired subsequently based on the attention data is better matched with the target user.
Fig. 1 introduced above is an illustration of an application scenario of the candidate commodity ranking method according to the present application, and an application scenario of the candidate commodity ranking method is not specifically limited in the embodiment of the present application, and the application scenario of the candidate commodity ranking method is only one embodiment of the application scenario of the candidate commodity ranking method provided by the present application, and the application scenario is provided to facilitate understanding of the candidate commodity ranking method provided by the present application, and is not used to limit the candidate commodity ranking method provided by the present application. Other application scenarios of the candidate commodity ordering method in the embodiment of the application are not repeated one by one.
First embodiment
A first embodiment of the present application provides a candidate product ranking method, which is described below with reference to fig. 2.
Please refer to fig. 2, which is a flowchart illustrating a candidate product sorting method according to a first embodiment of the present application.
The candidate commodity ordering method provided by the embodiment of the application comprises the following steps:
step S201: a plurality of candidate items are recalled for the target user.
The candidate commodity sorting method of the embodiment can be applied to a scene of obtaining sorting information of a plurality of candidate commodities in a certain promotion activity.
After a target user sends a request message for entering a meeting place home page through a user side page, the user side sends a request message for requesting to acquire page information to a server side, and the server side recalls a plurality of candidate commodities for the target user after obtaining the request message for requesting to acquire the page information sent by the user side.
When multiple candidate items are recalled, a variety of approaches may be taken. For example, the target user may recall a product near the target user, or a product personalized for the target user may be recalled, and of course, a plurality of candidate products may be recalled in other manners.
Step S202: and obtaining historical effective operation data of the target user for the commodity, wherein the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity.
Meanwhile, the server side obtains historical effective operation data of the target user aiming at the commodities. Specifically, as one way to obtain the historical valid operation data of the target user for the commodity, the following may be mentioned: and obtaining historical effective operation data of the target user for the commodity within the preset time and for preset times.
For example, the behavior log of the target user for the commodity within the preset number of days may be traced back, and the historical valid operation data of the target user for the commodity for the preset number of times is obtained based on the behavior log.
In this embodiment, the valid operation may refer to operations such as clicking, collecting, purchasing, placing an order, paying, and the like, and of course, other operations are also possible. Of course, the above cases of valid operation and invalid operation are merely applicable cases in the present embodiment.
As an example of the history valid operation data, the history click data of the target user clicking on the commodity may be used, and when the history click data of the target user clicking on the commodity is used as the history valid operation data, the history valid operation data includes the history time information and the history space information of the target user when the target user performs valid operation on the commodity, and the history time information and the history space information correspond to each other when the target user clicks on the commodity.
Step S203: and acquiring behavior data of the target user aiming at the target candidate commodity according to the historical effective operation data and the attention weight of the target user aiming at any one target candidate commodity in the plurality of candidate commodities.
The attention weight of the target candidate commodity is obtained from the history valid operation data and the feature information of the target candidate commodity.
Before obtaining the behavior data, it is necessary to obtain in advance an attention weight of the target user with respect to any one target candidate product among the plurality of candidate products.
As a way of obtaining the attention weight of the target user for any one target candidate product in the plurality of candidate products, the multiple times of history valid operation data which are preset times in the history valid operation data may refer to: and obtaining the attention weight of the target candidate commodity for each time of the historical effective operation data.
Specifically, in obtaining the attention weight of the target candidate product for each time of the history valid operation data, the following steps need to be performed:
first, each vector of history valid operation data for vector representation of each time of history valid operation data of a preset number of times is obtained.
Then, a feature information vector of the target candidate commodity for vector representation of the feature information of the target candidate commodity is obtained.
And finally, obtaining the attention weight of the target candidate commodity aiming at the historical effective operation data of each time according to each historical effective operation data vector and the characteristic information vector of the target candidate commodity.
More specifically, obtaining the attention weight of the target candidate product for each time of the historical valid operation data according to each historical valid operation data vector and the feature information vector of the target candidate product respectively may be: and respectively carrying out attention operation on each history effective operation data vector and the characteristic information vector of the target candidate commodity to obtain the attention weight of the target candidate commodity aiming at each history effective operation data.
Further, performing attention operation on each history effective operation data vector and the feature information vector of the target candidate commodity by adopting the following formula:
Figure DEST_PATH_IMAGE001
wherein,Qfor a history of valid operation data vectors,K、Vis the feature information vector of the target candidate commodity,d k is composed ofKThe dimensions of the vector.
The above attention operation is actually performed based on a target attention mechanism, that is: the above formula is a formula based on a target attention mechanism. The target (target attention) attention mechanism is used for learning the correlation between the target user behavior sequence and the target candidate commodity and constructing an expression vector of the target user behavior sequence. Considering the physical meaning of the vector, the inner product is used here to calculate the Attention in the above formula, so that the more similar the two items are, the larger the inner product, the larger the Attention gain. The Attention weight is calculated by means of inner Product using Scaled Dot-Product Attention. Fig. 1A is a schematic diagram illustrating the principle of zooming the click attention.
As can be seen in FIG. 1A, the principle of calculating attention weight by inner product for scaled click attention is as follows: firstly, the method is carried outQ(linear vector) andK(Linear vector) carrying out Scaled Dot-Product operation, then carrying out SoftMax operation, and carrying out the operation result after the SoftMax operation and the linear vectorVMatMul operation is carried out, and finally attention weight is calculated through Concat operation.
To more clearly understand how to calculate the attention weight of the target candidate product for each time of the historical valid operation data based on the above formula, a more specific example is given.
Assuming that there are three candidate products recalled in advance and the historically valid operation data is the historically valid operation data of the target user for the product twenty times, it is necessary to obtain the attention weight of each of the historically valid operation data for the target candidate product twenty times.
Assuming that the target user is user a, the recalled candidate commodities are candidate commodity 1, candidate commodity 2 and candidate commodity 3; for the user a, it is necessary to obtain the attention weight of each of the twenty times of the historically valid operation data for the candidate commodity 1, the attention weight of each of the twenty times of the historically valid operation data for the candidate commodity 2, and the attention weight of each of the twenty times of the historically valid operation data for the candidate commodity 3, respectively.
When the attention weight of each time of the historical valid operation data for the candidate commodity 1 in the twenty times of historical valid operation data is respectively obtained, the attention weight of the first time of the historical valid operation data for the candidate commodity 1 can be respectively obtained
Figure DEST_PATH_IMAGE003
Attention weighting of the second time of the historically valid operation data for candidate item 1
Figure 706860DEST_PATH_IMAGE004
Attention weight of third time history valid operation data for candidate commodity 1
Figure DEST_PATH_IMAGE005
And so on until obtaining the attention weight of the twentieth historical valid operation data for the candidate product 1
Figure 687323DEST_PATH_IMAGE006
。
Attention weight of candidate commodity 1 with the obtained first-time history valid operation data
Figure 743004DEST_PATH_IMAGE003
For example, first, a first vector of historically valid operation data for vector representation of first historically valid operation data is obtainedQ 1. At the same time, a feature information vector of the candidate product 1 for vector representation of the feature information of the candidate product 1 is obtainedK 1 、V 1 WhereinK 1 、V 1 are used for carrying out vector representation on the characteristic information of the candidate commodity 1, but the expression forms of the characteristic information and the candidate commodity are different,K 1 andQ 1the dimensions of the vectors are the same as each other,V 1 the vector dimension may be equal toK 1 AndQ 1different.d k Is composed ofKDimension of vector toQ 1、K 1 AndV 1 substituting the formula, the formula can be used for calculation
Figure 575962DEST_PATH_IMAGE003
。
In a similar manner as described above, willQ 1Is replaced byQ 2Can be obtained by calculation
Figure 109711DEST_PATH_IMAGE004
. By analogy, the calculation can be obtained
Figure 490488DEST_PATH_IMAGE006
Thus, attention weights of each of the twenty times of the historical valid operation data with respect to the candidate product 1 are obtained, respectively.
The first vector of historically valid operation data for vector representation of the first historically valid operation dataQ 1Obtained by the following method:
the historical effective operation data also comprises characteristic information of the operated commodity aimed at by the effective operation of the target user; for example, the user a may click on the product M in the first-time history valid operation data, and the product M is the operated product.
Thus, the above-described obtaining of the first vector of history valid operation data for vector representation of the first history valid operation dataQ 1Can mean that: firstly, aiming at first-time historical valid operation data, obtaining a characteristic information vector of a commodity M for vector representation of characteristic information of the commodity M in the first-time historical valid operation data; meanwhile, obtaining a first historical time information vector used for carrying out vector representation on historical time information in the first-time historical effective operation data and obtaining a first historical space information vector used for carrying out vector representation on historical space information in the first-time historical effective operation data; then, based on the feature information vector, the first historical time information vector and the first historical space information vector of the commodity M, a first historical valid operation data vector for performing vector representation on the first historical valid operation data is obtainedQ 1. For example, the feature information vector of the article M, the first historical time information vector, and the first historical space information vector may be added to obtain a first historical valid operation data vector for vector representation of the first historical valid operation dataQ 1。
According to the acquisition for the firstFirst vector of historically valid operation data for vector representation of sub-historically valid operation dataQ 1In the same manner, a second vector of historically valid operation data for vector representation of the second historically valid operation data may be obtainedQ 2By analogy, a twentieth vector of history valid operation data for vector representation of the twentieth vector of history valid operation data may be obtainedQ 20。
Assuming that the vector dimensions of the twenty history valid operation data vectors are all 32, the twenty history valid operation data vectors are vector sequences with a length of 20 and a node vector dimension of 32.
After obtaining each vector of the history valid operation data vector representing each time of the history valid operation data and the attention weight of the target candidate product for each time of the history valid operation data, the behavior data of the target user for the target candidate product may be obtained as the attention weight of the target user for any one target candidate product in the plurality of candidate products according to the history valid operation data and the attention weight of the target user for the target candidate product, and may be: and performing corresponding weighted summation operation on each historical effective operation data vector and the attention weight of the target candidate commodity aiming at the historical effective operation data of each time to obtain a behavior data vector for performing vector representation on the behavior data of the target user aiming at the target candidate commodity.
For example, for the candidate item 1, the behavior data of the user a for the candidate item 1 may refer to a behavior data vector for vector-representing the behavior data of the user a for the candidate item 1Q 1Wherein
Figure DEST_PATH_IMAGE007
(ii) a Similarly, the behavior data of the user a for the candidate commodity 2 may refer to a behavior data vector for vector-representing the behavior data of the user a for the candidate commodity 2Q 2Wherein
Figure 982649DEST_PATH_IMAGE008
(ii) a User A targets candidatesThe behavior data of the commodity 3 may refer to a behavior data vector for vector-representing the behavior data of the user a with respect to the candidate commodity 3Q 3Wherein
Figure DEST_PATH_IMAGE009
。
step S204: and obtaining the ranking information of the candidate commodities according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity.
Specifically, step S204 can be further divided into step S204-1 and step S204-2.
Specifically, step S204-1: and acquiring the attention data of the target user for the target candidate commodity according to the behavior data of the target user for the target candidate commodity and the characteristic information of the target candidate commodity.
After obtaining the behavior data vector, as a way of obtaining the attention data of the target user for the target candidate commodity according to the behavior data of the target user for the target candidate commodity and the feature information of the target candidate commodity: and taking the behavior data vector and the characteristic information vector of the target candidate commodity as input data of an attention data acquisition model to acquire the attention data of the target user for the target candidate commodity, wherein the attention data acquisition model is used for acquiring the attention data of the user for the candidate commodity according to the behavior data vector of the user for the candidate commodity and the characteristic information vector of the candidate commodity.
For example, a behavior data vector of the user A for the candidate commodity 1 is obtainedQ 1Then, willQ 1And inputting the attention data acquisition model together with the characteristic information vector of the candidate commodity 1, and further acquiring the attention data of the user A on the candidate commodity 1. If it is determined in the history valid operation data of the user a that the user a is used to shop on saturday at a store within one kilometer from home, the obtained attention data of the user a on the candidate commodity 1 refers to the attention data of the user a on the candidate commodity 1 under the time condition of saturday and under the space condition within one kilometer from home, and the attention data may refer to the preferenceThe degree is embodied as a preference score.
More specifically, taking the behavior data vector and the feature information vector of the target candidate commodity as input data of the attention data obtaining model to obtain the attention data of the target user for the target candidate commodity may refer to: splicing the behavior data vector with the characteristic information vector of the target candidate commodity to obtain a spliced vector; and taking the spliced vector as the input data of the attention data acquisition model to acquire the attention data of the target user for the target candidate commodity.
Of course, in the same manner, the attention data of the user a on the candidate product 2 may be obtained, and similarly, the attention data of the user a on the candidate product 2 refers to the attention data of the user a on the candidate product 2 under the time condition of saturday and under the space condition within one kilometer from home.
In the present embodiment, the attention data obtaining model is obtained as follows:
firstly, obtaining historical operation sample data of a user for a commodity, an attention weight sample of the user for a candidate commodity, a feature information sample of the candidate commodity and an attention sample data of the user for the candidate commodity; the historical operation sample data, the attention weight sample of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity correspond to each other.
And then training the deep convolutional neural network based on the historical operation sample data, the attention weight sample of the user for the candidate commodity, the characteristic information sample of the candidate commodity and the attention sample data of the user for the candidate commodity to obtain an attention data obtaining model.
Specifically, the training of the deep convolutional neural network based on the historical operation sample data, the attention weight sample of the user for the candidate product, the feature information sample of the candidate product, and the attention sample data of the user for the candidate product may refer to: firstly, acquiring behavior sample data of a user for a candidate commodity according to historical operation sample data and an attention weight sample of the user for the candidate commodity; and then training the deep convolutional neural network based on the behavior sample data of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity.
Further, training the deep convolutional neural network based on the behavior sample data of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity refers to: firstly, acquiring a behavior sample data vector for vector representation of behavior sample data of a candidate commodity by a user; then, obtaining a characteristic information sample vector of the candidate commodity for carrying out vector representation on the characteristic information sample of the candidate commodity; and finally, training the deep convolutional neural network based on the behavior sample data vector, the characteristic information sample vector of the candidate commodity and the attention sample data of the user aiming at the candidate commodity.
In this embodiment, the historical operation sample data includes historical valid operation sample data and historical invalid operation sample data; when the deep convolutional neural network is trained, the historical effective operation sample data is used as a positive sample of the training data, and the historical invalid operation sample data is used as a negative sample of the training data.
For example, if a user is exposed to a certain product and clicks on the product when the valid operation is regarded as a click operation, the click operation of the user on the product may be a history valid operation. If a certain product is exposed to the user and the user does not click on the product, the un-clicked operation by the user on the product may be a history invalidation operation.
Step S204-2: and obtaining ranking information of the candidate commodities according to the attention data of the target user aiming at each candidate commodity in the candidate commodities.
After obtaining the attention data of the target user for each of the plurality of candidate items, ranking information of the plurality of candidate items may be obtained. For example, after obtaining the attention data of the user a for each of the candidate item 1, the candidate item 2, and the candidate item 3, the candidate item 1, the candidate item 2, and the candidate item 3 may be ranked, and ranking information of the candidate item 1, the candidate item 2, and the candidate item 3 may be obtained. Assuming that the value of the attention data of the user a for the candidate item 1 is the largest, the candidate item 1 is ranked first.
The server side sends the ranking information of the candidate commodities to the client side after obtaining the ranking information of the candidate commodities, and can obtain a first request message sent by the client side for requesting to acquire the page information in advance before sending the ranking information of the candidate commodities to the client side; and sending the sequencing information of the candidate commodities to the user terminal aiming at the first request message.
In this embodiment, before obtaining ranking information of a plurality of candidate commodities, the method recalls a plurality of candidate commodities for a target user in advance, and obtains historical effective operation data of the target user for the commodities, wherein the historical effective operation data includes historical time information and historical space information when the target user effectively operates the commodities; and then, acquiring behavior data of the target user for the target candidate commodity based on the historical effective operation data and the attention weight of the target user for any one target candidate commodity in the plurality of candidate commodities, wherein the historical effective operation data comprises historical time information and historical space information when the target user effectively operates the commodity, and the subsequently acquired behavior data of the target user for the target candidate commodity is fused with the historical time information and the historical space information. Therefore, the ranking information of the candidate commodities, which is obtained based on the characteristic information of the candidate commodity and the behavior data of the target user fusing the historical time information and the historical space information and aiming at the candidate commodity, can embody the preference of the target user under the condition of the time information and the space information matched with the historical time information and the historical space information, and the obtained ranking information of the candidate commodities is better matched with the target user.
Second embodiment
Corresponding to the first embodiment, a second embodiment of the present application provides a method for displaying a candidate commodity. The main execution body of this embodiment is the user side, and the same parts of the second embodiment as those of the first embodiment will not be described again, specifically refer to the relevant parts of the first embodiment.
Please refer to fig. 3, which is a flowchart illustrating a candidate product displaying method according to a second embodiment of the present application.
The candidate commodity display method comprises the following steps:
step S301: and obtaining the ranking information of the candidate commodities sent by the server.
In this embodiment, the ranking information of the plurality of candidate commodities is obtained according to the behavior data of the target user for the target candidate commodity and the feature information of the target candidate commodity, and the target candidate commodity is any one of the plurality of candidate commodities.
In this embodiment, the behavior data of the target user for the target candidate commodity is obtained according to the history valid operation data of the target user for the commodity and the attention weight of the target user for any one target candidate commodity in the plurality of candidate commodities, the attention weight of the target candidate commodity is obtained according to the history valid operation data and the feature information of the target candidate commodity, and the history valid operation data includes the history time information and the history space information when the target user performs valid operation for the commodity.
Step S302: and displaying recommended candidate commodities recommended to the target user based on the ranking information of the candidate commodities.
In this embodiment, in order to obtain the sorting information of the multiple candidate commodities, the user terminal further sends a first request message for requesting to obtain the page information to the server terminal.
As an example of obtaining the ranking information of the multiple candidate commodities sent by the server, the following may be mentioned: and obtaining the ranking information of the candidate commodities sent by the server aiming at the first request message.
In this embodiment, presenting, as ranking information based on a plurality of candidate products, recommended candidate products for recommendation to a target user may refer to: firstly, determining recommended candidate commodities for recommending to a target user based on the ranking information of a plurality of candidate commodities; and then, displaying the determined recommended candidate commodity recommended to the target user on a page of the user side.
After the ranking information of the candidate commodities is obtained, if the number of the candidate commodities is large, there may be a case where it is not necessary to display all of the candidate commodities on the page of the user side, and for this case, a part of the candidate commodities may be preferentially selected for display. For example, the user side obtains the ranking information of one hundred candidate commodities, and only ten candidate commodities can be displayed on the page of the user side, so that the top ten candidate commodities can be used as recommended candidate commodities recommended to the target user.
Of course, in this embodiment, all the candidate products may be used as recommendation candidate products recommended to the target user.
In the embodiment, the user end also obtains a second request message which is sent by the target user through a page of the user end and used for requesting to display the to-be-recommended commodity; and displaying the recommended candidate commodity recommended to the target user on the page based on the second request message. In a scenario of obtaining ranking information for a plurality of candidate items in a certain promotional activity, the second request message may be a request message for entering a meeting place page issued by a target user.
In this embodiment, since the ranking information of the candidate goods presented to the target user is obtained based on the attention data of the target user for the target candidate goods, and the attention data of the target user for the target candidate goods is obtained based on the feature information of the target candidate goods and the behavior data of the target user for the target candidate goods, which combines the historical time information and the historical space information, the attention data of the target user for the target candidate goods can embody the preference of the target user under the condition of the time information and the space information which are matched with the historical time information and the historical space information, and the ranking information of a plurality of candidate goods obtained based on the attention data is more matched with the target user.
Third embodiment
Corresponding to the candidate commodity ranking method provided in the first embodiment of the present application, a third embodiment of the present application also provides a candidate commodity ranking device. Since the device embodiment is substantially similar to the first embodiment, it is relatively simple to describe, and reference may be made to some descriptions of the first embodiment for relevant points. The device embodiments described below are merely illustrative.
Please refer to fig. 4, which is a schematic diagram of a candidate product sorting apparatus according to a third embodiment of the present application.
The candidate commodity ranking device comprises: a recall unit 401 that recalls a plurality of candidate commodities for a target user; a history valid operation data obtaining unit 402, configured to obtain history valid operation data for a commodity of a target user, where the history valid operation data includes history time information and history space information when the target user performs a valid operation for the commodity; a behavior data obtaining unit 403, configured to obtain behavior data of the target user for any target candidate product in the multiple candidate products according to the historical valid operation data and the attention weight of the target user for the target candidate product; a ranking information obtaining unit 404, configured to obtain ranking information of the multiple candidate commodities according to the behavior data of the target user for the target candidate commodity and the feature information of the target candidate commodity.
Optionally, the history valid operation data obtaining unit is specifically configured to: and obtaining historical effective operation data of the target user for the commodity within the preset time and for preset times.
Optionally, the method further includes: the attention weight obtaining unit is specifically configured to: and obtaining the attention weight of the target candidate commodity according to the historical effective operation data and the characteristic information of the target candidate commodity.
Optionally, the attention weight obtaining unit is specifically configured to: obtaining each historical effective operation data vector used for carrying out vector representation on each time of historical effective operation data in the historical effective operation data of preset times; obtaining a characteristic information vector of the target candidate commodity for vector representation of the characteristic information of the target candidate commodity; and obtaining the attention weight of the target candidate commodity aiming at the historical effective operation data of each time according to each historical effective operation data vector and the characteristic information vector of the target candidate commodity.
Optionally, the history valid operation data further includes: the target user effectively operates the characteristic information of the operated commodity; the attention weight obtaining unit is specifically configured to: aiming at the historical effective operation data of each time, obtaining a characteristic information vector of the operated commodity, wherein the characteristic information vector is used for carrying out vector representation on the characteristic information of the operated commodity in the historical effective operation data of each time; obtaining a historical time information vector used for carrying out vector representation on the historical time information in the historical effective operation data each time; obtaining a historical spatial information vector used for carrying out vector representation on the historical spatial information in the historical effective operation data each time; and obtaining each history effective operation data vector used for carrying out vector representation on each history effective operation data in the history effective operation data of preset times based on the characteristic information vector, the history time information vector and the history space information vector of the operated commodity.
Optionally, the attention weight obtaining unit is specifically configured to: and performing attention operation on each historical effective operation data vector and the characteristic information vector of the target candidate commodity respectively to obtain the attention weight of the target candidate commodity aiming at each historical effective operation data.
Optionally, the attention weight obtaining unit is specifically configured to: performing attention operation on each historical effective operation data vector and the characteristic information vector of the target candidate commodity by adopting the following formula:
Figure 958695DEST_PATH_IMAGE001
wherein,Qfor a history of valid operation data vectors,K、Vis a target candidate quotientThe feature information vector of the product is obtained,d k is composed ofKThe dimensions of the vector.
Optionally, the behavior data obtaining unit is specifically configured to: and performing corresponding weighted summation operation on each historical effective operation data vector and the attention weight of the target candidate commodity aiming at the historical effective operation data of each time to obtain a behavior data vector for performing vector representation on the behavior data of the target user aiming at the target candidate commodity.
Optionally, the sorting information obtaining unit is specifically configured to: acquiring attention data of the target user for the target candidate commodity according to the behavior data of the target user for the target candidate commodity and the characteristic information of the target candidate commodity; obtaining ranking information of the candidate commodities according to the attention data of the target user for each candidate commodity in the candidate commodities.
Optionally, the sorting information obtaining unit is specifically configured to: and obtaining the attention data of the target user for the target candidate commodity by taking the behavior data vector and the feature information vector of the target candidate commodity as input data of an attention data obtaining model, wherein the attention data obtaining model is used for obtaining the attention data of the user for the candidate commodity according to the behavior data vector of the user for the candidate commodity and the feature information vector of the candidate commodity.
Optionally, the sorting information obtaining unit is specifically configured to: splicing the behavior data vector with the characteristic information vector of the target candidate commodity to obtain a spliced vector; and taking the spliced vector as input data of an attention data acquisition model to acquire the attention data of the target user for the target candidate commodity.
Optionally, the method further includes: a training unit, specifically configured to: the attention data acquisition model is obtained as follows: obtaining historical operation sample data of a user for a commodity, an attention weight sample of the user for a candidate commodity, a feature information sample of the candidate commodity and an attention sample data of the user for the candidate commodity; wherein the historical operation sample data, the attention weight sample of the user for the candidate commodity, the characteristic information sample of the candidate commodity and the attention sample data of the user for the candidate commodity correspond to each other; training a deep convolutional neural network based on the historical operation sample data, the attention weight sample of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity to obtain the attention data obtaining model.
Optionally, the training unit is specifically configured to: acquiring behavior sample data of the user for the candidate commodity according to the historical operation sample data and the attention weight sample of the user for the candidate commodity; training the deep convolutional neural network based on the behavior sample data of the user for the candidate commodity, the feature information sample of the candidate commodity and the attention sample data of the user for the candidate commodity.
Optionally, the training unit is specifically configured to: obtaining a behavior sample data vector used for performing vector representation on the behavior sample data of the user aiming at the candidate commodity; obtaining a characteristic information sample vector of the candidate commodity for carrying out vector representation on the characteristic information sample of the candidate commodity; training the deep convolutional neural network based on the behavior sample data vector, the characteristic information sample vector of the candidate commodity and the attention sample data of the user for the candidate commodity.
Optionally, the historical operation sample data includes historical valid operation sample data and historical invalid operation sample data; when the deep convolutional neural network is trained, the historical effective operation sample data is used as a positive sample of training data, and the historical ineffective operation sample data is used as a negative sample of the training data.
Optionally, the method further includes: the sequencing information sending unit is specifically configured to: acquiring a first request message sent by a user side and used for requesting to acquire page information; and sending the sequencing information of the candidate commodities to the user terminal aiming at the first request message.
Fourth embodiment
Corresponding to the candidate commodity display method provided by the second embodiment of the present application, a fourth embodiment of the present application further provides a candidate commodity display device. Since the apparatus embodiment is substantially similar to the second embodiment, it is relatively simple to describe, and reference may be made to some descriptions of the second embodiment for relevant points. The device embodiments described below are merely illustrative.
Please refer to fig. 5, which is a schematic diagram of a candidate product display apparatus according to a fourth embodiment of the present application.
The candidate commodity display device includes: a ranking information obtaining unit 501, configured to obtain ranking information of multiple candidate commodities sent by a server; the ranking information of the candidate commodities is obtained according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity; the behavior data of the target user for the target candidate commodity is obtained according to historical effective operation data of the target user for the commodity and attention weight of the target user for any one target candidate commodity in the candidate commodities, and the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity; a display unit 502, configured to display recommended candidate commodities for recommendation to a target user based on the ranking information of the multiple candidate commodities.
Optionally, the target candidate commodity is any one of the candidate commodities; the attention weight of the target candidate commodity is obtained according to the historical valid operation data and the characteristic information of the target candidate commodity.
Optionally, the method further includes: a first request message sending unit, specifically configured to: sending a first request message for requesting to acquire page information to a server; the sorting information obtaining unit is specifically configured to: and obtaining the ranking information of the candidate commodities sent by the server aiming at the first request message.
Optionally, the display unit is specifically configured to: determining recommended candidate commodities for recommendation to a target user based on the ranking information of the candidate commodities; and displaying the determined recommended candidate commodity recommended to the target user on a page of the user side.
Optionally, the method further includes: the second request message obtaining unit is specifically configured to: acquiring a second request message which is sent by the target user through a page of a user side and used for requesting to display the to-be-recommended commodity; the display unit is specifically configured to: and displaying recommended candidate commodities recommended to the target user on the page based on the second request message.
Fifth embodiment
Corresponding to the methods of the first to second embodiments of the present application, a fifth embodiment of the present application further provides an electronic device.
As shown in fig. 6, fig. 6 is a schematic view of an electronic device provided in a fifth embodiment of the present application. The electronic device includes: a processor 601; a memory 602 for storing a computer program to be executed by the processor for performing the methods of the first to second embodiments.
Sixth embodiment
In correspondence with the methods of the first to second embodiments of the present application, a sixth embodiment of the present application also provides a computer storage medium storing a computer program that is executed by a processor to perform the methods of the first to second embodiments.
Although the present invention has been described with reference to the preferred embodiments, it should be understood that the scope of the present invention is not limited to the embodiments described above, and that various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the present invention.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory. The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
1. Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include non-transitory computer-readable storage media (non-transitory computer readable storage media), such as modulated data signals and carrier waves.
2. As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

Claims (9)

1. A method for ranking candidate items, comprising:
recalling a plurality of candidate items for the target user;
obtaining historical valid operation data of a target user for a commodity, comprising: obtaining historical effective operation data of a target user for commodities within preset time and for preset times; the historical effective operation data comprises historical time information and historical space information when the target user effectively operates the commodity;
obtaining behavior data of the target user for any target candidate commodity in the plurality of candidate commodities according to the historical valid operation data and the attention weight of the target user for the target candidate commodity, wherein the behavior data comprises: performing corresponding weighted summation operation on each historical effective operation data vector used for performing vector representation on each historical effective operation data in the historical effective operation data of preset times and the attention weight of the target candidate commodity aiming at each historical effective operation data to obtain behavior data of the target user aiming at the target candidate commodity; the attention weight of the target candidate commodity for each time of the historical effective operation data is respectively obtained according to each historical effective operation data vector and a characteristic information vector of the target candidate commodity for vector representation of the characteristic information of the target candidate commodity;
and obtaining the ranking information of the candidate commodities according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity.
2. The method of claim 1, further comprising: and obtaining the attention weight of the target candidate commodity according to the historical effective operation data and the characteristic information of the target candidate commodity.
3. The method of claim 2, wherein the obtaining the attention weight of the target candidate good according to the historical valid operation data and the feature information of the target candidate good comprises:
obtaining each historical effective operation data vector used for carrying out vector representation on each time of historical effective operation data in the historical effective operation data of preset times;
obtaining a characteristic information vector of the target candidate commodity for vector representation of the characteristic information of the target candidate commodity;
and obtaining the attention weight of the target candidate commodity aiming at the historical effective operation data of each time according to each historical effective operation data vector and the characteristic information vector of the target candidate commodity.
4. The method of claim 3, wherein the historical valid operation data further comprises: the target user effectively operates the characteristic information of the operated commodity;
the obtaining of each history valid operation data vector for performing vector representation on each history valid operation data in the history valid operation data of the preset times comprises:
aiming at the historical effective operation data of each time, obtaining a characteristic information vector of the operated commodity, wherein the characteristic information vector is used for carrying out vector representation on the characteristic information of the operated commodity in the historical effective operation data of each time;
obtaining a historical time information vector used for carrying out vector representation on the historical time information in the historical effective operation data each time;
obtaining a historical spatial information vector used for carrying out vector representation on the historical spatial information in the historical effective operation data each time;
and obtaining each history effective operation data vector for performing vector representation on each history effective operation data in the history effective operation data of preset times based on the characteristic information vector, the history time information vector and the history space information vector of the operated commodity.
5. A candidate commodity display method is characterized by comprising the following steps:
obtaining the sequencing information of a plurality of candidate commodities sent by a server; the ranking information of the candidate commodities is obtained according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity; the behavior data of the target user for the target candidate commodity is obtained by performing corresponding weighted summation operation on each historical effective operation data vector which is used for performing vector representation on each historical effective operation data in the historical effective operation data of the target user for the commodity for preset times and the attention weight of any one target candidate commodity in the plurality of candidate commodities for each historical effective operation data, wherein the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity; the attention weight of the target candidate commodity for each time of the historical effective operation data is respectively obtained according to each historical effective operation data vector and a characteristic information vector of the target candidate commodity for vector representation of the characteristic information of the target candidate commodity;
and displaying recommended candidate commodities recommended to the target user based on the ranking information of the candidate commodities.
6. A candidate commodity ranking device, comprising:
a recall unit that recalls a plurality of candidate commodities for a target user;
a history valid operation data obtaining unit for obtaining history valid operation data of a target user for a commodity, comprising: obtaining historical effective operation data of a target user for commodities within preset time and for preset times; the historical effective operation data comprises historical time information and historical space information when the target user effectively operates the commodity;
a behavior data obtaining unit, configured to obtain behavior data of the target user for any target candidate product in the plurality of candidate products according to the historical valid operation data and the attention weight of the target user for the target candidate product, where the behavior data obtaining unit includes: performing corresponding weighted summation operation on each historical effective operation data vector used for performing vector representation on each historical effective operation data in the historical effective operation data of preset times and the attention weight of the target candidate commodity aiming at each historical effective operation data to obtain behavior data of the target user aiming at the target candidate commodity; the attention weight of the target candidate commodity for each time of the historical effective operation data is respectively obtained according to each historical effective operation data vector and a characteristic information vector of the target candidate commodity for vector representation of the characteristic information of the target candidate commodity;
and the ranking information obtaining unit is used for obtaining ranking information of the candidate commodities according to the behavior data of the target user for the target candidate commodity and the characteristic information of the target candidate commodity.
7. A candidate merchandise display device, comprising:
the system comprises a ranking information obtaining unit, a ranking information obtaining unit and a ranking information obtaining unit, wherein the ranking information obtaining unit is used for obtaining ranking information of a plurality of candidate commodities sent by a server; the ranking information of the candidate commodities is obtained according to the behavior data of the target user aiming at the target candidate commodity and the characteristic information of the target candidate commodity; the behavior data of the target user for the target candidate commodity is obtained by performing corresponding weighted summation operation on each historical effective operation data vector which is used for performing vector representation on each historical effective operation data in the historical effective operation data of the target user for the commodity for preset times and the attention weight of any one target candidate commodity in the plurality of candidate commodities for each historical effective operation data, wherein the historical effective operation data comprises historical time information and historical space information when the target user performs effective operation on the commodity; the attention weight of the target candidate commodity aiming at the historical effective operation data of each time is respectively obtained according to each historical effective operation data vector and a characteristic information vector of the target candidate commodity for carrying out vector representation on the characteristic information of the target candidate commodity;
and the display unit is used for displaying recommended candidate commodities recommended to the target user based on the ranking information of the candidate commodities.
8. An electronic device, comprising:
a processor;
a memory for storing a computer program for execution by the processor to perform the method of any one of claims 1 to 5.
9. A computer storage medium, characterized in that it stores a computer program that is executed by a processor to perform the method of any one of claims 1-5.
CN202111577701.8A 2021-12-22 2021-12-22 Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment Active CN113962753B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111577701.8A CN113962753B (en) 2021-12-22 2021-12-22 Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111577701.8A CN113962753B (en) 2021-12-22 2021-12-22 Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment

Publications (2)

Publication Number Publication Date
CN113962753A CN113962753A (en) 2022-01-21
CN113962753B true CN113962753B (en) 2022-05-13

Family

ID=79473563

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111577701.8A Active CN113962753B (en) 2021-12-22 2021-12-22 Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment

Country Status (1)

Country Link
CN (1) CN113962753B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10448120B1 (en) * 2016-07-29 2019-10-15 EMC IP Holding Company LLC Recommending features for content planning based on advertiser polling and historical audience measurements
CN113139850A (en) * 2021-04-26 2021-07-20 西安电子科技大学 Commodity recommendation model for relieving data sparsity and commodity cold start
CN113360816A (en) * 2020-03-05 2021-09-07 北京沃东天骏信息技术有限公司 Click rate prediction method and device

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108335137B (en) * 2018-01-31 2021-07-30 北京三快在线科技有限公司 Sorting method and device, electronic equipment and computer readable medium
CN110458649A (en) * 2019-07-11 2019-11-15 北京三快在线科技有限公司 Information recommendation method, device, electronic equipment and readable storage medium storing program for executing
CN112417207B (en) * 2020-11-24 2023-02-21 未来电视有限公司 Video recommendation method, device, equipment and storage medium

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10448120B1 (en) * 2016-07-29 2019-10-15 EMC IP Holding Company LLC Recommending features for content planning based on advertiser polling and historical audience measurements
CN113360816A (en) * 2020-03-05 2021-09-07 北京沃东天骏信息技术有限公司 Click rate prediction method and device
CN113139850A (en) * 2021-04-26 2021-07-20 西安电子科技大学 Commodity recommendation model for relieving data sparsity and commodity cold start

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
An Attention-Based Recommendation Algorithm;Yan Chu;《2019 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Big Data & Cloud Computing, Sustainable Computing & Communications, Social Computing & Networking (ISPA/BDCloud/SocialCom/SustainCom)》;20200326;第1505-1510页 *
融合时空感知GRU和注意力的下一个地点推荐;李全 等;《计算机应用》;20200310;第40卷(第3期);第678-682页 *

Also Published As

Publication number Publication date
CN113962753A (en) 2022-01-21

Similar Documents

Publication Publication Date Title
JP6725549B2 (en) Save communication session state
US8909568B1 (en) Predictive analytic modeling platform
KR100961783B1 (en) Apparatus and method for presenting personalized goods and vendors based on artificial intelligence, and recording medium thereof
CN110298716B (en) Information pushing method and device
US20150310388A1 (en) Local couriers network in the context of an on-line trading platform
CN106779940B (en) Method and device for confirming display commodity
TW201804400A (en) Data object pushing method, device and system
US20140156394A1 (en) Targeted Advertisement Generation For Travelers
CN111144874A (en) Payment mode recommendation method, device and equipment
WO2020221022A1 (en) Service object recommendation method
US10445813B2 (en) Tap walls that recommend products based on in-store and out-of-store interactions of anonymous users
CN110827101B (en) Shop recommending method and device
CN114168843A (en) Search word recommendation method, device and storage medium
CN111680213A (en) Information recommendation method, data processing method and device
CN113837843A (en) Product recommendation method, device, medium and electronic equipment
US10096045B2 (en) Tying objective ratings to online items
CN113962753B (en) Candidate commodity ordering method, candidate commodity display method, candidate commodity ordering device, candidate commodity display device and electronic equipment
CN110796505B (en) Business object recommendation method and device
CN112632389B (en) Information processing method, information processing apparatus, storage medium, and electronic device
CN113158032B (en) Information pushing method and device
CN111652674B (en) Resource recommendation method and device
CN111369315A (en) Resource object recommendation method and device, and data prediction model training method and device
CN113052664B (en) Service information recall method and device, electronic equipment and storage medium
CN116452303B (en) Electronic commerce data management method based on big data
CN113420214B (en) Electronic transaction object recommendation method, device and equipment

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