CN110909176A

CN110909176A - Data recommendation method and device, computer equipment and storage medium

Info

Publication number: CN110909176A
Application number: CN201911141806.1A
Authority: CN
Inventors: 胡乐
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2019-11-20
Filing date: 2019-11-20
Publication date: 2020-03-24
Anticipated expiration: 2039-11-20
Also published as: CN110909176B

Abstract

The embodiment of the application provides a data recommendation method, a data recommendation device, computer equipment and a storage medium, wherein the method comprises the following steps: acquiring a first service object set and acquiring a knowledge graph; the first service object set comprises basic description information and historical behavior parameters corresponding to service objects, and the knowledge graph comprises an association relation between detail description information corresponding to each historical service object in an object library; acquiring detail description information which has an incidence relation with the basic description information from the knowledge graph, taking the detail description information as target detail description information corresponding to the business object, and predicting pre-estimated behavior parameters corresponding to the business object based on the target detail description information; and selecting a target business object for recommendation from the first business object set according to the historical behavior parameters and the estimated behavior parameters corresponding to the business objects. By adopting the embodiment of the application, the accuracy of data recommendation can be improved.

Description

Data recommendation method and device, computer equipment and storage medium

Technical Field

The present application relates to the field of internet technologies, and in particular, to a data recommendation method and apparatus, a computer device, and a storage medium.

Background

Dynamic Product Ads (DPA) refers to a personalized advertisement for each user, which is established based on user preferences and characteristics, and can achieve the effect of thousands of people. For all commodities, the recommended commodities for the user can be accurately determined based on the behavior data of the user (namely the behaviors of clicking, purchasing, browsing and the like of the user) and the commodity information (namely the description information, displaying and clicking data of the commodities), and then the dynamic commodity advertisement is established.

However, for a new commodity, in the process of recommending to a user, since the new commodity does not display and click data, the commodity information is too simple to recommend to the user only by means of random sampling. Therefore, the whole recommendation process lacks a commodity data reference basis, the interest degree of the user on the commodity cannot be controlled, and the accuracy of the commodity recommended to the user is too low.

Disclosure of Invention

The embodiment of the application provides a data recommendation method and device, computer equipment and a storage medium, which can improve the accuracy of data recommendation.

An embodiment of the present application provides a data recommendation method, including:

acquiring a first service object set and acquiring a knowledge graph; the first service object set comprises basic description information and historical behavior parameters corresponding to service objects, and the knowledge graph comprises an incidence relation between detail description information corresponding to each historical service object in an object library;

acquiring detail description information which has an incidence relation with the basic description information from the knowledge graph, taking the detail description information as target detail description information corresponding to the business object, and predicting pre-estimated behavior parameters corresponding to the business object based on the target detail description information;

and selecting a target business object for recommendation from the first business object set according to the historical behavior parameters and the estimated behavior parameters corresponding to the business objects.

Wherein the method further comprises:

generating the first set of business objects and the second set of business objects based on the object library; the first business object set and the second business object set are both determined by historical behavior parameters corresponding to historical business objects in the object library;

if the service triggering request hits the first service object set, executing the step of acquiring the first service object set and acquiring a knowledge graph;

and if the service triggering request hits the second service object set, acquiring the second service object set, and selecting a target service object from the second service object set based on the recommended evaluation value corresponding to each historical service object in the second service object set.

Wherein the generating the first set of business objects and the second set of business objects based on the object library comprises:

acquiring historical behavior parameters corresponding to each historical service object in the object library; the historical behavior parameters comprise click quantity and display quantity;

dividing historical service objects contained in the object library based on the historical behavior parameters to obtain a first to-be-processed set and a second to-be-processed set;

according to the click rate and the display amount, counting click rates corresponding to the historical service objects respectively, and sequencing the historical service objects contained in the first to-be-processed set and the second to-be-processed set respectively based on the click rates;

selecting M historical service objects from the sorted first to-be-processed set according to a sorting order to serve as the second service object set; m is a positive integer less than or equal to K, and K is the number of historical service objects contained in the first set to be processed;

selecting N/2 historical business objects from the sorted second to-be-processed set according to a sorting sequence, selecting N/2 non-shown business objects from all non-shown business objects in the object library, and determining the N/2 historical business objects and the N/2 non-shown business objects as the first business object set; n is a positive integer less than or equal to I, and I is the sum of the number of the historical business objects contained in the second to-be-processed set and the number of all the business objects which are not shown.

Wherein the selecting N/2 non-shown business objects from all non-shown business objects in the object library comprises:

acquiring all non-displayed business objects from the object library, and determining a first category corresponding to each non-displayed business object; the first category comprises a second category;

according to the click rate corresponding to each historical object, counting a first average click rate corresponding to the object library, a second average click rate corresponding to the first category and a third average click rate corresponding to the second category;

determining click evaluation values corresponding to each non-displayed business object according to the first average click rate, the second average click rate and the third average click rate;

and sorting all the non-displayed business objects based on the click evaluation value, and selecting the N/2 non-displayed business objects from all the sorted non-displayed business objects according to a sorting sequence.

The acquiring, from the knowledge graph, detail description information having an association relationship with the basic description information as target detail description information corresponding to the service object, and predicting an estimated behavior parameter corresponding to the service object based on the target detail description information includes:

extracting target keywords from the basic description information, and searching entity character strings matched with the target keywords in the knowledge graph;

determining the detail description information associated with the entity character string in the knowledge graph as target detail description information corresponding to the business object;

inputting the target detail description information into a behavior prediction model, acquiring label information corresponding to the target detail description information based on the behavior prediction model, and determining behavior parameters corresponding to the label information as pre-estimated behavior parameters corresponding to the business object.

The estimated behavior parameters comprise estimated click quantity and estimated display quantity;

the selecting a target business object for recommendation from the first business object set according to the historical behavior parameters and the pre-estimated behavior parameters of the business object comprises:

overlapping the click rate in the historical behavior parameters corresponding to the business object with the estimated click rate to obtain a first parameter;

overlapping the display quantity and the estimated display quantity in the historical behavior parameters corresponding to the business object to obtain a second parameter;

constructing beta distribution corresponding to the business object based on the first parameter and the second parameter;

and determining a sampling reference value corresponding to the business object according to the beta distribution, and selecting a target business object for recommendation from the first business object set based on the sampling reference value.

Determining a sampling reference value corresponding to the business object according to the beta distribution, and selecting a target business object for recommendation from the first business object set based on the sampling reference value, wherein the selecting comprises:

determining a sampling reference value corresponding to the business object according to the beta distribution, sequencing all business objects in the first business object set based on the sampling reference value, and selecting C objects to be recommended from all the sequenced business objects; c is a positive integer less than or equal to N, and N is the number of the business objects contained in the first business object set;

and sequentially inputting the target detail description information corresponding to the C objects to be recommended into a recommendation model, acquiring recommendation evaluation values corresponding to the C objects to be recommended respectively based on the recommendation model, and determining a target service object for recommendation based on the recommendation evaluation values.

The sequentially inputting the target detail description information corresponding to the C objects to be recommended into a recommendation model, obtaining recommendation evaluation values corresponding to the C objects to be recommended respectively based on the recommendation model, and determining a target service object for recommendation based on the recommendation evaluation values includes:

sequentially inputting the target detail description information corresponding to the C objects to be recommended into the recommendation model to generate object feature vectors corresponding to the C objects to be recommended;

determining target users corresponding to the C objects to be recommended, and acquiring user portrait data of the target users;

inputting the user portrait data into the recommendation model, and generating a user feature vector corresponding to the user portrait data;

performing inner product operation on the user characteristic vectors and object characteristic vectors corresponding to the C objects to be recommended respectively to obtain recommendation evaluation values corresponding to the C objects to be recommended respectively;

and determining the object to be recommended with the maximum recommendation evaluation value as a target business object for recommendation.

Wherein the method further comprises:

and acquiring an object recommendation template, fusing the object recommendation template and the target service object, generating recommendation information for the target user, and displaying the recommendation information to the target user.

Wherein the method further comprises:

recording real-time behavior data of the recommendation information based on a public gateway interface, and uploading the real-time behavior data to a log system;

and updating historical behavior parameters corresponding to all historical behavior objects in the object library based on the real-time behavior data on the log system.

An embodiment of the present application provides a data recommendation device in one aspect, including:

the acquisition module is used for acquiring a first service object set and acquiring a knowledge graph; the first service object set comprises basic description information and historical behavior parameters corresponding to service objects, and the knowledge graph comprises an incidence relation between detail description information corresponding to each historical service object in an object library;

the prediction module is used for acquiring detail description information which has an incidence relation with the basic description information from the knowledge graph, using the detail description information as target detail description information corresponding to the business object, and predicting the predicted behavior parameters corresponding to the business object based on the target detail description information;

and the selection module is used for selecting a target business object for recommendation from the first business object set according to the historical behavior parameters and the estimated behavior parameters corresponding to the business objects.

Wherein the apparatus further comprises:

a set generating module for generating the first business object set and the second business object set based on the object library; the first business object set and the second business object set are both determined by historical behavior parameters corresponding to historical business objects in the object library;

the object triggering module is used for executing the steps of acquiring the first service object set and acquiring the knowledge graph if the service triggering request hits the first service object set;

the object triggering module is further configured to, if the service triggering request hits the second service object set, obtain the second service object set, and select a target service object from the second service object set based on a recommended evaluation value corresponding to each historical service object in the second service object set.

Wherein the set generation module comprises:

the parameter acquisition unit is used for acquiring historical behavior parameters corresponding to each historical service object in the object library; the historical behavior parameters comprise click quantity and display quantity;

the dividing unit is used for dividing historical service objects contained in the object library based on the historical behavior parameters to obtain a first to-be-processed set and a second to-be-processed set;

the first sequencing unit is used for counting the click rate corresponding to each historical service object according to the click rate and the display amount, and sequencing the historical service objects contained in the first to-be-processed set and the second to-be-processed set based on the click rate;

a first selecting unit, configured to select M historical service objects from the sorted first set to be processed according to a sorting order, where the M historical service objects serve as the second service object set; m is a positive integer less than or equal to K, and K is the number of historical service objects contained in the first set to be processed;

a second selecting unit, configured to select N/2 historical business objects from the sorted second to-be-processed set according to a sorting order, select N/2 business objects not shown from all business objects not shown in the object library, and determine the N/2 historical business objects and the N/2 business objects not shown as the first business object set; n is a positive integer less than or equal to I, and I is the sum of the number of the historical business objects contained in the second to-be-processed set and the number of all the business objects which are not shown.

Wherein the second selection unit includes:

the category determining subunit is used for acquiring all the non-displayed business objects from the object library and determining a first category corresponding to each non-displayed business object; the first category comprises a second category;

an average click rate determining subunit, configured to count a first average click rate corresponding to the object library, a second average click rate corresponding to the first category, and a third average click rate corresponding to the second category according to the click rate corresponding to each history object;

a click evaluation value determining subunit, configured to determine, according to the first average click rate, the second average click rate, and the third average click rate, a click evaluation value corresponding to each non-displayed service object;

and the object selection subunit is used for sorting all the non-displayed business objects based on the click evaluation value and selecting the N/2 non-displayed business objects from all the sorted non-displayed business objects according to a sorting sequence.

Wherein the prediction module comprises:

the extraction unit is used for extracting a target keyword from the basic description information and searching an entity character string matched with the target keyword in the knowledge graph;

a detail determining unit, configured to determine, as target detail description information corresponding to the service object, detail description information associated with the entity character string in the knowledge graph;

and the estimation parameter determining unit is used for inputting the target detail description information into a behavior prediction model, acquiring label information corresponding to the target detail description information based on the behavior prediction model, and determining the behavior parameter corresponding to the label information as the estimation behavior parameter corresponding to the business object.

the selection module comprises:

the first parameter determining unit is used for superposing the click rate in the historical behavior parameters corresponding to the business object and the estimated click rate to obtain a first parameter;

the second parameter determining unit is used for superposing the display quantity and the estimated display quantity in the historical behavior parameters corresponding to the business object to obtain a second parameter;

a beta distribution construction unit, configured to construct a beta distribution corresponding to the business object based on the first parameter and the second parameter;

and the sampling unit is used for determining a sampling reference value corresponding to the business object according to the beta distribution and selecting a target business object for recommendation from the first business object set based on the sampling reference value.

Wherein the sampling unit includes:

a to-be-recommended object determining subunit, configured to determine, according to the beta distribution, a sampling reference value corresponding to the service object, sort all service objects in the first service object set based on the sampling reference value, and select C to-be-recommended objects from all sorted service objects; c is a positive integer less than or equal to N, and N is the number of the business objects contained in the first business object set;

and the information input subunit is used for sequentially inputting the target detail description information corresponding to the C objects to be recommended into a recommendation model, acquiring recommendation evaluation values corresponding to the C objects to be recommended respectively based on the recommendation model, and determining a target service object for recommendation based on the recommendation evaluation values.

Wherein the information input subunit includes:

the first vector generation subunit is configured to sequentially input the target detail description information corresponding to the C objects to be recommended into the recommendation model, and generate object feature vectors corresponding to the C objects to be recommended;

the user portrait determining subunit is used for determining target users corresponding to the C objects to be recommended and acquiring user portrait data of the target users;

the second vector generation subunit is used for inputting the user portrait data into the recommendation model and generating a user feature vector corresponding to the user portrait data;

the operation subunit is configured to perform inner product operation on the user feature vectors and object feature vectors corresponding to the C objects to be recommended, so as to obtain recommendation evaluation values corresponding to the C objects to be recommended;

and the target business object determining subunit is used for determining the object to be recommended with the maximum recommendation evaluation value as the target business object for recommendation.

Wherein the apparatus further comprises:

and the display module is used for acquiring an object recommendation template, fusing the object recommendation template and the target service object, generating recommendation information for the target user and displaying the recommendation information to the target user.

Wherein the apparatus further comprises:

the recording module is used for recording the real-time behavior data of the recommendation information based on a public gateway interface and uploading the real-time behavior data to a log system;

and the updating module is used for updating the historical behavior parameters corresponding to all the historical behavior objects in the object library based on the real-time behavior data on the log system.

An aspect of the embodiments of the present application provides a computer device, including a memory and a processor, where the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the steps of the method in an aspect of the embodiments of the present application.

An aspect of embodiments of the present application provides a computer-readable storage medium storing a computer program comprising program instructions that, when executed by a processor, perform the steps of the method as described in an aspect of embodiments of the present application.

The method and the device for recommending the business objects can acquire a first business object set and a knowledge graph, wherein the first business object set can comprise basic description information corresponding to the business objects, the knowledge graph comprises an incidence relation between detail description information corresponding to each historical business object in an object library, the detail description information which is associated with the basic description information can be acquired from the knowledge graph and is used as target detail description information corresponding to the business objects, and the estimated behavior parameters corresponding to the business objects are predicted based on the target detail description information, so that the target business objects for recommending can be selected from the first business object set according to the historical behavior parameters and the estimated behavior parameters corresponding to the business objects. Therefore, all the service objects in the first service object set, no matter the service objects are displayed or not, can obtain corresponding detailed description information from the knowledge graph, predict the estimated behavior parameters corresponding to each service object, further determine the recommended service objects from the first service object set, when the service objects are new commodities, obtain complete description information of the new commodities based on the knowledge graph, predict the estimated behavior parameters (including estimated display amount and click amount) of the new commodities, provide more accurate parameter data for the recommendation of the new commodities, and further improve the accuracy of commodity recommendation.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings 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 of the present application, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a schematic diagram of a network architecture provided in an embodiment of the present application;

FIG. 2 is a schematic diagram of a scenario of data recommendation provided in an embodiment of the present application;

fig. 3 is a schematic flowchart of a data recommendation method provided in an embodiment of the present application;

FIG. 4 is a schematic diagram of a knowledge-graph provided by an embodiment of the present application;

FIG. 5 is a diagram illustrating detailed description information provided by an embodiment of the present application;

FIG. 6 is a schematic diagram of a method for predicting predicted behavior parameters according to an embodiment of the present disclosure;

FIG. 7 is a schematic diagram illustrating obtaining a recommended evaluation value based on a recommendation model according to an embodiment of the present application;

FIG. 8 is a flowchart illustrating another data recommendation method according to an embodiment of the present application;

FIG. 9 is a schematic diagram of generating a first set of business objects and a second set of business objects according to an embodiment of the present application;

FIG. 10 is a schematic diagram of a merchandise recommendation system according to an embodiment of the present invention;

FIG. 11 is a schematic diagram of another data recommendation scenario provided in an embodiment of the present application;

fig. 12 is a schematic structural diagram of a data recommendation device according to an embodiment of the present application;

fig. 13 is a schematic structural diagram of a computer device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Dynamic Product Advertisement (DPA) refers to an advertisement format that selects the most suitable product using an algorithm according to user characteristics, i.e., user's taste and characteristics, and presents the selected product to the user. In other words, dynamic merchandise advertisements are those that present the correct merchandise to the correct user. In the process of putting the dynamic commodity advertisement, commodity data and user behaviors can be analyzed based on artificial intelligence, so that accurate matching between user appeal and advertiser promoted commodities is achieved, and one-to-one advertisement display between the user and the commodities is completed.

Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

The scheme provided by the embodiment of the application relates to Machine Learning (ML) belonging to the field of artificial intelligence.

Machine learning is a multi-field cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

Please refer to fig. 1, which is a schematic diagram of a network architecture according to an embodiment of the present application. The network architecture may include a server 10a and a plurality of terminal devices (specifically, as shown in fig. 1, including a terminal device 10b, a terminal device 10c, and a terminal device 10d), where the server 10a may perform data transmission with each terminal device through a network.

The server 10a may obtain object data of all the displayed business objects (such as commodities), and count the display amount and click amount of the displayed business objects based on the object data, and further may obtain a business object set (the business objects included in the business object set may also be referred to as candidate objects) from all the business objects according to the display amount and click amount; the database corresponding to the server 10a stores a knowledge graph, which includes an association relationship between the detailed description information corresponding to each exposed business object in the object library. When the server 10a receives the service triggering request, the target detail description information corresponding to each service object in the service object set may be obtained according to the knowledge graph, and the estimated behavior parameter (which may also be referred to as estimated click rate) corresponding to each service object is predicted based on the target detail description information, so that the target service object recommended for the user is selected from the service object set based on the estimated behavior parameter, and the target service object is encapsulated into a dynamic commodity advertisement to be displayed to the user, that is, the dynamic commodity advertisement is sent to the terminal device corresponding to the user, and the terminal device may display the received dynamic commodity advertisement. The display amount and click rate generated by the dynamic commodity advertisement can be returned to the server 10a, and the server 10a can update the object data according to the returned display amount and click rate. Optionally, the scheme provided in the embodiment of the present application may also be directly implemented by the terminal device, that is, the terminal device predicts the estimated behavior parameters of each service object in the service object set, and selects the target service object for recommendation based on the estimated behavior parameters. The following description will be made by taking an example of how the server 10a predicts the predicted behavior parameters and selects the target business object. The terminal device 10b, the terminal device 10c, the terminal device 10d, and the like may include a mobile phone, a tablet computer, a notebook computer, a palm computer, a Mobile Internet Device (MID), a wearable device (e.g., a smart watch, a smart band, and the like), and the like.

Please refer to fig. 2, which is a schematic view of a data recommendation scenario provided in an embodiment of the present application. As shown in fig. 2, taking the service object as a commodity as an example, the server 10a may obtain all already displayed commodities and all not displayed commodities in a commodity library (also referred to as an object library), and count the display amount (i.e., the number of times that the commodity is displayed as an advertisement to a user) and the click amount (i.e., the number of times that the commodity is displayed as an advertisement to a user and clicked by the user) corresponding to each displayed commodity, respectively, the server 10a may select N/2 commodities from the already displayed commodities according to the display amount and the click amount (e.g., select N/2 commodities from the already displayed commodities whose display amount and click amount are higher than a quantity threshold value), and select N/2 commodities from the not displayed commodities to form a service object set 20a, where the service object set 20a may include basic description information of the commodities, such as basic description information of the commodity 1 is "aaa frost", the basic description information of the product 2 is "bbb facial cleanser", the basic description information of the product 3 is "ccc paper extraction", and the like, and the displayed product included in the business object set 20a may further include the display amount and click amount of the displayed product.

The server 10a may obtain a knowledge graph 20b, where the knowledge graph 20b may include an association relationship between detailed description information of each displayed commodity in a commodity library, and the knowledge graph 20b may further include a plurality of entity character strings, and since the current scene is a commodity scene, the knowledge graph 20b is a knowledge graph about commodities, that is, the entity character strings in the knowledge graph 20b are character strings about commodity brand names, commodity category names, commodity attribute names, and commodity price description manners. Different commodities in the commodity library may have the same brand, the same category, the same attribute, the same price, and the like, and the same commodity may have different attributes, different prices, and the like, and the relationship between different attributes of the same commodity and the relationship between the same information of different commodities may be included in the knowledge map 20 b.

The server 10a may search the knowledge graph 20b for detailed description information having an association relationship with the basic description information of each product based on the basic description information of each product in the business object set 20a, taking the product 1 in the business object set 20a as an example, the basic description information of the product 1 in the business object set 20a is "aaa cream", and may obtain the detailed description information 20c corresponding to the product 1 according to the knowledge graph 20b, where the detailed description information 20c may include a brand name (e.g., "aaa"), a product class name (e.g., "skin care product", "cream"), an attribute name (e.g., "night use", "new money", "lady use"), and price information (e.g., "￥ 90"), where the price information may also be represented by a label corresponding to a price interval, the label may be a number or a alphabetic character, e.g., a number "1" represents a price of a product below 10 yuan, a number "9" represents a price of a product above a yuan, or a letter "a price of a product below 10 yuan, and the like.

The server 10a obtains the behavior prediction model 20d, where the behavior prediction model 20d has been trained based on the detailed description information of the commodity, and may be used to predict the predicted behavior parameters corresponding to each commodity of the business object set 20 a. The server 10a may input the detailed description information 20c corresponding to the product 1 into the behavior prediction model 20d, and since the brand name, the class name, the attribute name, and the price information in the detailed description information 20c are described in natural language, it is necessary to convert the input detailed description information 20c into an information vector using an Embedding layer (Embedding layer) in the behavior prediction model 20 d. In other words, each word (such as "cream", "skin care product", "night use", "woman", "application", and the like) in the detail description information 20c may be converted into a word vector, and then the word vectors corresponding to each word are spliced into an information vector corresponding to the detail description information 20c, so that label information corresponding to the detail description information 20c may be obtained based on a full connection layer in the behavior prediction model 20d, and a behavior parameter corresponding to the label information is determined as an estimated behavior parameter corresponding to the product 1, where the estimated behavior parameter may be understood as an estimated click quantity and an estimated display quantity corresponding to the product 1. It should be noted that, since the non-displayed merchandise included in the business object set 20a has no display amount and click amount, and the click amount and display amount data corresponding to the displayed merchandise is not sufficient, it may be necessary to predict the estimated click amount and estimated display amount of the merchandise based on the behavior prediction model 20 d.

The server 10a may construct a beta distribution for the commodity based on the estimated click rate and the estimated display rate, where the beta distribution includes a first parameter and a second parameter, the first parameter is expressed as the click rate of the commodity, and the second parameter may be expressed as the number of times the commodity is not clicked. From the beta distribution, a sample reference value for the commodity may be determined. When the server 10a obtains the sampling reference value corresponding to each commodity in the business object set 20a, all commodities in the business object set 20a may be sorted based on the sampling reference value, and the server 10a may select, from the sorted commodities, a target commodity 20e recommended for the user based on the user information.

The server 10a may obtain a recommendation template in the information publishing platform, fuse the target product 20e with the recommendation template, and send the fused product advertisement to the terminal device 10b of the user, and the terminal device 10b may display the product advertisement in a display page of the information publishing platform after receiving the product advertisement sent by the server 10a, where the product advertisement may include the target product 20e and the document information (e.g., "money order, surprise attack", etc.).

Please refer to fig. 3, which is a flowchart illustrating a data recommendation method according to an embodiment of the present application. As shown in fig. 3, the data recommendation method may include the steps of:

step S101, acquiring a first service object set and acquiring a knowledge graph; the first service object set comprises basic description information and historical behavior parameters corresponding to service objects, and the knowledge graph comprises an incidence relation between detail description information corresponding to each historical service object in an object library;

specifically, a server (e.g., the server 10a in the embodiment corresponding to fig. 2) may obtain a first service object set (e.g., the service object set 20a in the embodiment corresponding to fig. 2), where the obtaining manner of the first service object set may be that the server determines, as the first service object set, a plurality of service objects with smaller behavior parameters or without behavior parameters (the display amount and/or the click amount belong to a number threshold range) based on behavior parameters of the service objects (including the display amount and the click amount corresponding to the service objects). The first business object set may include basic description information (information such as names of business objects) and historical behavior parameters (recorded display amount and click amount) corresponding to the multiple business objects, and the business objects may include: merchandise, scenery, articles, caricatures, etc.

The server obtains a knowledge graph (such as the knowledge graph 20b in the embodiment corresponding to fig. 2), wherein the knowledge graph uses a visualization technology to describe knowledge resources and carriers thereof, and mines, analyzes, constructs, draws and displays knowledge and mutual relations between the knowledge resources and the carriers. The knowledge graph includes an association relationship between detailed description information corresponding to each historical service object in the object library, that is, the knowledge graph may include entity character strings corresponding to a plurality of entities and a category name corresponding to each entity character string, where the category name represents a service category where the corresponding entity character string is located.

For example, the service type corresponding to the entity character string HW in the knowledge graph is a brand, the service type corresponding to the entity character string clothing in the knowledge graph is a category, the service type corresponding to the entity character string mobile phone in the knowledge graph is a category, the service type corresponding to the entity character string digital product in the knowledge graph is a category, the service type corresponding to the entity character string double shot in the knowledge graph is an attribute, the service type corresponding to the entity character string ￥ 5000 in the knowledge graph is a price, the service type corresponding to the kylin 990 in the knowledge graph is an attribute, the service type corresponding to the xxx9 in the knowledge graph is a model, and the service type corresponding to the android in the knowledge graph is a model.

Please refer to fig. 4, which is a schematic diagram of a knowledge graph according to an embodiment of the present application. In the knowledge graph, complex relationships among multiple entities, such as short sleeves, grades, clothes, trousers, grades and clothes, can be represented in the form of triples. Taking the example of constructing a knowledge graph of the article "clothes", the entity string associated with the article "clothes" may be from entities involved in the electronic goods repository, such as clothes style, clothes material, gender, age, etc. The connection line between every two entity character strings can be represented as a corresponding logical relationship, i.e. the respective service categories of the two entity character strings are identified, and the knowledge graph of the article "clothes" corresponding to fig. 4 includes 4 styles: short sleeves, long sleeves, one-piece dress, short skirt; 2, sex: male and female; 2 ages: children, young adults; 4 materials: pure cotton, terylene, acrylic fiber and chinlon. Thus, the number of item-style logical relationships is 4, the number of item-gender logical relationships is 2, the number of item-age logical relationships is 2, and the number of item-material logical relationships is 4.

Step S102, acquiring detail description information which has an incidence relation with the basic description information from the knowledge graph, taking the detail description information as target detail description information corresponding to the business object, and predicting estimated behavior parameters corresponding to the business object based on the target detail description information;

specifically, the server may extract the target keyword from the basic description information, and search the knowledge graph for an entity string matching the target keyword. Extracting the target keyword can abandon meaningless characters in the basic description information, for example, the basic description information of a certain business object is as follows: "summer undershirts in the simple atmosphere", then the target keywords are: an undershirt. Since the target keyword is a descriptive character input by the user or the operator for the service object, there may be a case where the used word is irregular, such as "undershirt" in the foregoing example, the server further needs to convert the target keyword into a standard synonym, such as converting the keyword "undershirt" into an entity character string "T-shirt", and the obtained synonym is the entity character string in the knowledge graph. Of course, if the basic description information of the business object is only the name of the business object, the basic description information can be used as the target keyword.

The server may determine the detail description information associated with the entity character string in the knowledge graph as the target detail description information corresponding to the business object. Please refer to fig. 5, which is a schematic diagram of detailed description information provided in an embodiment of the present application. As shown in fig. 5, taking the service object as a mobile phone as an example, the basic description information of the service object is: "abcxxx 8," through a knowledge graph, may obtain target detail description information corresponding to the service object, where the target detail description information specifically is: the brands are abc and xxx, the categories are digital products and mobile phones, the models are xxx8 and android, the attributes are four shots, new money and xx990, and the price is 9 (high range).

The server may input the found target detail description information into a behavior prediction model (such as the behavior prediction model 20d in the embodiment corresponding to fig. 2), obtain tag information corresponding to the target detail information based on the behavior prediction model, and determine a behavior parameter corresponding to the tag information as an estimated behavior parameter corresponding to the service object. After the server obtains the target detail description information corresponding to the business object, the server can predict the predicted behavior parameters corresponding to the business object through the behavior prediction model, namely predict the predicted click quantity and the predicted display quantity corresponding to the business object. The prediction behavior model may be a deep neural network model and may be used to predict the prediction behavior parameters corresponding to each service object in the first service object set.

The following describes in detail how to predict the predicted behavior parameters based on the target detail description information: the server can convert each descriptor contained in the target detail description information into a one-dimensional vector, called a unit vector, in a one-hot (one-hot) coding mode, acquire a behavior prediction model, classify an input vector consisting of a plurality of descriptors by the behavior prediction model, obtain label information corresponding to the target detail description information, and determine an estimated behavior parameter corresponding to the business object based on the behavior parameter corresponding to the label information. It can be understood that a plurality of descriptors contained in the target detail description information can be understood as character strings in the knowledge graph, a unit vector corresponding to each descriptor is an one-hot code corresponding to the corresponding character string, the one-hot code is a vector in which the vector only contains one 1 and the rest are 0, and the number of dimensions of the one-hot code is the same as the number of the character strings contained in the knowledge graph. When the knowledge graph contains a large number of character strings, the dimension of the unit vector corresponding to the descriptor is large, and unnecessary storage space is occupied, for example, when the knowledge graph contains 10000 character strings for all business objects, the dimension of the unique hot code corresponding to each character string is 10000.

Please refer to fig. 6, which is a schematic diagram of a method for predicting predicted behavior parameters according to an embodiment of the present disclosure. As shown in fig. 6, taking the business object as a commodity as an example, after the server converts all the descriptors in the target detail description information 30a into unit vectors, that is, all the information such as commodity types, commodity names, commodity models, commodity prices, etc. in the target detail description information into unit vectors, a behavior prediction model is obtained, which may include an embedded layer 30b and a plurality of fully connected layers (e.g., a fully connected layer 30 c); the server inputs a plurality of unit vectors into the behavior prediction model, and first inputs the embedded layer 30b of the behavior prediction model, and the embedded layer 30b can be regarded as a word vector conversion model in nature, and the embedded layer 30b can reduce a unit vector of a high dimension to a word vector of a low dimension. The vectors obtained by the embedding layer 30b may be referred to as word vectors, word vectors corresponding to each descriptor are spliced, and the spliced vectors are input to the fully-connected layer 30c of the behavior prediction model, and of course, if the behavior prediction model includes a plurality of fully-connected layers, the spliced vectors are input to the first fully-connected layer, the input result of the first fully-connected layer is used as the input data of the second fully-connected layer, and the output result of the second fully-connected layer is used as the input data of the third fully-connected layer, and so on. Through the full connection layer 30c, the feature vector 30d corresponding to the target detail description information finally obtained matches the feature vector 30d with a plurality of preset feature vectors contained in the behavior prediction model, determines the label corresponding to the preset feature vector with the highest matching degree as the label information corresponding to the target detail description information, and searches the behavior parameter matched with the label information from the stored label-parameter correspondence table 30g as the estimated behavior parameter corresponding to the commodity.

The behavior prediction model is trained and completed based on the mapping relation between the detail description information corresponding to each displayed commodity in the commodity library and the actual behavior parameters. In the training process, the actual behavior parameters corresponding to the commodity can be divided into several grades, and each grade is provided with label information, for example, the actual behavior parameters less than or equal to 100 can be divided into one grade and represented by a label "1"; actual behavior parameters greater than 100 and less than or equal to 500 are classified as a level, represented by the label "2", and so on. In the tag-parameter correspondence table 30g, each tag and the behavior parameter corresponding to each tag may be stored, where the behavior parameter may be represented as an average value of all actual behavior parameters in the level corresponding to the tag information, for example, the behavior parameter a corresponding to the tag information 1 represents: the average value corresponding to all the actual behavior parameters in the class to which the label "1" belongs, if the class to which the label "1" belongs is the actual behavior parameter less than or equal to 100, and the class includes: 10. 20, 30, 40, 50, 60, 70, the behavior parameter a corresponding to the tag information 1 is: 40. and storing the trained behavior prediction model for predicting the predicted behavior parameters corresponding to each commodity in the first business object set, wherein the trained behavior prediction model can comprise feature vectors corresponding to a plurality of labels. After target detail description information corresponding to the commodity in the first business object set is input into the behavior prediction model, label information matched with the commodity can be obtained from the behavior prediction model, and the estimated behavior parameters of the commodity are determined according to behavior parameters corresponding to the matched label information. The server can predict the predicted behavior parameters corresponding to each commodity in the first service set through the behavior prediction model.

Optionally, the server may also directly predict, by using the behavior prediction model, an estimated click rate corresponding to each commodity in the first commodity set, where the estimated click rate is a ratio between the estimated click rate and the estimated display amount, and the server may use the value 100 as a default value of the display amount corresponding to the commodity, and multiply the estimated click rate predicted by the behavior prediction model and the default value of the display amount to obtain the estimated click amount corresponding to the commodity. When the behavior prediction model is trained based on the mapping relation between the detailed description information corresponding to each displayed commodity in the commodity library and the actual click rate, the behavior prediction model can predict the estimated click rate corresponding to the commodity in the first business object set.

Optionally, the estimated behavior parameter corresponding to each service object in the first service object set may also be determined based on the following formula (1):

wherein, C₀(k) Represents the estimated click rate, lambda, corresponding to the non-displayed business object k₁As initial value weight corresponding to the object library, C_d(k) And I_d(k) Respectively representing the total click rate (namely the sum of the click rates corresponding to all historical business objects contained in the object library) and the total display rate (namely the sum of the display rates corresponding to all historical business objects contained in the object library) corresponding to the object library to which the business object k which is not displayed belongs, and C_i(k) And I_i(k) And respectively representing the total click quantity and the total display quantity corresponding to the industry to which the non-displayed business object k belongs. Based on the formula (1), the estimated click rate corresponding to each non-displayed business object in the first business object set can be obtained. The estimated display amount corresponding to each non-displayed service object in the first service object set can default to a fixed value (for example, the estimated display amount is 100).

Step S103, selecting a target business object for recommendation from the first business object set according to the historical behavior parameters and the estimated behavior parameters corresponding to the business objects.

Specifically, the server may construct a Beta distribution (that is, a Beta distribution, which may also be considered as a probability distribution) for each business object in the first business object set based on predicted estimated behavior parameters obtained through prediction and historical behavior parameters corresponding to the business objects (that is, actual display amounts and click amounts statistically obtained for the displayed business objects), since the construction of the Beta distribution involves two parameters, that is, the first parameter α and the second parameter β represent the number of times that the business object is clicked, and β represents the number of times that the business object is not clicked, for the displayed business object in the first business object set, the click amount and the display amount corresponding to the displayed business object may be obtained, the sum of the click amount and the predicted click amount in the predicted behavior parameters is determined as the first parameter α, the sum of the display amount and the predicted display amount in the predicted behavior parameters is determined as the second parameter β, and the value obtained by subtracting the first parameter α is determined as the second parameter β, and the displayed business object distributed corresponding to the first business object is constructed according to the first parameter α and the second βSince historical behavior parameters are not generated for business objects which are not shown in the contract, the estimated click rate can be used as a first parameter α, the difference value between the estimated display amount and the estimated click rate can be used as a second parameter β, and then Beta distribution of the business objects which are not shown can be constructed, wherein the Beta distribution can be represented as Beta (α), wherein α ═ C₀(k)+C(k)，β＝I₀(k) + I (k) - α, C (k) and I (k) are expressed as the actual click volume and the display volume of the business object k (for the business object not shown, C (k) and I (k) are both 0), C₀(k) And I₀(k) The beta distribution has different shapes aiming at different first parameters α and second parameters β, but the beta distribution is in [0,1 ] no matter what shape]Within the interval, a beta distribution can therefore be used to describe events within various 0-1 intervals, i.e. a uniform distribution when α -1 and β -1.

The server may generate a random number for each service object based on Thompson Sampling (TS) algorithm by using the Beta distribution of the service object, which may also be referred to as a TS score of the service object, to represent a Sampling reference value corresponding to the service object in the current environment. Based on the beta distribution corresponding to each business object, a sampling reference value corresponding to each business object in the first business object set can be generated. All the business objects contained in the first business object set are sorted according to the sampling reference value, and C (for example, C is 10) to-be-recommended objects are selected from the sorted business objects, that is, the business object C before the sampling reference value is selected from the first business object set as the to-be-recommended object. It should be noted that C represents the number of objects to be recommended, and C is a natural number smaller than N, and N is the number of business objects included in the first business object set.

The server can obtain a recommendation model, the recommendation model is trained based on detailed description information corresponding to the displayed business objects in the object library and display user information corresponding to the displayed business objects, the trained recommendation model can be used for obtaining recommendation evaluation values corresponding to the C objects to be recommended respectively, and the recommendation model can be a deep neural network model and is composed of an embedded layer and a full connection layer. The server can sequentially input the target detail description information corresponding to the C objects to be recommended into the recommendation model to generate object feature vectors corresponding to the C objects to be recommended; determining target users corresponding to the C objects to be recommended, and acquiring user portrait data of the target users; inputting user portrait data into a recommendation model, and generating a user feature vector corresponding to the user portrait data; performing inner product operation on the user characteristic vector and object characteristic vectors corresponding to the C objects to be recommended respectively to obtain recommendation evaluation values corresponding to the C objects to be recommended respectively; and determining the object to be recommended with the maximum recommendation evaluation value as a target business object for recommendation. In other words, the server may obtain target users corresponding to the C objects to be recommended (i.e., users to be presented with the objects to be recommended, such as the user a when determining the recommended object for the user a, the user a is the target user), and obtain user portrait data corresponding to the target users, such as information about the age, sex, geographical location, interests, and the like of the users; the server can also obtain target detail description information corresponding to the C objects to be recommended respectively, a user feature vector corresponding to the user portrait data and an object feature vector corresponding to the target detail description information can be obtained through a recommendation model, the object feature vector and the user feature vector are subjected to inner product, and a recommendation evaluation value (also called a recommendation score corresponding to each object to be recommended) corresponding to each object to be recommended respectively can be obtained. It should be noted that the recommendation model may be divided into two branches, one branch is used to obtain a user feature vector corresponding to the user portrait data, and the other branch is used to obtain an object feature vector corresponding to the object to be recommended.

Please refer to fig. 7, which is a schematic diagram of obtaining a recommended evaluation value based on a recommendation model according to an embodiment of the present application. As shown in fig. 7, taking the business object as an example, the process of acquiring the recommended evaluation value will be described in detail: the recommendation model may include two branches, i.e. branch 1 and branch 2, each of branch 1 and branch 2 may be composed of an embedded layer and a plurality of fully-connected layers, e.g. branch 1 may include embedded layer 1, fully-connected layer 2, etc., and branch 2 may include embedded layer 2, fully-connected layer 3, fully-connected layer 4, etc. The server can obtain C target users corresponding to objects to be recommended, and obtain user portrait data corresponding to the target users (for example, age is 25 years old, gender is female, geographic location of the user is hong Kong, interest and hobbies are favorite of Chinese style, etc.), generate user features 40a based on the user portrait data, namely determine description information contained in the user portrait data as the user features 40a, such as '25 years old', 'female', 'China hong Kong', 'Chinese style', and input the user features 40a into a branch 1 of a recommendation model, and obtain user feature vectors corresponding to the user portrait data based on the embedded layer 1 and a plurality of full connection layers (such as the full connection layer 1 and the full connection layer 2).

The server may input the commodity features 40C (i.e., a plurality of descriptors included in the target detail description information, such as "abc", "xxx", "digital product", "mobile phone" in the embodiment corresponding to fig. 5) corresponding to the C objects to be recommended into the branch 2 of the recommendation model, respectively, and obtain a commodity feature vector (i.e., an object feature vector) corresponding to each object to be recommended based on the embedded layer 2 and a plurality of fully-connected layers (e.g., the fully-connected layer 3 and the fully-connected layer 4). The server may perform inner product on the user feature vector obtained in the branch 1 and each commodity feature vector obtained in the branch 2, respectively, to obtain a recommendation evaluation value 40c corresponding to each object to be recommended. The user features 40a and the commodity features 40c are text information described in natural language, so the server also needs to convert the user features 40a and the commodity features 40c into vector representations, and the vector conversion process and the data processing process of the embedding layer may refer to the description in the embodiment corresponding to fig. 6, which is not described herein again.

The server may rank the C objects to be recommended based on the recommendation evaluation value, and determine the object to be recommended with the largest recommendation evaluation value as a target product for recommendation, in other words, the object to be recommended with the largest recommendation evaluation value is a recommended product corresponding to the target user.

Please refer to fig. 8, which is a flowchart illustrating another data recommendation method according to an embodiment of the present application. As shown in fig. 8, the data recommendation method may include the steps of:

step S201, obtaining historical behavior parameters corresponding to each historical service object in the object library;

specifically, the server may obtain, based on the reflowed object data, historical behavior parameters corresponding to each historical service object in the object library, where the object library may include all service objects corresponding to a certain promotion party, that is, the object library may include displayed service objects and non-displayed service objects, the displayed service objects may also be referred to as historical service objects, the non-displayed service objects refer to objects that are not displayed on a common platform, such as newly produced products, for the displayed service objects, historical behavior parameters, that is, display amount and click amount, may be counted, the display amount of the service object is increased once per display, the click amount of the service object is increased once per click and browse of the user, when a service object is displayed on a certain platform, the user does not click and browse the service object, the display amount of the business object is increased once, and the click rate is kept unchanged. For example, the object library may refer to all commodities in a certain e-commerce platform (e.g., naobao, kyoto, etc.), the historical service object may refer to a commodity which has been displayed on the e-commerce platform or other platforms and has a user behavior, the reflowed object data is the number of times of display of the commodity and the operation behavior data of the user when the commodity is displayed (e.g., user's behavior such as clicking, browsing, purchasing, paying attention to, adding a shopping cart, etc.), and based on the reflowed object data, historical behavior parameter statistics may be performed on the displayed commodity, that is, the display amount and click amount of the commodity are counted.

Step S202, based on the historical behavior parameters, dividing historical service objects contained in the object library to obtain a first to-be-processed set and a second to-be-processed set;

specifically, the server may divide the historical service objects included in the object library according to the size of the historical behavior parameter, so as to obtain a first to-be-processed set and a second to-be-processed set. The basis for dividing the historical business objects contained in the object library includes but is not limited to: determining the historical business objects with the sum of the display amount and the click amount larger than a first quantity threshold (such as 300) in the object library as a first set to be processed, and determining the residual historical business objects in the object library as a second set to be processed; or determining the historical service objects with the display amount larger than a second threshold and the click amount larger than a third threshold in the object library as a first set to be processed, and determining the remaining historical service objects in the object library as a second set to be processed; or all the historical business objects in the object library are sorted from large to small based on the display amount and the click amount, a certain proportion (such as 60%) of the historical business objects are selected according to the sorting order and determined as a first set to be processed, and the remaining historical business objects in the object library are determined as a second set to be processed. The first quantity threshold, the second quantity threshold, the third quantity threshold and the specific proportion may be set manually according to actual needs, and are not limited specifically here.

Optionally, the business objects not shown in the object library may also be added to the second to-be-processed set. In other words, the second to-be-processed set may be composed of a part of the exposed business objects in the object library (i.e., except the historical business objects in the first to-be-processed set) and all the unexposed business objects.

Step S203, according to the click rate and the display amount, counting the click rate corresponding to each historical service object, and sorting the historical service objects contained in the first to-be-processed set and the second to-be-processed set based on the click rate;

specifically, the server may count the click rate corresponding to each historical service object according to the display amount and the click amount corresponding to the historical service object, where the click rate is expressed as a ratio of the click amount to the display amount, i.e., the click amount/the display amount. And respectively sequencing all the historical service objects contained in the first to-be-processed set and all the historical service objects contained in the second to-be-processed set according to the sequence of the click rate from large to small.

Step S204, selecting M historical service objects from the sorted first to-be-processed set according to a sorting order, and using the M historical service objects as the second service object set;

specifically, the server may select, from the sorted first set to be processed, the top M historical service objects according to the sorting order as a second service object set, that is, the second service object set includes the historical service objects of M before the click rate in the first set to be processed, where M is a positive integer less than or equal to K, and K is the number of the historical service objects included in the first set to be processed.

Step S205, selecting N/2 historical business objects from the sorted second to-be-processed set according to a sorting order, selecting N/2 business objects which are not shown from all business objects which are not shown in the object library, and determining the N/2 historical business objects and the N/2 business objects which are not shown as the first business object set;

specifically, the server may select N/2 historical service objects from the sorted second to-be-processed set according to the sorting order, that is, select a historical service object N/2 before the click rate from the second to-be-processed set; the server may further select N/2 non-shown business objects from all non-shown business objects in the object library, and form a first business object set together with the historical business objects N/2 before the click rate, that is, the first business object set may include N business objects, where N is a positive integer less than or equal to I, and I is a sum of the number of the historical business objects included in the second to-be-processed set and the number of all non-shown business objects. Optionally, when the second to-be-processed set includes a business object not shown in the object library, I at this time is the total number of business objects included in the second to-be-processed set.

The following describes the selection strategy of N/2 non-shown business objects: acquiring all non-displayed business objects from an object library, and determining a first category corresponding to each non-displayed business object, wherein the first category comprises a second category; according to the click rate corresponding to each historical object, counting a first average click rate corresponding to the object library, a second average click rate corresponding to the first category and a third average click rate corresponding to the second category; determining click evaluation values corresponding to each non-displayed business object according to the first average click rate, the second average click rate and the third average click rate; and sorting all the non-displayed business objects based on the click evaluation value, and selecting N/2 non-displayed business objects from all the sorted non-displayed business objects according to a sorting sequence. The server can acquire a first category and a second category to which each non-displayed business object belongs from the object library, wherein the first average click rate is a result obtained by accumulating click rates corresponding to each historical business object in the object library and dividing the result by the number of the historical business objects; the second average click rate is a result obtained by accumulating the click rate corresponding to each historical service object in the first category and dividing the result by the number of the historical service objects contained in the first category; the third average click rate is a result obtained by accumulating the click rate corresponding to each historical service object in the second category and dividing the result by the number of the historical service objects included in the second category. The first average click rate, the second average click rate, and the third average click rate are subjected to weighted summation, so as to obtain a click evaluation value corresponding to the business object that is not shown, which can be specifically expressed as formula (2):

wherein p (k) represents a click evaluation value corresponding to the non-shown business object k; c_d(k) And I_d(k) Respectively representing the total click quantity and the total display quantity corresponding to the object library to which the business object k which is not displayed belongs, C_d(k) And I_d(k) The ratio between the average click rates is a first average click rate; c₁(k) And I₁(k) Respectively representing the total click rate (namely the sum of the click rates corresponding to all the historical business objects contained in the first category) and the total display rate (namely the sum of the display rates corresponding to all the historical business objects contained in the first category) corresponding to the first category to which the business object k which is not displayed belongs, and C₁(k) And I₁(k) The ratio between the first average click rate and the second average click rate is a second average click rate; c₂(k) And I₂(k) Respectively representing the total click rate (namely the sum of the click rates corresponding to all the historical business objects contained in the second category) and the total display amount (namely the sum of the display amounts corresponding to all the historical business objects contained in the second category) corresponding to the second category to which the business object k which is not displayed belongs, and C₂(k) And I₂(k) The ratio between the average click rates is the third average click rate; theta₁、θ₂、θ₃Respectively expressed as the weight coefficients corresponding to the first average click rate, the second average click rate and the third average click rate.

Based on the above formula (2), the click evaluation value corresponding to each non-displayed business object in the object library can be calculated, the server can sort all non-displayed business objects contained in the object library based on the click evaluation value, and select N/2 non-displayed business objects from the sorted non-displayed business objects according to the sorting order (i.e. according to the order of the click evaluation values from large to small), namely select the first N/2 non-displayed business objects with the largest click evaluation value. For example, if the business objects not shown are: and for the female short sleeves, the object library to which the non-displayed business objects belong is a commodity library to which the female short sleeves belong, the first category to which the non-displayed business objects belong is 'female clothes' which the female short sleeves belong, and the second category to which the non-displayed business objects belong is 'coats' which the female short sleeves belong.

Please refer to fig. 9, which is a schematic diagram of generating a first business object set and a second business object set according to an embodiment of the present application. As shown in fig. 9, after acquiring the object library 41a, the server may count the exposure amount and the click amount corresponding to each service object in the object library 41a based on the log system, and store the counted exposure amount and click amount in the behavior parameter table 41 b. The log system may be configured to store original display and click data corresponding to all displayed service objects, and the behavior parameter table 41b may include the service objects and display amounts and click amounts corresponding to the service objects, where, for example, the click amount corresponding to the service object 1 is 2, and the corresponding display amount is 5; the business object 2 has no click volume and display volume, which indicates that the business object 2 is a new business object, namely, an undisplayed business object; the click volume corresponding to the business object 3 is 30, the display volume corresponding to the business object 3 is 500, and so on.

The server may divide the service objects in the object library 41a based on the click quantity and the display quantity stored in the behavior parameter table 41b, and divide all the service objects included in the object library 41a into a first to-be-processed set 41c, a second to-be-processed set 41d, and an unexposed service object set 41e, where all the service objects in the first to-be-processed set 41c are service objects that have been displayed and have display quantity and click quantity greater than a certain set threshold (e.g., a set threshold is 150), such as a service object 3, a service object 4, and the like; all the business objects in the second to-be-processed set 41d are the business objects which have been displayed and the display amount and the click amount are less than or equal to the set threshold (for example, the set threshold is 150), such as the business object 1 and the business object 2; all the business objects in the non-shown business object set 41e are non-shown business objects, i.e. business objects without click volume and display volume.

The server may count click rates (i.e., ratios of click amounts to display amounts) corresponding to the service objects in the first to-be-processed set 41c, sort the service objects in the first to-be-processed set 41c according to the size of the click rates to obtain a service object sorting table 41f, and select the first M service objects from the service object sorting table 41f as a second service object set 41 i; similarly, the server may count the click rate corresponding to the service objects in the second to-be-processed set 41d, sort the service objects in the second to-be-processed set 41d based on the click rate to obtain a service object sorting table 41g, and select the first N/2 service objects from the service object sorting table 41 g; the server may further calculate a click evaluation value corresponding to the service object in the non-displayed service object set 41e based on the formula (2), rank the service objects in the non-displayed service object set 41e based on the click evaluation value to obtain a service object ranking table 41h, select the first N/2 non-displayed service objects from the service object ranking table 41h, and form the first service object set 40j together with the first N/2 service objects selected in the service object ranking table 41 g. It should be noted that the service objects in the service object ordering table 41f and the service object ordering table 41g are ordered from the largest click rate to the smallest click rate, and the service objects in the service object ordering table 41h are ordered from the largest click evaluation value to the smallest click evaluation value.

Step S206, if the service triggering request hits the first service object set, acquiring the first service object set and acquiring a knowledge graph;

specifically, after generating the first service object set and the second service object set, the server may set hit probabilities for the first service object set and the second service object set, respectively, where the hit probability of the first service object set is P (e.g., P ═ 20%), and the hit probability of the second service object set is 1-P (e.g., 1-P ═ 80%). When the hit probability of the second service object set is greater than the hit probability of the first service object set, it indicates that the probability of selecting a service object in the second service object set as a target service object is high when a request is triggered, for each trigger request, the object set hit by the request is determined based on the hit probability, and when the service trigger request hits the first service object set, the server may obtain the first service object set and obtain a knowledge graph, and the specific implementation process may refer to step S101 in the embodiment corresponding to fig. 3, which is not described herein again.

Step S207, acquiring detail description information which has an association relation with the basic description information from the knowledge graph, taking the detail description information as target detail description information corresponding to the business object, and predicting estimated behavior parameters corresponding to the business object based on the target detail description information;

step S208, selecting a target business object for recommendation from the first business object set according to the historical behavior parameters and the estimated behavior parameters of the business object;

the specific implementation process of step S207 to step S208 may refer to step S102 to step S103 in the embodiment corresponding to fig. 3, which is not described herein again.

Step S209, if the service triggering request hits the second service object set, acquiring the second service object set, and selecting a target service object from the second service object set based on a recommended evaluation value corresponding to each historical service object in the second service object set;

specifically, when the service trigger request hits the second service object set, the server may obtain the second service object set, and obtain the detailed description information corresponding to each historical service object in the second service object set, and the user portrait data corresponding to the target service object, input the detailed description information into the recommendation model, may obtain the object feature vector corresponding to each historical service object in the second service object set, input the user portrait data into the recommendation model, may obtain the user feature vector corresponding to the target user, perform an inner product on the object feature vector and the user feature vector, may obtain the recommended evaluation value corresponding to each historical service object in the second service object set, and use the historical service object corresponding to the maximum recommended evaluation value as the target service object for recommendation, where the object feature vector and the user feature vector are obtained based on two branches of the recommendation model, for a specific obtaining process, reference may be made to the description of step S103 in the embodiment corresponding to fig. 3, which is not described herein again.

Step S210, acquiring an object recommendation template, fusing the object recommendation template and the target service object, generating recommendation information for the target user, and displaying the recommendation information to the target user;

specifically, the server determines that the target service object aims at recommending the target user, so that the server can obtain an object recommendation template in an information publishing platform corresponding to the target user, fuse the object recommendation template with the target service object to obtain recommendation information for the target user, send the recommendation information to the information publishing platform, and display the recommendation information on a display page of the information publishing platform. The recommendation information may be a dynamic commodity advertisement, and the dynamic commodity advertisement may need to be ranked and competed with other advertisements in the process of being displayed to the user, and the dynamic commodity advertisement can be displayed in the information publishing platform after being won. The information publishing platform may include, but is not limited to: e-commerce platform, news information platform, and social platform.

Step S211, recording the real-time behavior data of the recommendation information based on a public gateway interface, and uploading the real-time behavior data to a log system;

specifically, after the recommendation information for the target user is displayed, the target user can click and browse the recommendation information, and after the target user clicks and browses the recommendation information, the server can record real-time display data and real-time click data (the real-time display data and the real-time click data are collectively referred to as real-time behavior data) of the recommendation information based on the public gateway interface and upload the real-time behavior data to the log system, wherein the log system can record the real-time behavior data of all recommendation information associated with the object library.

Optionally, the log system may be a blockchain service system, that is, the real-time behavior data may be uploaded to a blockchain network, the real-time behavior data of all recommendation information associated with the object library is recorded in the service system of the blockchain network, the data recorded in the blockchain service system may not be tampered, and the security of the data may be improved.

The Blockchain (Blockchain) is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism and an encryption algorithm. The block chain is essentially a decentralized database, which is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, and is used for verifying the validity (anti-counterfeiting) of the information and generating a next block. The blockchain may include a blockchain underlying platform, a platform product services layer, and an application services layer.

The block chain underlying platform can comprise processing modules such as user management, basic service, intelligent contract and operation monitoring. The user management module is responsible for identity information management of all blockchain participants, and comprises public and private key generation maintenance (account management), key management, user real identity and blockchain address corresponding relation maintenance (authority management) and the like, and under the authorization condition, the user management module supervises and audits the transaction condition of certain real identities and provides rule configuration (wind control audit) of risk control; the basic service module is deployed on all block chain node equipment and used for verifying the validity of the service request, recording the service request to storage after consensus on the valid request is completed, for a new service request, the basic service firstly performs interface adaptation analysis and authentication processing (interface adaptation), then encrypts service information (consensus management) through a consensus algorithm, transmits the service information to a shared account (network communication) completely and consistently after encryption, and performs recording and storage; the intelligent contract module is responsible for registering and issuing contracts, triggering the contracts and executing the contracts, developers can define contract logics through a certain programming language, issue the contract logics to a block chain (contract registration), call keys or other event triggering and executing according to the logics of contract clauses, complete the contract logics and simultaneously provide the function of upgrading and canceling the contracts; the operation monitoring module is mainly responsible for deployment, configuration modification, contract setting, cloud adaptation in the product release process and visual output of real-time states in product operation, such as: alarm, monitoring network conditions, monitoring node equipment health status, and the like.

Step S212, updating historical behavior parameters corresponding to all historical behavior objects in the object library based on the real-time behavior data on the log system.

Specifically, according to the real-time behavior data on the log system, data updating can be performed on a corresponding service object in the object library, for example, a target service object in the recommendation information is a commodity 1, after the recommendation information is displayed on the information platform, a target user clicks and browses the recommendation information, display data and click data generated by the recommendation information can be stored in the log system, and then the server can update historical behavior data of the commodity 1 based on the data stored in the log system, for example, the display amount corresponding to the commodity 1 is increased by 1 time, and the click amount corresponding to the commodity 1 is increased by one time.

Please refer to fig. 10, which is a schematic diagram of a product recommendation system according to an embodiment of the present invention. The scheme provided by the embodiment of the application is applied to a commodity recommendation scene, and an overall recommendation system is shown in fig. 10 and can include: the method comprises four links of offline commodity recommendation 50a, online commodity triggering, sampling, sequencing 50b, dynamic commodity advertisement exposure 50c and commodity data backflow 50 d. The offline commodity recommendation 50a refers to a process in which the server generates a first business object set and a second business object set based on an object library, that is, a process in which N/2 historical commodities and N/2 non-displayed commodities included in the first business object set are selected from the commodity library, and M historical commodities are selected from the commodity library, and a specific implementation process may refer to descriptions in steps S201 to S205; the online triggering, sampling and sorting 50b refers to a process of selecting a target commodity for recommendation from the first service object set or the second service object set by the server, and includes a prediction process of an estimated behavior parameter, a recommendation process of a recommended evaluation value, and the like, and the specific implementation process may refer to the description in step S206-step S209; the dynamic commodity advertisement exposure 50c is a process in which the server fuses the target commodity and the commodity recommendation template, assembles the target commodity and the commodity recommendation template into a dynamic commodity advertisement and displays the dynamic commodity advertisement, and the specific implementation process can be described in step S210; the commodity data reflux 50d means that the server records real-time display data and real-time click data of commodities based on a public gateway interface, updates the display amount and click amount of the commodities based on the recorded real-time display data and real-time click data, and transmits the updated data to the offline commodity recommendation 50a to form a closed-loop commodity recommendation updating system.

Please refer to fig. 11, which is a schematic diagram of another data recommendation scenario provided in the embodiment of the present application. Taking a dynamic commodity advertisement scene as an example, the data recommendation scheme provided in the embodiment of the present application is explained, as shown in fig. 11, in a commodity recommendation setting page of the terminal device 60a, an advertisement delivery form of a dynamic commodity advertisement (e.g., a single commodity advertisement delivery form, a multi-commodity advertisement delivery form, etc.), a commodity series (e.g., a snack beverage series, a daily article series, etc.), a recommendation form (e.g., a re-marketing recommendation form, a pull-up recommendation form, etc.) may be selected, where the re-marketing recommendation form is a commodity recommendation form based on user behavior data, and the pull-up recommendation form is a commodity recommendation form based on advertisement platform free data without user behavior data. In the advertisement delivery stage, after the user selects the delivery mode as a multi-commodity advertisement, the commodity library is a snack beverage commodity library, and the recommendation mode is a pull-up recommendation 60b, the terminal device 60a may notify the selected recommendation mode and commodity library to the server 10a, and the server 10a may obtain a commodity library 60c (i.e., an object library) based on the received information, where the commodity library 60c includes already displayed snack beverage commodities (i.e., already displayed business objects) and not yet displayed snack beverage commodities (i.e., not yet displayed business objects). The server 10a may perform statistics on the display amount and the click amount of all the commodities included in the commodity library 60c based on a log system to obtain a behavior parameter table 60d corresponding to the commodity library 60c, where the behavior parameter table 60d includes the click amount and the display amount corresponding to each commodity in the commodity library 60 c. The server 10a may obtain a first commodity set 60e (i.e., a first business object set) and a second commodity set 60f (i.e., a second business object set) from the commodity library 60c based on the click quantity and the display quantity.

The server may set hit probabilities for the first commodity set 60e and the second commodity set 60f, respectively, and determine to which commodity set the target commodity for recommendation belongs based on the hit probabilities. If the trigger request hits in the first set of items 60e, the server 10a may obtain a stored knowledge graph 60g, where the knowledge graph 60g includes an association relationship between detailed description information of all items in the item library 60 c. The detailed description information corresponding to each commodity in the first commodity set 60e can be obtained based on the knowledge graph 60g, and the detailed description information is subjected to data processing through a behavior prediction model to obtain an estimated behavior parameter corresponding to each commodity in the first commodity set 60 e; constructing a beta distribution for each commodity in the first commodity set 60e based on the estimated behavior parameters and the actual behavior parameters in the behavior parameter table 60d, obtaining a sampling reference value corresponding to each commodity in the first commodity set 60e according to the beta distribution, and selecting a commodity to be recommended from the first commodity set 60e based on the sampling reference value; the server 10a may acquire user profile data corresponding to a target user (an object of commodity recommendation), sort commodities to be recommended based on commodity characteristics corresponding to the commodities to be recommended and the user profile data corresponding to the target user, and select a target commodity for recommendation.

If the trigger request hits the second product set 60f, the server 10a may directly obtain user portrait data corresponding to the target user, sort all the products in the second product set 60f based on the product features corresponding to each product in the second product set 60f and the user portrait data corresponding to the target user, and further select a target product for recommendation.

Regardless of the target product selected from the first product set 60e or the target product selected from the second product set 60f, the server 10a needs to obtain a product recommendation template, fuse and assemble the product recommendation template and the target object to obtain a dynamic product advertisement (i.e., recommendation information), and recommend the dynamic product advertisement to a corresponding target user. Based on the above implementation process, different dynamic commercial advertisements may be generated for different users and displayed on the terminal devices corresponding to the users, such as the terminal device 60h corresponding to the user 1, the terminal device 60i corresponding to the user 2, and the terminal device 60j corresponding to the user 3.

In this embodiment, the candidate service objects for recommendation may be divided into a first service object set and a second service object set, where the first service object set is used to store new service objects to be explored (i.e., service objects that are not shown and service objects that have a small click-through rate), and the second service object set is used to store high-quality service objects (i.e., service objects that have been shown and have a large click-through rate). The first business object set and the second business object set can be selected according to the set probability aiming at the business trigger request, and the probability of selecting the second business object set is higher than that of selecting the first business object set, so that most recommendation opportunities are reserved for high-quality business objects, and the display opportunities are not wasted. The service objects in the first service object set can obtain estimated behavior parameters based on the knowledge graph and the behavior prediction model, beta distribution is built for each service object in the first service object set based on the estimated behavior parameters, random numbers, namely sampling reference values, are generated through the beta distribution, target service objects for recommendation are selected from the first service object set based on the sampling reference values, the existing behavior parameters of the service objects are fully considered, and the accuracy of service object recommendation can be improved.

Please refer to fig. 12, which is a schematic structural diagram of a data recommendation device according to an embodiment of the present application. As shown in fig. 12, the data recommendation apparatus 1 may include: the device comprises an acquisition module 11, a prediction module 12 and a selection module 13;

the acquisition module 11 is configured to acquire a first service object set and acquire a knowledge graph; the first service object set comprises basic description information and historical behavior parameters corresponding to service objects, and the knowledge graph comprises an incidence relation between detail description information corresponding to each historical service object in an object library;

the prediction module 12 is configured to obtain detail description information having an association relationship with the basic description information from the knowledge graph, use the detail description information as target detail description information corresponding to the service object, and predict an estimated behavior parameter corresponding to the service object based on the target detail description information;

and the selecting module 13 is configured to select a target business object for recommendation from the first business object set according to the historical behavior parameter and the estimated behavior parameter corresponding to the business object.

The specific functional implementation manners of the obtaining module 11, the predicting module 12, and the selecting module 13 may refer to steps S101 to S103 in the embodiment corresponding to fig. 3, which is not described herein again.

Referring to fig. 12, the data recommendation apparatus 1 may further include: the system comprises a set generation module 14, an object triggering module 15, a display module 16, a recording module 17 and an updating module 18;

a set generating module 14, configured to generate the first set of business objects and the second set of business objects based on the object library; the first business object set and the second business object set are both determined by historical behavior parameters corresponding to historical business objects in the object library;

an object triggering module 15, configured to execute the step of obtaining the first service object set and obtaining a knowledge graph if the service triggering request hits the first service object set;

the object triggering module 15 is further configured to, if the service triggering request hits the second service object set, obtain the second service object set, and select a target service object from the second service object set based on a recommended evaluation value corresponding to each historical service object in the second service object set;

the display module 16 is configured to obtain an object recommendation template, fuse the object recommendation template with the target service object, generate recommendation information for the target user, and display the recommendation information to the target user;

the recording module 17 is configured to record real-time behavior data of the recommendation information based on a public gateway interface, and upload the real-time behavior data to a log system;

and the updating module 18 is configured to update the historical behavior parameters corresponding to all the historical behavior objects in the object library based on the real-time behavior data on the log system.

For specific functional implementation manners of the set generating module 14, the object triggering module 15, the displaying module 16, the recording module 17, and the updating module 18, reference may be made to steps S201 to S205, and steps S209 to S211 in the embodiment corresponding to fig. 8, which are not described herein again.

Referring also to fig. 12, the prediction module 12 may include: an extraction unit 121, a detail determination unit 122, an estimated parameter determination unit 123;

an extracting unit 121, configured to extract a target keyword from the basic description information, and search the knowledge graph for an entity string matching the target keyword;

a detail determining unit 122, configured to determine, as target detail description information corresponding to the business object, detail description information associated with the entity character string in the knowledge graph;

the estimated parameter determining unit 123 is configured to input the target detail description information into a behavior prediction model, obtain tag information corresponding to the target detail description information based on the behavior prediction model, and determine a behavior parameter corresponding to the tag information as an estimated behavior parameter corresponding to the business object.

The specific functional implementation manners of the extracting unit 121, the detail determining unit 122, and the estimated parameter determining unit 123 may refer to step S102 in the embodiment corresponding to fig. 3, which is not described herein again.

Referring to fig. 12, the estimated behavior parameters include an estimated click rate and an estimated display amount;

the selection module 13 may include: a first parameter determining unit 131, a second parameter determining unit 132, a beta distribution constructing unit 133, and a sampling unit 134;

a first parameter determining unit 131, configured to superimpose the click rate in the historical behavior parameter corresponding to the service object and the estimated click rate to obtain a first parameter;

a second parameter determining unit 132, configured to superimpose the display amount and the estimated display amount in the historical behavior parameter corresponding to the service object, so as to obtain a second parameter;

a beta distribution constructing unit 133, configured to construct a beta distribution corresponding to the business object based on the first parameter and the second parameter;

and the sampling unit 134 is configured to determine a sampling reference value corresponding to the business object according to the beta distribution, and select a target business object for recommendation from the first business object set based on the sampling reference value.

For specific functional implementation manners of the first parameter determining unit 131, the second parameter determining unit 132, the beta distribution constructing unit 133, and the sampling unit 134, reference may be made to step S103 in the embodiment corresponding to fig. 3, which is not described herein again.

Referring also to fig. 12, the set generation module 14 may include: a parameter obtaining unit 141, a dividing unit 142, a first sequencing unit 143, a first selecting unit 144, a second selecting unit 145;

a parameter obtaining unit 141, configured to obtain a historical behavior parameter corresponding to each historical service object in the object library; the historical behavior parameters comprise click quantity and display quantity;

a dividing unit 142, configured to divide the historical service objects included in the object library based on the historical behavior parameters to obtain a first to-be-processed set and a second to-be-processed set;

a first sorting unit 143, configured to count click rates corresponding to the historical service objects according to the click rate and the display amount, and sort the historical service objects included in the first to-be-processed set and the second to-be-processed set based on the click rates;

a first selecting unit 144, configured to select M historical service objects from the sorted first set to be processed according to a sorting order, as the second service object set; m is a positive integer less than or equal to K, and K is the number of historical service objects contained in the first set to be processed;

a second selecting unit 145, configured to select N/2 historical business objects from the sorted second to-be-processed set according to a sorting order, select N/2 non-shown business objects from all non-shown business objects in the object library, and determine the N/2 historical business objects and the N/2 non-shown business objects as the first business object set; n is a positive integer less than or equal to I, and I is the sum of the number of the historical business objects contained in the second to-be-processed set and the number of all the business objects which are not shown.

For specific functional implementation manners of the parameter obtaining unit 141, the dividing unit 142, the first ordering unit 143, the first selecting unit 144, and the second selecting unit 145, reference may be made to step S201 to step S209 in the embodiment corresponding to fig. 8, which is not described herein again.

Referring to fig. 12, the sampling unit 134 may include: an object to be recommended determination subunit 1341, an information input subunit 1342;

a to-be-recommended object determining subunit 1341, configured to determine, according to the beta distribution, a sampling reference value corresponding to the service object, sort all service objects in the first service object set based on the sampling reference value, and select C to-be-recommended objects from all sorted service objects; c is a positive integer less than or equal to N, and N is the number of the business objects contained in the first business object set;

an information input subunit 1342, configured to sequentially input the target detail description information corresponding to the C objects to be recommended into a recommendation model, obtain recommendation evaluation values corresponding to the C objects to be recommended respectively based on the recommendation model, and determine a target service object for recommendation based on the recommendation evaluation values.

The specific functional implementation manners of the object to be recommended determining subunit 1341 and the information inputting subunit 1342 may refer to step S103 in the embodiment corresponding to fig. 3, which is not described herein again.

Referring to fig. 12, the second selecting unit 145 may include: a category determining subunit 1451, an average click rate determining subunit 1452, a click evaluation value determining subunit 1453, an object selecting subunit 1454;

a category determining subunit 1451, configured to obtain all non-shown service objects from the object library, and determine a first category corresponding to each non-shown service object; the first category comprises a second category;

an average click rate determining subunit 1452, configured to count a first average click rate corresponding to the object library, a second average click rate corresponding to the first category, and a third average click rate corresponding to the second category according to the click rate corresponding to each history object;

a click evaluation value determining subunit 1453, configured to determine, according to the first average click rate, the second average click rate, and the third average click rate, a click evaluation value corresponding to each non-displayed service object;

an object selecting subunit 1454, configured to sort all the non-shown business objects based on the click evaluation value, and select the N/2 non-shown business objects from the sorted all non-shown business objects according to a sorting order.

The specific functional implementation manners of the category determining subunit 1451, the average click rate determining subunit 1452, the click evaluation value determining subunit 1453, and the object selecting subunit 1454 may refer to step S205 in the embodiment corresponding to fig. 8, which is not described herein again.

Referring to fig. 12, the information input subunit 1342 includes: a first vector generation subunit 13421, a user portrait determination subunit 13422, a second vector generation subunit 13423, an operations subunit 13424, a target business object determination subunit 13425;

a first vector generation subunit 13421, configured to sequentially input the target detail description information corresponding to the C objects to be recommended into the recommendation model, and generate object feature vectors corresponding to the C objects to be recommended;

a user portrait determination subunit 13422, configured to determine target users corresponding to the C objects to be recommended, and acquire user portrait data of the target users;

a second vector generation subunit 13423, configured to input the user portrait data into the recommendation model, and generate a user feature vector corresponding to the user portrait data;

an operation subunit 13424, configured to perform inner product operation on the user feature vector and object feature vectors corresponding to the C objects to be recommended, respectively, to obtain recommendation evaluation values corresponding to the C objects to be recommended, respectively;

a target business object determining subunit 13425, configured to determine the object to be recommended having the largest recommendation evaluation value as the target business object for recommendation.

The specific functional implementation manners of the first vector generation subunit 13421, the user image determination subunit 13422, the second vector generation subunit 13423, the operation subunit 13424, and the target service object determination subunit 13425 may refer to step S103 in the embodiment corresponding to fig. 3, which is not described herein again.

Fig. 13 is a schematic structural diagram of a computer device according to an embodiment of the present application. As shown in fig. 13, the computer apparatus 1000 may include: the processor 1001, the network interface 1004, and the memory 1005, and the computer apparatus 1000 may further include: a user interface 1003, and at least one communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display) and a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a standard wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1004 may be a high-speed RAM memory or a non-volatile memory (e.g., at least one disk memory). The memory 1005 may optionally be at least one memory device located remotely from the processor 1001. As shown in fig. 13, a memory 1005, which is a kind of computer-readable storage medium, may include therein an operating system, a network communication module, a user interface module, and a device control application program.

In the computer device 1000 shown in fig. 13, the network interface 1004 may provide a network communication function; the user interface 1003 is an interface for providing a user with input; and the processor 1001 may be used to invoke a device control application stored in the memory 1005 to implement:

It should be understood that the computer device 1000 described in this embodiment of the application may perform the description of the data recommendation method in the embodiment corresponding to any one of fig. 3 and fig. 8, and may also perform the description of the data recommendation apparatus 1 in the embodiment corresponding to fig. 7, which is not described herein again. In addition, the beneficial effects of the same method are not described in detail.

Further, here, it is to be noted that: an embodiment of the present application further provides a computer-readable storage medium, where a computer program executed by the aforementioned data recommendation apparatus 1 is stored in the computer-readable storage medium, and the computer program includes program instructions, and when the processor executes the program instructions, the description of the data recommendation method in any one of the embodiments corresponding to fig. 3 and fig. 8 can be executed, so that details are not repeated here. In addition, the beneficial effects of the same method are not described in detail. For technical details not disclosed in embodiments of the computer-readable storage medium referred to in the present application, reference is made to the description of embodiments of the method of the present application.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.

The above disclosure is only for the purpose of illustrating the preferred embodiments of the present application and is not to be construed as limiting the scope of the present application, so that the present application is not limited thereto, and all equivalent variations and modifications can be made to the present application.

Claims

1. A method for recommending data, comprising:

2. The method of claim 1, further comprising:

3. The method of claim 2, wherein generating the first set of business objects and the second set of business objects based on the object library comprises:

4. The method of claim 3, wherein the selecting N/2 non-exposed business objects from all non-exposed business objects in the object library comprises:

5. The method according to claim 1, wherein the obtaining of detail description information having an association relationship with the basic description information from the knowledge graph as target detail description information corresponding to the business object, and predicting predicted behavior parameters corresponding to the business object based on the target detail description information comprises:

6. The method of claim 1, wherein the estimated behavior parameters comprise an estimated click volume and an estimated display volume;

7. The method of claim 6, wherein the determining a sampling reference value corresponding to the business object according to the beta distribution, and selecting a target business object for recommendation from the first business object set based on the sampling reference value comprises:

8. The method according to claim 7, wherein the sequentially inputting the target detail description information corresponding to the C objects to be recommended into a recommendation model, acquiring recommendation evaluation values corresponding to the C objects to be recommended respectively based on the recommendation model, and determining a target business object for recommendation based on the recommendation evaluation values comprises:

9. The method according to any one of claims 1-8, further comprising:

10. The method of claim 9, further comprising:

11. A data recommendation device, comprising:

12. A computer arrangement comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program comprising program instructions which, when executed by a processor, perform the steps of the method according to any one of claims 1 to 10.