CN117972214A

CN117972214A - Knowledge-graph-based real-time film recommendation processing method and device

Info

Publication number: CN117972214A
Application number: CN202410221616.5A
Authority: CN
Inventors: 陈干文
Original assignee: Shenzhen Coocaa Network Technology Co Ltd
Current assignee: Shenzhen Coocaa Network Technology Co Ltd
Priority date: 2024-02-28
Filing date: 2024-02-28
Publication date: 2024-05-03

Abstract

The invention discloses a real-time film recommendation processing method and device based on a knowledge graph, comprising the following steps: acquiring user operation behavior data, and constructing a user portrait drawing based on the collected user operation behavior data; based on the constructed user portrait drawing, establishing association between the user portrait drawing and media resource data, and constructing a joint knowledge graph of the user; converting the entity and relation parameters of the combined knowledge graph into vectors by using a knowledge graph embedding method, and preparing data for subsequent entity and relation retrieval; based on the user operation behavior data, constructing real-time recommendation pool data of each user by using a real-time recommendation algorithm; based on the construction of the user image map, combining the real-time recommendation pool data of each user, filtering the content which is not interested by the user, and outputting a recommendation prediction result of the recommendation content updated in real time according to the user operation. The invention can recommend in real time: the recommended content can be updated in real time according to the user operation, and convenience is provided for the use of the user.

Description

Knowledge-graph-based real-time film recommendation processing method and device

Technical Field

The invention relates to the technical field of internet television film recommendation, in particular to a real-time film recommendation processing method and device based on a knowledge graph, an intelligent terminal and a storage medium.

Background

Along with the development of technology and the continuous improvement of living standard of people, the use of various intelligent terminals is becoming more and more popular.

The existing internet television film recommendation algorithm mostly adopts a traditional database to store media assets, and a recommendation model integrates part of media asset attributes to recommend, for example, a film watching record accumulated by long-term film watching of a user and a user label are combined to recommend according to the name, director, actors, label information and the like of a film. But these indicators do not reflect the actual interests and needs of the user. Therefore, the recommendation results obtained by the algorithms may be inaccurate, and the recommendation results cannot be recommended to the users or the recommendation results cannot be recommended to the users to be watched or not watched by many users, so that the user watching experience is affected; or the recommended content can not be updated in real time according to the operation of the user, so that the use of the user is inconvenient.

Accordingly, there is a need for improvement and development in the art.

Disclosure of Invention

The technical problem to be solved by the invention is to provide a real-time film recommendation processing method, device, intelligent terminal and storage medium based on a knowledge graph aiming at the defects in the prior art, wherein the real-time film recommendation processing method and device can be used for recommending in real time: the recommended content can be updated in real time according to the user operation, and convenience is provided for the use of the user.

The technical scheme adopted by the invention for solving the problems is as follows:

A real-time film recommendation processing method based on a knowledge graph comprises the following steps:

Acquiring user operation behavior data, and constructing a user portrait drawing based on the collected user operation behavior data;

based on the constructed user portrait drawing, establishing association between the user portrait drawing and media resource data, and constructing a joint knowledge graph of the user;

Converting the entity and relation parameters of the combined knowledge graph into vectors by using a knowledge graph embedding method, and preparing data for subsequent entity and relation retrieval;

based on the user operation behavior data, constructing real-time recommendation pool data of each user by using a real-time recommendation algorithm;

Based on the construction of the user image map, combining the real-time recommendation pool data of each user, filtering the content which is not interested by the user, and outputting a recommendation prediction result of the recommendation content updated in real time according to the user operation.

The method for recommending the real-time film based on the knowledge graph, wherein the step of constructing the real-time recommendation pool data of each user by using a real-time recommendation algorithm based on the user operation behavior data comprises the following steps:

cleaning and preprocessing the collected user operation behavior data;

extracting features for expressing user interests and preferences from the cleaned and preprocessed user operation behavior data,

Based on the extracted features for expressing the interests and preference of the user, selecting a corresponding real-time recommendation algorithm, and constructing a real-time recommendation pool according to the features and behavior data of the user.

The knowledge graph-based real-time film recommendation processing method is characterized in that the characteristics for expressing the interests and the preferences of the user comprise the favorite fields, the browsing frequency and the purchasing preference information of the user; the real-time recommendation algorithm comprises the following steps: collaborative filtering algorithm, content recommendation algorithm, deep learning algorithm.

The method for processing the real-time film recommendation based on the knowledge graph, wherein the steps of constructing the user image graph, combining real-time recommendation pool data of each user, filtering the content which is not interested by the user, and outputting a recommendation prediction result for updating the recommendation content in real time according to the user operation comprise the following steps:

combining the constructed user image with the real-time recommendation pool data of each user to form a comprehensive user information data set for the prediction and recommendation of a real-time recommendation algorithm;

Filtering out content that the user has viewed or marked as uninteresting according to the historical behavior data of the user, and filtering out films/features that the user has viewed and is uninterested;

Based on the user image of the user, the real-time recommendation pool data and the filtered content data, using a real-time recommendation algorithm to conduct recommendation prediction, and outputting recommendation content updated in real time according to user operation;

The recommended content updated in real time according to the user operation is presented.

The method for recommending and processing the real-time film based on the knowledge graph, wherein the step of acquiring the user operation behavior data and constructing the user portrait graph based on the collected user operation behavior data comprises the following steps:

acquiring user operation behavior data through embedded point collection; the user operation behavior data comprise browsing history, searching history, playing records, layout film exposure and layout film clicking operation behavior data;

User preference analysis is carried out based on the collected user operation behavior data, long-term portrait of the user and collaborative ordering data of each film are calculated, and a portrait drawing of the user is constructed; the user portrait graph comprises interest fields, preference labels and behavior habit information of the user.

The method for processing the real-time film recommendation based on the knowledge graph, wherein the step of establishing the association between the user portrait graph and the media resource data based on the constructed user portrait graph and constructing the joint knowledge graph of the user comprises the following steps:

based on the constructed user portrait drawing, the big model and the media analysis technology are adopted to enhance the view points and the classification characteristics, the association between the user portrait drawing and the media resource data is established, and the joint knowledge graph of the user is constructed;

Wherein the joint knowledge graph comprises: and the relationship among the user search, the film exposure, the film clicking and the film playing is established between the user and the film map metadata, and the relationship among the film, the film person, the roles, the labels and the user is described through the knowledge map record.

The method for processing the real-time film recommendation based on the knowledge graph, wherein the step of converting the entity and the relation parameter of the combined knowledge graph into the vector by using the knowledge graph embedding method comprises the following steps:

collecting the entity and relation data of the combined knowledge graph, wherein the relation data comprise the relation between the entities and the attribute information of the relation;

Preprocessing the collected entity and relationship data;

training the preprocessed entity and relation data by selecting a matched knowledge graph embedding model, and obtaining vector representation of the entity and relation through learning;

The entity and relation vector representation obtained by training is stored in a database or a file for data preparation for subsequent entity and relation retrieval.

A knowledge-graph-based real-time movie recommendation processing device, wherein the device comprises:

the user portrait module is used for acquiring user operation behavior data and constructing a user portrait based on the collected user operation behavior data;

The knowledge graph construction module is used for establishing association between the user portrait graph and the media resource data based on the constructed user portrait graph and constructing a joint knowledge graph of the user;

The vector conversion module is used for converting the entity and the relation parameter of the combined knowledge graph into a vector by using a knowledge graph embedding method and preparing data for subsequent entity and relation retrieval;

The real-time recommendation pool construction module is used for constructing real-time recommendation pool data of each user by using a real-time recommendation algorithm based on the user operation behavior data;

And the real-time recommendation module is used for combining the real-time recommendation pool data of each user based on the construction of the user image graph, filtering the content which is not interested by the user, and outputting a recommendation prediction result for updating the recommendation content in real time according to the user operation.

A smart terminal comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs comprising means for performing any of the methods.

A non-transitory computer readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform any one of the methods.

The invention has the beneficial effects that: the invention provides a real-time film recommendation processing method and device based on a knowledge graph, and provides a film recommendation algorithm designed based on factors such as a film graph, a film person graph, a label graph, a role graph, a user graph (comprising long-term images, time-sharing images and real-time images of users, and establishing relationships of user searching, film exposure, film clicking, film playing and the like between users and film graph metadata). Based on the current film watching record and preference of the user, the relation among various patterns of films, film persons, roles and labels, long-term portrait of the user and real-time portrait of the user, the film real-time recommendation based on the patterns can be provided, the recommendation content can be updated in real time according to the operation of the user, and convenience is provided for the user.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings may be obtained according to the drawings without inventive effort to those skilled in the art.

Fig. 1 is a flow chart of a real-time movie recommendation processing method based on a knowledge graph according to an embodiment of the present invention.

Fig. 2 is a flowchart of a real-time movie recommendation processing method based on a knowledge graph provided in embodiment 1 of the present invention.

Fig. 3 is a schematic block diagram of an implementation of the real-time movie recommendation processing method based on a knowledge graph according to embodiment 2 of the present invention.

Fig. 4 is a schematic block diagram of a real-time movie recommendation processing device based on a knowledge graph according to an embodiment of the present invention.

Fig. 5 is a schematic block diagram of an internal structure of an intelligent terminal according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more clear and clear, the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

It should be noted that, if directional indications (such as up, down, left, right, front, and rear … …) are included in the embodiments of the present invention, the directional indications are merely used to explain the relative positional relationship, movement conditions, etc. between the components in a specific posture (as shown in the drawings), and if the specific posture is changed, the directional indications are correspondingly changed.

For the existing internet television film recommendation algorithm in the prior art, most of the existing internet television film recommendation algorithm adopts a traditional database to store media assets, and a recommendation model integrates part of media asset attributes to recommend, for example, a film watching record accumulated by long-term film watching of a user and a user label are combined to recommend according to the name, director, actors, label information and the like of a film. But these indicators do not reflect the actual interests and needs of the user. Therefore, the recommendation results obtained by the algorithms may be inaccurate, and the recommendation results cannot be recommended to the users or the recommendation results cannot be recommended to the users to be watched or not watched by many users, so that the user watching experience is affected; or the recommended content can not be updated in real time according to the operation of the user, so that the use of the user is inconvenient.

Exemplary method

As shown in fig. 1, the real-time movie recommendation processing method based on a knowledge graph in embodiment 1 of the present invention includes the following steps:

Step S100, acquiring user operation behavior data, and constructing a user portrait drawing based on the collected user operation behavior data;

In the embodiment of the present invention, first, operation behavior data of a user needs to be collected, where the operation behavior data includes, but is not limited to, browsing records, searching records, clicking records, purchasing records, collecting records, and the like. Such data may be recorded and stored by a background system of the website or application. And then constructing a user portrait drawing based on the collected user operation behavior data. Wherein the user portrait includes a long-term portrait of the user, a time-sharing portrait, and a real-time portrait.

The user portrait drawing is constructed based on the collected user operation behavior data, and can be understood as constructing a user portrait about the detailed description and characteristics of the user by analyzing and integrating the collected user operation behavior data in the system.

In the embodiment of the invention, the method for constructing the user portrait graph has the following advantages:

personalized recommendation: by knowing the interests, preferences and behavior patterns of the user, more personalized recommended content can be provided for the user, and user experience and satisfaction are improved.

Accurate marketing: the user portrait graph is used for knowing the buying habit, preference and other information of the user, so that marketing activities can be performed pertinently, and the marketing effect and the conversion rate are improved.

And (3) product optimization: through analyzing the user portrait, the use condition and feedback of the user to the product can be known, the product design and functions can be improved, and the product quality can be improved.

User subdivision: based on the user portrait drawing, users can be subdivided, user groups of different categories are identified, and customized services and experiences are provided for the different user groups in a targeted manner.

Enhancing user viscosity: by knowing the user portraits, the user requirements can be better met, the user satisfaction can be improved, and the dependence and loyalty of the user on products or services can be enhanced.

In short, constructing the user image is helpful for understanding the user in depth, optimizing the user experience, and improving the competitiveness of the product or service, thereby promoting the development and growth of enterprises.

The step S100 specifically includes:

S101, acquiring user operation behavior data through embedded point collection; the user operation behavior data comprise browsing history, searching history, playing records, layout film exposure and layout film clicking operation behavior data;

Various operation behavior data of the user on the platform are collected in the step by embedding codes (buried points) in the application program or the website. Such data includes user browsing history, search history, play records, and operational behavior data such as video or content that the user sees (exposes) and clicks on the page.

Benefits of this data collection method include:

1) The collection of the user operation behavior data can help to analyze the behavior pattern of the user on the platform, know the interests and the preferences of the user, and provide basis for personalized recommendation and customized service.

2) Content optimization is convenient: by analyzing the browsing history, playing record and other data of the user, the content recommendation on the platform can be optimized, and the satisfaction degree and viscosity of the user on the content can be improved.

4) By collecting the operation behavior data of the user, the interaction mode of the user and the product can be known, and references are provided for product design and improvement of user experience.

5) The collected user operation behavior data can be used for constructing user portraits, so that a platform is helped to better know user groups, and more targeted services and experiences are provided.

The method is an effective data collection mode by collecting user operation behavior data through buried points, can help a platform to optimize products and services, improves user experience, enhances user viscosity, and supports business requirements such as personalized recommendation and accurate marketing.

S102, carrying out user preference analysis based on the collected user operation behavior data, calculating long-term portrait of the user and collaborative ordering data of each film, and constructing a portrait drawing of the user; the user portrait graph comprises interest fields, preference labels and behavior habit information of the user.

The user preference analysis is performed based on the collected user operation behavior data so as to calculate the long-term portrait of the user and the collaborative ordering data of each film, thereby constructing a portrait of the user. The user portrait map comprises information such as interest fields, preference labels, behavior habits and the like of the user.

Wherein the user preference analysis is: user preference analysis is performed by analyzing operational behavior data of a user on a platform, such as browsing history, search history, play records, etc., and knowing user preferences and preferences.

Wherein, the long-term portrait of the user is: by calculating the long-term portrait of the user, the user's interests, favorites and other information can be more comprehensively known, and the user's interests, favorites and other information can be more comprehensively understood than the behavior data in a short period. This helps to more accurately grasp the needs and behavior patterns of the user.

Wherein, the film collaborative ordering data is: according to the user operation behavior data, the preference degree or the correlation of the user to each film can be calculated, the collaborative ordering is further carried out, the films possibly interested by the user are recommended, and the accuracy of recommendation and the satisfaction degree of the user are improved.

Wherein, the user portrait graph is constructed by: integrating the information of interest fields, preference labels, behavior habits and the like of the users to construct a user portrait drawing. The user portrait graph can help the platform to better know the user group, and provides basis for personalized recommendation, accurate marketing and the like.

And the interest field, preference label and behavior habit information of the user: the user portrait graph comprises the interest field of the user, namely the theme or the field of interest of the user; preference tags, i.e., preference tags or keywords of a user for specific content; behavior habit information, namely behavior habits and habitual operations of the user on the platform, such as a time period browsed every day, common functions and the like.

The user operation behavior data are analyzed, the user portrait graph is constructed, the user requirements and behaviors can be better understood, more personalized and accurate service is provided, and therefore user experience and platform business effect are improved.

Step 200, based on the constructed user portrait drawing, establishing association between the user portrait drawing and media resource data, and constructing a joint knowledge graph of the user;

In the embodiment of the invention, based on the constructed user portrait drawing, the association between the user portrait drawing and the media resource data is established, and the joint knowledge graph of the user is constructed, specifically, the association analysis is carried out on the personal information, the hobbies and the behavior habit and other data of the user and the media resource data, so that a comprehensive user knowledge graph is constructed, the requirements and the behaviors of the user are better understood, and more personalized services and recommendations are provided for the user.

For example: when one e-commerce platform has a large amount of user data and commodity data, a user portrait drawing can be constructed by analyzing information such as purchase history, browsing behaviors, favorites and the like of a user. Meanwhile, the user portraits are associated with the attributes of the commodities, sales data and the like, and an associated knowledge graph between the user and the commodities can be constructed. In this way, the platform may better understand the user's purchasing preferences, providing personalized merchandise recommendations and customized services.

The method has the advantages that: user experience can be improved: by establishing the joint knowledge graph of the user, the user requirements and behaviors can be known more accurately, personalized and accurate recommendation service is provided for the user, and user experience is improved. And through analyzing the portrait and behavior data of the user, the positioning marketing can be more accurately performed, and the marketing effect and the conversion rate are improved.

And moreover, by analyzing the preference and the preference of the user, references can be provided for product design and function optimization, and the user requirements can be better met. And the association and integration among different data sources can be realized by constructing the joint knowledge graph of the user, so that the utilization efficiency and the value of the data are improved.

Further, the step S200 specifically includes:

Based on the constructed user portrait drawing, the big model and the media analysis technology are adopted to enhance the view points and the classification characteristics, the association between the user portrait drawing and the media resource data is established, and the joint knowledge graph of the user is constructed, as shown in figure 3;

In the refinement step, the interest points and classification characteristics of the user are enhanced by constructing the association of the user portrait drawing and the media resource data and utilizing a large model and a media analysis technology, so that the joint knowledge graph of the user is established. As shown in fig. 3, the knowledge graph includes long-term representation, time-sharing representation and real-time representation of the user, and is associated with film graph metadata, and records describing the relationship between the user and the film, such as searching, exposure, clicking and playing, and the relationship between the film, the movie person, the character, the label and the user.

For example: when an online video platform exists, the image graph of the user is constructed by analyzing data such as viewing history, preference, praise, comments and the like of the user. Meanwhile, the platform has a large amount of film data, including metadata such as film type, actor information, labels and the like.

By establishing a joint knowledge graph of the user, the following functions can be realized:

1) And recommending films related to user preference according to the portrait and the interest points of the user.

2) And analyzing the viewing preferences of the user in different time periods, and providing personalized recommendation services.

3) Updating the user portrait in real time, and adjusting the recommendation strategy according to the latest behavior and preference of the user.

4) And establishing the association between the user and the film, recording the searching, exposing, clicking, playing and other behaviors of the user on the film, and providing deeper user behavior analysis and personalized recommendation service for the platform.

5) Describing the relation among films, shadow persons, roles and labels, and providing more diversified recommendation and information presentation for users.

Through the user joint knowledge graph, the video platform can better understand user demands and behaviors, provide more personalized and accurate recommendation service, and improve user experience and service quality of the platform.

Step S300, converting the entity and the relation parameter of the combined knowledge graph into a vector by using a knowledge graph embedding method, and preparing data for subsequent entity and relation retrieval;

In this embodiment, the knowledge graph embedding method is a technology of representing entities and relationships in the knowledge graph as continuous vectors, and by learning these vectors, semantic relationships and associations between entities can be captured. The method can help better understand complex relationships between the entities and the relationships in the knowledge graph, and is further used for tasks such as retrieval, recommendation and the like of the entities and the relationships.

For example: there is a knowledge graph that includes entities such as movies, actors, directors, etc., and relationships between these entities, such as "director", etc. Knowledge graph embedding methods can be used to translate these entities and relationships into vector representations.

First, the entities and relationships in the knowledge graph are represented as unique IDs, such as ID 1 for movies, ID 2 for actors, ID 3 for the relationship "lead actor", etc. These IDs can then be converted into vector representations by knowledge graph embedding methods, such as converting movie ID 1 into a vector [0.2,0.5, -0.3], actor ID 2 into a vector [0.6, -0.1,0.4], and the relationship "lead actor" ID 3 into a vector [0.8,0.2, -0.5].

By learning these vectors, semantic relationships and associations between entities, such as relationships between actors and directors, similarity between movies, etc., can be captured. These vectorized representations may be used for subsequent entity and relationship retrieval tasks, such as finding movies associated with an actor; or given a movie, find a similar movie, etc.

Therefore, the knowledge graph embedding method of the embodiment can convert the entities and the relations in the knowledge graph into vector representations for subsequent tasks such as entity and relation retrieval, and the like, so that better understanding of the relations and semantics among the entities in the knowledge graph is facilitated.

Further, the step S300 specifically includes:

s301, collecting entity and relation data of a combined knowledge graph, wherein the relation and attribute information of the relation are included;

In this step, entity and relationship data in different data sources are collected, and a joint knowledge graph is established. An entity may be any particular thing or concept, a relationship describes a relationship between entities, and attribute information of the relationship provides more details of the relationship. For example, in a knowledge graph of a movie, the movie may be an entity, the actor may be an entity, and the "lead actor" may be a relationship, whose attribute information may include the actor's role in the movie, etc.

S302, preprocessing the collected entity and relationship data:

in the step, the collected data are subjected to cleaning, de-duplication, standardization and the like so as to ensure the quality and consistency of the data. The preprocessing work helps to improve the effect of subsequent model training.

S303, training the preprocessed entity and relation data by selecting a matched knowledge graph embedding model, and obtaining vector representation of the entity and relation through learning:

In this step, a suitable knowledge graph embedding model, such as TransE, transR, complEx, is selected, and the preprocessed entity and relationship data are trained to learn to obtain vector representations of the entity and relationship. These vector representations capture semantic relationships and associations between entities, facilitating subsequent entity, relationship retrieval tasks.

S304, storing the entity and the relation vector representation obtained by training in a database or a file for preparing data for subsequent entity and relation retrieval:

In this step, the entity and relationship vector representations obtained by training are stored in a database or file for use by subsequent entity and relationship retrieval tasks. These vector representations may be used for applications such as entity similarity calculation, relationship reasoning, knowledge graph completion, etc.

Through the refinement step, the invention can construct a combined knowledge graph with rich semantic information, and model entities and relations by utilizing a knowledge graph embedding method, thereby improving the expression capacity and application effect of the knowledge graph.

Step S400, based on the user operation behavior data, constructing real-time recommendation pool data of each user by using a real-time recommendation algorithm;

In the embodiment of the invention, based on the user operation behavior data, the real-time recommendation pool data of each user is constructed by using a real-time recommendation algorithm, and specifically, according to the real-time behavior data of the user on a platform, a personalized recommendation result for each user is generated by using the real-time recommendation algorithm. This process includes monitoring the user's behavior in real-time, analyzing the user's interests and preferences, and then applying this information to a recommendation algorithm to provide customized recommended content to the user.

For example, there is one e-commerce platform that wants to recommend goods for each user in real time. The platform may collect behavior data such as browsing, searching, purchasing, etc. of the user on the website, such as goods browsed by the user, links clicked, goods added to the shopping cart, etc. Based on these user operational behavior data, the platform may use real-time recommendation algorithms, such as collaborative filtering, content recommendation, deep learning, etc., to construct real-time recommendation pool data for each user.

Specifically, when the user browses on the platform, the real-time recommendation system can calculate interest preference of the user in real time according to current browsing behavior and historical behavior data of the user, then select commodities related to the user interests from the commodity library, and build a personalized recommendation pool of the user. The recommendation pool is updated continuously according to the real-time behavior of the user, so that the recommendation content is ensured to be consistent with the interest of the user, and the purchase conversion rate and the user experience of the user are improved.

By constructing the real-time recommendation pool based on the user operation behavior data, the platform can better meet the personalized requirements of the user, and the accuracy and the real-time performance of recommendation are improved, so that the participation degree and the loyalty of the user are increased.

Further, the step S400 specifically includes:

S401, cleaning and preprocessing the collected user operation behavior data;

In the embodiment of the invention, the operation behavior data of the user needs to be collected first, including but not limited to browsing records, searching records, clicking records, purchasing records, collecting records and the like. These data may be recorded and stored by a background system of the website or application; and then cleaning and preprocessing the collected user operation behavior data, including removing noise data, processing missing values, performing data format conversion and the like, so as to ensure the quality and the integrity of the data.

S402, extracting characteristics for expressing user interests and preferences from the cleaned and preprocessed user operation behavior data;

In this step, features are extracted from the user operation behavior data to describe interests and preferences of the user. These features may include information about the user's preference area, browsing frequency, purchasing preferences, etc.

Namely, the characteristics for expressing the interests and the preferences of the user comprise the favorite fields, browsing frequency and purchasing preference information of the user; the real-time recommendation algorithm comprises the following steps: collaborative filtering algorithm, content recommendation algorithm, deep learning algorithm.

S403, based on the extracted characteristics for expressing the interests and preferences of the user, selecting a corresponding real-time recommendation algorithm, and constructing a real-time recommendation pool according to the characteristics and behavior data of the user.

In this step, a suitable real-time recommendation algorithm, such as collaborative filtering, content recommendation, deep learning, etc., may be selected to construct a real-time recommendation pool according to the characteristics and behavior data of the user.

Specifically, model training may be performed using a selected real-time recommendation algorithm based on the user's operational behavior data and the extracted features to construct real-time recommendation pool data for each user. And then updating the recommendation pool data of each user in real time according to the latest operation behavior data of the user and the model training result, so as to ensure the accuracy and real-time performance of the recommended content.

Through the steps, based on the user operation behavior data, the real-time recommendation pool data of each user can be constructed by using a real-time recommendation algorithm, personalized, real-time and accurate recommendation contents are provided for the users, and user experience and platform value are improved.

And S500, based on the construction of the user image graph, combining the real-time recommendation pool data of each user, filtering the content which is not interested by the user, and outputting a recommendation prediction result for updating the recommendation content in real time according to the user operation.

In the embodiment of the invention, the real-time recommendation pool data of each user is combined based on the construction of the user image graph, the content which is not interested by the user is filtered, and the recommendation prediction result of the recommendation content is updated in real time according to the user operation is output: the method is characterized in that real-time updating is carried out by combining a recommendation algorithm according to personalized image information of a user, real-time recommendation pool data and feedback information of the user, so that a real-time recommendation prediction result of the user is generated. In this process, content that has been seen by the user or that the user expresses no interest may be filtered out to ensure accuracy and user experience of the recommended content.

For example, when a video platform recommends videos according to the interests of a user. Firstly, the platform constructs a user portrait drawing according to the behavior data of the user such as viewing history, praise, collection and the like, and knows the interests and the preferences of the user. And then, updating the recommended content of the user in real time according to the real-time operation behavior of the user and the personalized recommendation pool data. In this process, the platform filters out videos that the user has watched and videos that the user marks as not of interest, ensuring that the recommended content obtained by the user is novel and meets the user's interests.

According to the embodiment of the invention, the platform can generate more accurate real-time recommendation results by combining the user image map, the real-time recommendation pool data and filtering the content which is not interested by the user. The recommendation system can improve the watching experience of the user, reduce the situation that the user is recommended to uninteresting content, increase the residence time and the interaction rate of the user, and improve the user participation and the retention rate of the platform.

In general, based on the recommendation prediction results of the uninteresting contents of the user through combining the user portrait and the real-time recommendation pool data, more personalized and accurate recommendation contents can be provided for the user, and the user experience and the user participation of the platform are improved.

Further, the step S500 specifically includes:

s501, combining the constructed user image graph with real-time recommendation pool data of each user to form a comprehensive user information data set for prediction and recommendation of a real-time recommendation algorithm;

In the step, the constructed user image graph is combined with the real-time recommendation pool data of each user to form a comprehensive user information data set for predicting and recommending a real-time recommendation algorithm, specifically, static attribute information (user portrait) and dynamic behavior data (real-time recommendation pool data) of the user are combined together to form a more comprehensive and rich user information data set so as to support the real-time recommendation algorithm to conduct more accurate and personalized recommendation.

For example, when a music streaming platform wants to recommend music to a user. Firstly, the platform constructs a user portrait according to the registration information, preference labels, song listening history and other data of the user, and knows the music preference and preference type of the user. Meanwhile, the platform can construct real-time recommendation pool data of the user according to real-time operation behaviors of the user, such as recently heard songs, searched singers and the like, and reflects current interests and demands of the user.

The user pictorial and real time recommendation pool data are combined to form a comprehensive user information data set. The data set contains static attribute information and dynamic behavior data of the user, and can describe interests and preferences of the user more fully. Based on the comprehensive data set, the real-time recommendation algorithm can more accurately predict the next behavior of the user and recommend the music content which is more in line with the interests of the user.

By combining the user image and the real-time recommendation pool data, the platform can realize more personalized and accurate real-time recommendation, and the satisfaction and participation of the user are improved. Users can find favorite music more easily, and the user retention rate and user experience of the platform are improved.

S502, filtering out content which is watched or marked as uninteresting by the user according to the historical behavior data of the user, and filtering out films/features which are watched and uninterested by the user;

Namely, in the embodiment of the invention, the content which is not interested in the user is filtered: according to the historical behavior data of the user, the content which is watched or marked as uninteresting by the user is filtered, so that repeated recommendation to the user is avoided.

S503, based on the user image graph of the user, the real-time recommendation pool data and the filtered content data, using a real-time recommendation algorithm to conduct recommendation prediction, and outputting recommendation content updated in real time according to user operation;

In the embodiment of the invention, based on the user image graph, the real-time recommendation pool data and the filtered content data of the user, a real-time recommendation algorithm is used for recommendation prediction, the recommendation content updated in real time according to the user operation is output, specifically, the user interest and the user demand are predicted by combining the static attribute information, the dynamic behavior data and the filtered content data of the user through the real-time recommendation algorithm, and the real-time updated personalized recommendation content is output so as to meet the real-time requirement of the user and improve the user experience.

For example, the music streaming platform continues to be taken as an example. The platform constructs a comprehensive dataset based on the user profile of the user (e.g., music preferences, favorites types), real-time recommendation pool data (e.g., recently listened to songs, searched singers), and filtered content data (filtering out music that is not of interest to the user). The data set contains static attribute information, dynamic behavior data and filtered content data of the user, and can describe interests and preferences of the user more fully.

Based on the comprehensive data set, the real-time recommendation algorithm analyzes the behavior mode and preference trend of the user, predicts music content which the user is likely to be interested in, and outputs real-time updated recommended content to the user. The recommended contents are dynamically generated according to the real-time requirements and behaviors of the user, so that the personalized requirements of the user can be better met, and the satisfaction degree and participation degree of the user are improved.

According to the embodiment of the invention, based on the user image graph, the real-time recommendation pool data and the filtered content data of the user, the real-time recommendation algorithm is used for recommendation prediction, so that more personalized and accurate recommended content can be provided for the user, the user experience and the user participation degree are enhanced, and the user retention rate and the development potential of the platform are improved.

S504, presenting the recommendation content updated in real time according to the user operation.

The recommended content updated in real time according to the user operation is then presented.

Through the steps, the method and the device can output the recommendation prediction result of the recommendation content updated in real time according to the user operation based on the user image graph, the real-time recommendation pool data and the user historical behavior data, provide personalized, real-time and accurate recommendation content for the user, and improve user experience and platform value.

And can also realize the following functions:

1) The cooperative function: the recommended result is recommended according to the viewing preference trend of the full-platform user, and is suitable for global recommendation of homepage;

2) Item recommendation function: the recommendation according to the article relativity is suitable for making related recommendation of movie details;

3) Real-time recommendation function: the method is suitable for being used as a feed stream recommendation scene for updating recommendation contents in real time according to user operation;

4) Time-sharing recommendation function: the method is suitable for time-sharing personalized recommendation of fixed positions, for example, the watching time points of a plurality of users on weekdays and weekends can be different.

The invention is further illustrated by the following specific examples of application:

The real-time film recommendation processing method based on the knowledge graph can be applied to various television applications, such as intelligent televisions, set-top boxes, television boxes and the like. The collection of user behavior data and the algorithmic calculation may be performed via the internet. Various machine learning algorithms may be used for training and adjustment of the weight coefficients.

As shown in fig. 2 and fig. 3, the method for real-time movie recommendation processing based on knowledge graph provided in this embodiment of the present application includes the following steps:

step S11, starting and entering step S12;

step S12, collecting user behavior data and constructing a long-term and real-time portrait drawing of a user;

The user in the implementation refers to the user using the television, and the portrait contains basic attributes such as age, gender and the like of the user and various viewing preferences of the user;

s13, enhancing the viewpoint and classification characteristics by using a large model and a media analysis technology;

in this embodiment, after having the video medium, a media analysis algorithm may be used to extract information of the video content, and combine with reasoning capability of the large model to summarize more accurate information such as points, classification, etc., so as to perfect and enhance key attributes of the media resource;

Step S14, associating the user image and the media resource based on the user operation behavior data;

And S15, vectorizing the entities and parameters in the knowledge graph, and storing by milvus to prepare data for subsequent entity and relationship retrieval.

And S16, constructing a Real-Time recommendation pool of each user by using an Item-based Real Time algorithm based on user behaviors.

The method comprises the steps of constructing a Real-Time recommendation pool of each user by using an Item-based Real Time algorithm based on user behaviors, and the Real-Time recommendation pool of each user is constructed according to behavior data of the user by using the Item-based Real Time algorithm in a specific embodiment. This approach is mainly focused on dynamically generating personalized recommended content for the user based on the user's real-time behavior and preferences.

Specifically, an Item-based Real Time (Item-based Real Time) algorithm analyzes interests and preferences of users according to Real-Time behavior data of the users, such as recent clicking, searching, purchasing and the like, and builds a Real-Time recommendation pool of each user by combining similarity information among items. This real-time recommendation pool contains a list of items that the system believes are most likely to be of interest to the user.

By continuously updating the real-time recommendation pool of the user, the method and the device can timely reflect the interest change and behavior dynamics of the user, and provide more personalized recommendation content meeting real-time requirements for the user. The method can help to improve the accuracy of recommendation and the satisfaction degree of the user, and promote the user to participate in the interaction and use of the platform more.

In general, the real-time recommendation pool of each user is constructed by using a real-time recommendation algorithm based on user behaviors, so that the method is an effective personalized recommendation method, and timely updated personalized recommendation experience can be provided for the user according to the user's real-time behaviors and preferences.

And S17, based on the long-term portrait of the user, combining the real-time recommendation pool data of each user, eliminating films/features which are not interested in the user, predicting the preference, and outputting a recommendation prediction result.

Step S18, ending.

From the above, the present invention builds real-time recommendation pool data for each user based on user operation behavior data by using a real-time recommendation algorithm, which can bring the following benefits:

Personalized recommendation can be realized: by analyzing the operational behavior data of the users, the interests and preferences of the users can be better understood, thereby providing personalized recommended content for each user. Therefore, the satisfaction and the viscosity of the user can be improved, and the retention rate and the activity of the user are increased.

For example: when an e-commerce platform is used, according to behavior data such as browsing, collecting and purchasing of a user, a real-time recommendation algorithm is used for recommending commodities meeting the interests of the user, and the user can see more commodities meeting the favorites of the user when browsing a website, so that the purchasing will and experience are improved.

Real-time and accuracy of recommendation can be achieved: the recommendation content of the user can be updated in time through the real-time recommendation algorithm, the recommendation result is continuously adjusted according to the latest behavior data of the user, and the accuracy and the instantaneity of recommendation are guaranteed.

For example: when a news reading application updates recommended news content in real time according to behavior data such as clicking time and reading time of a user, the user is ensured to see the latest and most interesting news.

User engagement may be improved: the real-time recommendation pool data of each user is built through the real-time recommendation algorithm, so that the participation degree and interactivity of the users can be increased, and the liveness of the users on the platform is improved.

For example: when a social media platform updates the recommended friend circle dynamic and topics in real time according to the behavior data of praise, comment, sharing and the like of the user, the user is attracted to participate in interaction, and the user viscosity and loyalty are increased.

Therefore, the embodiment of the invention constructs the real-time recommendation pool data of each user by using the real-time recommendation algorithm based on the user operation behavior data, can provide personalized, real-time and accurate recommendation content, and improves user experience and participation, thereby increasing the user viscosity and value of the platform.

Exemplary apparatus

As shown in fig. 4, an embodiment of the present invention provides a real-time movie recommendation processing device based on a knowledge graph, where the device includes:

A user portrait module 310 for acquiring user operation behavior data and constructing a user portrait based on the collected user operation behavior data;

A knowledge graph construction module 320, configured to construct a joint knowledge graph of the user by associating the user graph with the media resource data based on the constructed user graph;

The vector conversion module 330 is configured to convert the entity and the relationship parameter of the joint knowledge graph into a vector by using the knowledge graph embedding method, and to prepare data for subsequent entity and relationship retrieval;

the real-time recommendation pool construction module 340 is configured to construct real-time recommendation pool data of each user using a real-time recommendation algorithm based on the user operation behavior data;

The real-time recommendation module 350 is configured to combine the real-time recommendation pool data of each user based on the user image graph, filter the content that the user has seen and is not interested in, and output a recommendation prediction result that the recommended content is updated in real time according to the user operation, as described above.

Based on the above embodiment, the present invention further provides an intelligent terminal, and a functional block diagram thereof may be shown in fig. 5. The intelligent terminal can be an intelligent television, a set top box, a television box and the like, and comprises a processor, a memory, a network interface, a display screen and a database which are connected through a system bus. The processor of the intelligent terminal is used for providing computing and control capabilities. The memory of the intelligent terminal comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the intelligent terminal is used for communicating with an external terminal through network connection. The computer program, when executed by the processor, implements a knowledge-graph-based real-time movie recommendation processing method. The database of the intelligent terminal is used for storing the collected user operation behavior data.

It will be appreciated by those skilled in the art that the schematic block diagram shown in fig. 5 is merely a block diagram of a portion of the structure associated with the present inventive arrangements and is not limiting of the smart terminal to which the present inventive arrangements are applied, and that a particular smart terminal may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In one embodiment, a smart terminal is provided that includes a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs comprising instructions for:

Based on the construction of the user image map, the real-time recommendation pool data of each user are combined, the content which is not interested in the user is filtered, and the recommendation prediction result of the recommendation content is updated in real time according to the user operation is output, and the method is specifically described above.

The step of constructing real-time recommendation pool data of each user by using a real-time recommendation algorithm based on the user operation behavior data comprises the following steps:

cleaning and preprocessing the collected user operation behavior data;

Wherein the characteristics for expressing the interests and the preferences of the user comprise the favorite fields, browsing frequency and purchasing preference information of the user; the real-time recommendation algorithm comprises the following steps: collaborative filtering algorithm, content recommendation algorithm, deep learning algorithm.

The step of outputting a recommendation prediction result for updating the recommended content in real time according to user operation comprises the following steps of:

The step of obtaining user operation behavior data and constructing a user portrait drawing based on the collected user operation behavior data comprises the following steps:

Wherein, based on the constructed user portrait drawing, the user portrait drawing and the media resource data are associated, and the step of constructing the joint knowledge graph of the user comprises the following steps:

The step of converting the entity and the relation parameter of the combined knowledge graph into the vector by using the knowledge graph embedding method comprises the following steps:

Preprocessing the collected entity and relationship data;

The trained entity and relationship vector representations are stored in a database or file for subsequent entity, relationship retrieval for data preparation, as described in detail above.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

In summary, the invention discloses a real-time film recommendation processing method, a device, an intelligent terminal and a storage medium based on a knowledge graph, which are used for acquiring user operation behavior data and constructing a user portrait graph based on the collected user operation behavior data; based on the constructed user portrait drawing, establishing association between the user portrait drawing and media resource data, and constructing a joint knowledge graph of the user; converting the entity and relation parameters of the combined knowledge graph into vectors by using a knowledge graph embedding method, and preparing data for subsequent entity and relation retrieval; based on the user operation behavior data, constructing real-time recommendation pool data of each user by using a real-time recommendation algorithm; based on the construction of the user image map, combining the real-time recommendation pool data of each user, filtering the content which is not interested by the user, and outputting a recommendation prediction result of the recommendation content updated in real time according to the user operation. The invention can recommend in real time: the recommended content can be updated in real time according to the user operation, and convenience is provided for the use of the user.

It is to be understood that the invention is not limited in its application to the examples described above, but is capable of modification and variation in light of the above teachings by those skilled in the art, and that all such modifications and variations are intended to be included within the scope of the appended claims.

Claims

1. The real-time film recommendation processing method based on the knowledge graph is characterized by comprising the following steps of:

2. The knowledge-graph-based real-time movie recommendation processing method according to claim 1, wherein the step of constructing real-time recommendation pool data for each user using a real-time recommendation algorithm based on the user operation behavior data comprises:

cleaning and preprocessing the collected user operation behavior data;

3. The knowledge-based real-time movie recommendation processing method as claimed in claim 2, wherein said characteristics for expressing user interests and preferences include user's preference field, browsing frequency, purchasing preference information; the real-time recommendation algorithm comprises the following steps: collaborative filtering algorithm, content recommendation algorithm, deep learning algorithm.

4. The knowledge-graph-based real-time movie recommendation processing method as claimed in claim 1, wherein the step of outputting a recommendation prediction result for updating the recommendation content in real time according to the user operation by combining the real-time recommendation pool data of each user based on the construction of the user image graph and filtering the content which is not interesting to the user, comprises:

5. The knowledge-graph-based real-time movie recommendation processing method as claimed in claim 1, wherein the step of acquiring user operation behavior data and constructing a user portrait graph based on the collected user operation behavior data includes:

6. The knowledge-graph-based real-time movie recommendation processing method as claimed in claim 1, wherein the step of constructing a joint knowledge graph of the user by associating the user graph with the media resource data based on the constructed user graph comprises:

7. The knowledge-graph-based real-time movie recommendation processing method according to claim 1, wherein the step of converting the entities and the relationship parameters of the joint knowledge graph into vectors using a knowledge-graph embedding method comprises:

Preprocessing the collected entity and relationship data;

8. A knowledge-graph-based real-time movie recommendation processing device, the device comprising:

9. An intelligent terminal comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs comprising instructions for performing the method of any of claims 1-7.

10. A non-transitory computer readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the method of any one of claims 1-7.