Content recommendation method and device and related equipment

Abstract

The invention provides a content recommendation method and device and related equipment. The method comprises the steps of obtaining real-time behavior data and historical behavior data of a user at a current time point in a service domain; based on the historical behavior data, generating an interest graph of the user, the interest graph being an interest representation of the user in the business domain; and based on the real-time behavior data and the interest graph, performing content recommendation on the user.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a content recommendation method, apparatus, and related equipment. Background Technology

[0002] Content recommendation involves analyzing user behavior, preferences, and needs, and then using recommendation algorithms to provide users with personalized content, such as service or product recommendations. The aim is to improve user satisfaction and service usage, thereby increasing operator revenue and user stickiness.

[0003] The core of content recommendation lies in data collection, processing, and analysis, achieved through techniques such as machine learning and data mining. Currently, the most commonly used content recommendation methods are content-based recommendation and collaborative filtering. Content-based recommendation analyzes product characteristics and user historical preferences to recommend products with similar characteristics to those the user has previously liked. Collaborative filtering, on the other hand, analyzes user behavioral data (such as purchase history and rating records) to discover similar users or products and then makes recommendations.

[0004] However, these content recommendation methods all rely on matching methods (such as user matching, product matching, etc.) to recommend content, resulting in relatively low quality of content recommendations. Summary of the Invention

[0005] This application provides a content recommendation method, apparatus, and related equipment, which can solve the technical problem that the quality of content recommendation is relatively low when using matching methods (such as user matching, product matching, etc.) in related technologies.

[0006] In a first aspect, embodiments of this application provide a content recommendation method, the method comprising:

[0007] Acquire real-time and historical behavioral data of users in the business domain at the current point in time;

[0008] Based on the historical behavior data, an interest map of the user is generated, which represents the user's interests in the business domain.

[0009] Based on the real-time behavioral data and the interest graph, content recommendations are made to the user.

[0010] Secondly, embodiments of this application provide a content recommendation device, the device comprising:

[0011] The first acquisition module is used to acquire real-time and historical behavior data of users in the business domain at the current point in time.

[0012] The generation module is used to generate the user's interest graph based on the historical behavior data, wherein the interest graph represents the user's interests in the business domain;

[0013] The recommendation module is used to recommend content to the user based on the real-time behavior data and the interest graph.

[0014] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the content recommendation method as described in the first aspect.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the content recommendation method as described in the first aspect.

[0016] Fifthly, embodiments of this application provide a computer program product including computer instructions that, when executed by a processor, implement the steps of the content recommendation method as described in the first aspect.

[0017] In this embodiment, by combining real-time and historical behavioral data of users in the business domain at the current time, and constructing an interest graph of users based on the historical behavioral data, and then combining the interest graph with the real-time behavioral data, content recommendations are made to users. Compared with content recommendations through matching methods (such as user matching, product matching, etc.), this approach can guide users to discover new needs, potentially recommending new products that differ from the services the user has previously subscribed to, and is not affected by the sparsity of user behavioral data, thus significantly improving recommendation accuracy and user satisfaction. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a content recommendation method provided in an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of the user interest mining process in an embodiment of this application;

[0021] Figure 3 This is a flowchart illustrating a specific example of a content recommendation method provided in an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of the structure of a content recommendation device provided in an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Content recommendation, such as service and product recommendations, analyzes user behavior, preferences, and needs, and uses recommendation algorithms to provide personalized service or product recommendations. Its aim is to improve user satisfaction and service usage, thereby increasing operator revenue and user stickiness.

[0026] The core of content recommendation systems lies in data collection, processing, and analysis, achieved through technologies such as machine learning and data mining. Currently, the most commonly used service product recommendations are content-based recommendation methods and collaborative filtering recommendations.

[0027] Content-based recommendation methods analyze product features and user history preferences to recommend products similar to those the user has previously liked. The core logic is as follows: Features are extracted from the user's past services and products, such as package type and data usage. These historical preference features are matched with the features of currently recommended products to find those with high similarity. Finally, products with the highest similarity are recommended to the user based on their similarity ranking. This recommendation method is intuitive and easy to understand, and the results are highly interpretable. It effectively utilizes users' historical service subscription data.

[0028] Collaborative filtering recommendation systems analyze user behavior data (such as purchase history, rating history, etc.) to discover similar users or similar products and then make recommendations. Its core logic is as follows: By calculating the similarity between users (e.g., the number of products purchased together), it finds users similar to the current user and then recommends products liked by these similar users. By calculating the similarity between products (e.g., the number of times they were jointly purchased), it finds products similar to those in the user's historical purchases and then makes recommendations. Based on similarity ranking, it recommends the products or services with the highest similarity to the user. This recommendation method does not require feature extraction from products and has wide applicability.

[0029] However, content-based recommendation methods lack exploratory capabilities and cannot recommend new products that differ from users' historical subscriptions. New users lack historical behavioral data, making effective recommendations difficult. Furthermore, they require large amounts of labeled data to extract product features. They are also prone to becoming trapped in "information cocoons," failing to guide users to discover new needs.

[0030] For collaborative filtering recommendations, there is a lack of sufficient data to support new users and new products, resulting in poor recommendation performance. When user behavior data is sparse, similarity calculation is inaccurate, affecting the quality of recommendations.

[0031] Based on this, the embodiments of this application provide a content recommendation method, which aims to solve the above-mentioned technical problems and improve the quality of content recommendation.

[0032] See Figure 1 , Figure 1 This is a flowchart of a content recommendation method provided in an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0033] Step 101: Obtain real-time and historical behavior data of users in the business domain at the current point in time;

[0034] Step 102: Based on the historical behavior data, generate the user's interest graph, which represents the user's interests in the business domain;

[0035] Step 103: Based on the real-time behavior data and the interest graph, recommend content to the user.

[0036] In step 101, the service domain refers to the operator's B-domain data. In the telecommunications industry, the B-domain (i.e., the service domain) refers to the data domain involved in the Business Support System (BSS). B-domain data mainly includes user data and service data, covering user consumption habits, terminal information, average revenue per user (ARPU) grouping, service content, and service target audience, etc.

[0037] B-domain data originates from multiple systems and platforms, including:

[0038] Billing system: Records users' consumption behavior and cost information.

[0039] Customer Relationship Management (CRM): Manages customer information and service processes.

[0040] Business analysis system: Analyzes business operations and provides decision support.

[0041] Value-added services integrated operation platform: manages and operates various value-added services.

[0042] B-domain data has wide applications in the communications industry, mainly in the following aspects:

[0043] Business support: By building a business support system, we ensure that telecommunications operators can support their business operations normally.

[0044] Customer Management: Improve customer satisfaction and loyalty through a CRM system.

[0045] Marketing: Utilize user consumption habits and business preference data to conduct precise marketing and promotion.

[0046] Business Analysis: By analyzing dynamic communication consumption data and static user base data, we can optimize business strategies and operational efficiency.

[0047] B-domain data is crucial for operators because it directly impacts the smooth operation of their business and customer satisfaction. By effectively managing and utilizing B-domain data, operators can improve service quality, enhance market competitiveness, and achieve sustained business growth.

[0048] It can retrieve real-time and historical behavioral data of users in domain B at the current point in time. The time point can be measured in hours, days, or other time units such as months; there is no specific limitation here. When measured in days, the current point in time refers to today, that is, it retrieves the user's real-time behavioral data for today and historical behavioral data up to today from domain B.

[0049] In some embodiments, historical behavioral data, including operator B-domain data, can be acquired for constructing a user's interest graph. This historical behavioral data may include call logs, internet browsing history, and application usage data. Call logs may include call frequency, call duration, and primary call recipients. Internet browsing history may include internet browsing frequency, internet content accessed, and internet traffic. Application usage data may include application usage frequency, application usage duration, and primary applications used.

[0050] For example, call logs include call frequency: an average of 5 calls per day; call duration: approximately 30 minutes per day; primary call recipients: family members and colleagues. Internet browsing logs include internet frequency: approximately 3 hours per day; internet content: primarily browsing technology news, social media, and online videos; data usage: approximately 1GB per day. App usage data includes app usage frequency: 5 different apps used per day; app usage duration: approximately 2 hours per day; primary apps: WeChat, Weibo, TikTok, news apps, and video streaming apps.

[0051] In some embodiments, real-time behavioral data and historical behavioral data may be of the same type. For example, real-time behavioral data may include a user's call logs, internet browsing history, and application usage data for the day.

[0052] In some embodiments, real-time and historical behavioral data can be preprocessed. For example, user behavior data can be cleaned to remove invalid and abnormal records. This includes deleting zero-duration calls from call logs and cleaning up duplicate browsing history. This ensures data accuracy and consistency and reduces the impact of noise on content recommendations.

[0053] User behavior data can be labeled. For example, sentiment analysis tools or natural language processing models such as pre-trained language models (Bidirectional Encoder Representations from Transformers, BERT) can be used to annotate text in SMS messages and browsing history with sentiment, extracting positive, negative, and neutral sentiment information. Enhancing the understanding of users' emotional states helps in context-aware content recommendation.

[0054] Next, the data can be normalized to ensure consistency in measurement across different sources (such as call duration and internet traffic), facilitating subsequent analysis. This improves the comparability between different data sources and prevents any single data feature from affecting the analysis results due to measurement issues.

[0055] In step 102, the user's interest graph, modeled from interest points and interest topics, is used to indicate the user's interest representation in the business domain.

[0056] In some embodiments, a user's interest map can be generated using a deep learning model based on historical behavioral data.

[0057] In some embodiments, such as Figure 2 As shown, preliminary interest extraction can be performed on the preprocessed historical behavior data, and an interest graph can be constructed to achieve interest mining through the fusion of multiple algorithms. Specifically, step 102 includes:

[0058] Step 1021: Using the first model, perform semantic analysis on the historical behavior data to obtain the user's points of interest;

[0059] Step 1022: Using the second model, perform topic modeling on the historical behavior data and the points of interest to obtain the user's interest topics;

[0060] Step 1023: Using the third model, construct a graph of the points of interest and the topics of interest to obtain the user's interest graph.

[0061] The multi-algorithm model is a fusion of a first model, a second model, and a third model. In some embodiments, the first model can be a BERT model. In some embodiments, the second model can be a Latent Dirichlet Allocation (LDA) model. In some embodiments, the third model can be a Graph Neural Network (GNN) model.

[0062] In other words, in some embodiments, the multi-algorithm model can be a fusion of BERT, LDA and GNN, which can make the constructed user interest map more comprehensive and accurate.

[0063] In step 1021, BERT can be used for deep semantic analysis to identify key points of interest for users in customer service data. BERT is a pre-trained language model based on the Transformer architecture that learns deep semantic representations from the context of text through a bidirectional encoder. It is pre-trained by masking parts of words and predicting those words, enabling it to understand the contextual relationships within sentences.

[0064] The BERT model can capture deep semantic relationships in text and process contextual information, making it excellent in various natural language processing tasks. BERT models can be used for semantic analysis on user feedback and complaint data collected through channels such as customer service hotlines and online customer service. These textual data can be processed using BERT models to extract key points of interest and sentiment information.

[0065] For example, a user's feedback "My 5G network is too slow" can be used to extract two points of interest: "5G network" and "slow speed," and negative sentiment can be identified. Collecting user posts and comments on social media platforms, forums, etc., and using the BERT model to perform semantic analysis on this text data, allows for the extraction of users' points of interest. For example, a user discussing "I am very interested in the newly launched smart home devices" on a forum can extract the point of interest "smart home devices," and positive sentiment can be identified.

[0066] Specifically, the BERT model can be used to perform semantic analysis on user call records, internet browsing records, and application usage data to extract points of interest and sentiment information.

[0067] There is a user's call log text. call (A) Internet browsing history text internet (A) and applications using data text. app (A) can obtain Interest Points BERT (A) = BERT(Text call (A)+Text internet (A)+Text app (A)), where Interest Points BERT (A) represents user A's interests.

[0068] Continuing with user A as an example, we will use the BERT model to perform deep semantic analysis on user A's internet browsing history and application usage data to extract points of interest. For example, we can extract the following points of interest from texts in technology news and social media using InterestPoints. BERT (A) = {Technology News, Social Media, Online Video} represents this.

[0069] Furthermore, in step 1022, the LDA model can be used to perform topic modeling on these points of interest and user behavior data, extracting the main interest topics. LDA is a generative statistical model used to discover topics from large-scale document sets. The LDA model assumes that each document is generated by a mixture of several topics, and each topic is generated by a distribution of several words. Through the LDA model, the implicit topic distribution in the document and the keywords in each topic can be obtained. The LDA model is good at topic modeling and can extract hidden topics from large amounts of text data, making it suitable for processing structured representations of text data. This is relevant for collecting the types of websites and content visited by users.

[0070] This text data can be input into an LDA model for topic modeling to identify the main topics accessed by users. For example, by analyzing users' visits to technology news websites and online video websites, the topics "technology news" and "online video" can be extracted. This also includes collecting information on the content and type of advertisements clicked by users.

[0071] Using the LDA model to perform thematic modeling on ad click data, we can extract which ad themes users are interested in. For example, by analyzing user clicks on technology product ads and travel ads, we can extract two themes: "technology products" and "travel".

[0072] By using topic modeling with LDA models, we can extract the main interest topics from user behavior data, provide structured interest representations, and facilitate further interest graph construction and personalized recommendations.

[0073] LDA (Local Theory Analysis) models are used to perform topic modeling on user call logs, internet browsing history, and application usage data to extract key topics of interest. For example, given a user's internet browsing history text... internet (A) and applications using data text. app (A) can be used to obtain Topics LDA (A)=LDA(Text internet (A)+ Text app (A)) Among them, Topics LDA (A) represents the topics of interest for user A.

[0074] Continuing with user A as an example, we use the LDA model to perform topic modeling on user A's internet browsing history and application usage data to extract primary interest topics. For example, by performing topic modeling on user A's browsing content and application usage, we obtain the following topics, which are then used as Topics. LDA (A) = {Technology, Social, Entertainment} represents this.

[0075] In step 1023, a GNN model can be used to construct a graph from the extracted interest points and interest topics, forming a user interest graph. Let user A have their interest points and interest topics. The resulting InterestGraph can be obtained. GNN (A) = GNN(InterestPoints) BERT (A), Topics LDA (A)), where Interest Graph GNN (A) represents user A's interest graph.

[0076] In this embodiment, by combining the interest points and interest topics extracted from the BERT and LDA models, a user interest graph is constructed, which can provide a structured interest representation for personalized recommendations. Thus, by combining operator B-domain data with various advanced algorithms to construct a user interest model, the accuracy of the recommendation system and user satisfaction can be significantly improved.

[0077] Furthermore, by combining the BERT and LDA models, the deep semantic analysis of the BERT model improves the accuracy of interest point extraction, while the topic modeling of the LDA model makes interest representation more structured and clear. Combining multiple data types (textual and behavioral data) allows for a comprehensive understanding of user interests and needs. The extracted interest points and topics can be directly used to construct the user's interest graph, providing reliable data support for subsequent personalized and context-aware recommendations.

[0078] In step 103, in some embodiments, content recommendations for users can be made using a deep learning model based on real-time behavioral data and interest graphs.

[0079] In some embodiments, step 103 specifically includes:

[0080] The fourth model is used to perform feature processing on the interest graph to obtain the user's interest features;

[0081] Based on the real-time behavioral data and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content.

[0082] The first recommended content is recommended to the user.

[0083] In some embodiments, the fourth model may be a convolutional neural network (CNN) model, and in some embodiments, the fifth model may be a reinforcement learning (RL) model.

[0084] CNNs are used for interest graph processing and are suitable for processing image or graph-structured data. A CNN model can be used to process user A's interest graph and extract the user's interest features.

[0085] For example, if the input is user A's interest graph, it can be represented as graph-structured data using the Interest Graph method. GNN (A) indicates that by inputting it into a CNN model, we can obtain Interest Features(A) = CNN(Interest Graph). GNN (A)), where Interest Features(A) are the interest features of user A.

[0086] Among them, RL models are suitable for decision-making problems. RL models can be used to learn the best recommendation strategy so that the recommended content can maximize user satisfaction and engagement.

[0087] In some embodiments, real-time behavioral data and interest features can be input into an RL model to make content recommendation decisions, outputting a first recommended content and recommending it to the user. In this way, by combining CNN and RL models, operators can analyze user behavior and interests in real time, generate personalized recommended content, and improve user satisfaction and loyalty.

[0088] In some embodiments, after recommending the first recommended content to the user, the method further includes:

[0089] Obtain the user's feedback information regarding the first recommended content;

[0090] Based on the feedback information and the user's first emotional information, decision-driving information is determined, wherein the first emotional information is obtained by semantic analysis of the feedback information using a first model.

[0091] Based on the decision-driven information, the model parameters of the fifth model are adjusted.

[0092] The RL model architecture is as follows:

[0093] 1. State: Composed of the user's real-time behavioral data and interest characteristics, denoted by S. t ={Real-timeBehavior(A,t),Interest Features(A)} means that Real-time Behavior(A,t) is the real-time behavior data of user A at time t.

[0094] 2. Action: Recommended content or service, such as the first recommended content, for example, recommending a call plan, a data plan, a value-added service, or a piece of content.

[0095] 3. Reward: Based on user feedback to the first recommended content and the user's initial emotional information, such as clicks, purchases, or interactions. Let the reward function be: R t =f(User Feedback, Emotion) User The reward can be decision-driven information, which is used to drive the RL model to make content recommendation decisions.

[0096] This decision-driven information can be based on feedback information and the initial emotional information determined based on that feedback. User (This is determined.) In some embodiments, if the first emotional information is a positive emotion, its decision-driving information may be positive. In this case, the RL model can be driven to output the content recommended to the user at the next time point according to the same strategy.

[0097] In some embodiments, if the first emotional information is negative, its decision-driving information may be negative. In this case, the model parameters of the RL model can be adjusted based on the decision-driving information so that the RL model can adjust its decision output and accordingly adjust the content recommended to the user at the next time point.

[0098] In this way, content recommendation can be achieved based on user context awareness, thereby enabling personalized recommendations.

[0099] In some embodiments, before adjusting the model parameters of the fifth model based on the decision-driven information, the method further includes:

[0100] Based on the historical behavior data, the sixth model is used to predict the user's behavior pattern at the next time point.

[0101] The step of adjusting the model parameters of the fifth model based on the decision-driven information includes:

[0102] Based on the behavioral pattern and the decision-driven information, the model parameters of the fifth model are adjusted so that the content recommendation decision of the fifth model at the next time point conforms to the behavioral pattern.

[0103] In some embodiments, the sixth model can be a recurrent neural network (RNN) model.

[0104] In some embodiments, RNN, CNN and RL can be combined for context-aware content recommendation, which can utilize users' real-time behavioral data and interest graphs to generate personalized recommended content.

[0105] The real-time behavior data of user A can be represented by Real-time Behavior(A,t)={Call(A,t),Internet(A,t),AppUsage(A,t)}.

[0106] You can use Recommendation(A,t)=f(Real-time Behavior(A,t),Interest Graph) GNN (A) indicates the first recommended content.

[0107] Here, f is a recommendation function that combines RNN, CNN and RL to generate recommended content based on real-time user behavior data and interest graphs.

[0108] For example, consider the following data:

[0109] Real-time behavior data: User A's real-time behavior data at the current point in time, for example: Real-time Behavior(A,t)={Call(A,t)=3 times, Internet(A,t)=2 hours, AppUsage(A,t)=WeChat 30 minutes, Douyin 1 hour}.

[0110] Interest Graph: User A's interest graph, derived from interest points and interest topics, for example, Interest GraphGNN (A) = GNN(Interest Points) BERT (A), Topics LDA (A)).

[0111] Among them, RNN models are used for processing time-series behavioral data. RNN models are particularly suitable for processing time-series data, such as user behavior sequences. An RNN model can be used to model user A's historical behavioral data and capture their behavioral patterns.

[0112] For example, the input is: a sequence of historical behavior data of user A, such as historical behavior data of the past week, which is represented by Behavior Sequence(A)={Behavior(A,t−6),Behavior(A,t−5),…,Behavior(A,t)}.

[0113] Output: Predicted behavior of user A at the next time point, represented by Predicted Behavior(A,t+1)=RNN(Behavior Sequence(A)), which gives the user's behavior pattern at the next time point.

[0114] CNNs are used for interest graph processing and are suitable for processing image or graph-structured data. A CNN model can be used to process user A's interest graph and extract the user's interest features.

[0115] For example, if the input is user A's interest graph, it can be represented as graph-structured data using the Interest Graph method. GNN (A) indicates that by inputting it into a CNN model, we can obtain Interest Features(A) = CNN(Interest Graph). GNN (A)), where Interest Features(A) are the interest features of user A.

[0116] In this context, behavioral patterns refer to an individual's reactions and behaviors in a specific situation, influenced by personality traits, beliefs, and values. These behavioral patterns can serve as constraints for adjusting the parameters of the RL model. Correspondingly, based on these behavioral patterns and decision-driven information, the parameters of the RL model can be adjusted to ensure that the content recommendation decisions made by the RL model at the next time point align with the user's behavioral patterns.

[0117] In other words, the RL model can first determine decision-driving information based on interest features and real-time user behavior data, driving the RL model to make content recommendation decisions and obtain the first recommended content, which is then recommended to the user. Next, it can obtain user feedback on the first recommended content and perform semantic analysis on the feedback to obtain the user's initial sentiment information, determining further decision-driving information. If the decision-driving information indicates a relatively strong penalty, the model parameters of the RL model can be adjusted by combining behavioral patterns and decision-driving information, so that the RL model's content recommendation decisions at the next time point are aligned with the user's behavioral patterns.

[0118] In this way, content recommendations can be made based on user context awareness and predicted user behavior patterns, thus achieving personalized recommendations.

[0119] In this embodiment, by combining RNN, CNN and RL, the operator can analyze user A's behavioral patterns and interests in real time, generate personalized recommendation content, and improve user satisfaction and loyalty.

[0120] In some embodiments, content recommendation can be performed based on user context awareness of real-time behavioral data. Before obtaining the first recommended content by making content recommendation decisions using a fifth model based on the real-time behavioral data and the interest features, the method further includes:

[0121] Sentiment analysis is performed on the real-time behavior data to obtain the user's second sentiment information regarding the real-time behavior data;

[0122] The first recommended content is obtained by using the fifth model to make content recommendation decisions based on the real-time behavior data and the interest features, including:

[0123] Based on the real-time behavioral data, the second sentiment information, and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content.

[0124] In other words, RL models can combine real-time behavioral data, users' secondary emotional information about the real-time behavioral data, and interest features to make content recommendation decisions, thereby achieving context-aware recommendation.

[0125] The RL model needs to be trained before context-aware recommendation, and the training process can be as follows:

[0126] 1. Initialization: Initialize the parameters and Q-value (value function) of the RL model.

[0127] 2. Training process:

[0128] S1: Obtain the current state S at time t. tThis includes real-time behavioral data, user sentiment information about real-time behavioral data, and interest characteristics.

[0129] S2: Select an action A based on the current strategy. t For example, recommending a calling plan.

[0130] S3: Execute action A t And observe user feedback to calculate reward R. t .

[0131] S4: Update state S t+1 The value function is then updated using the Q-learning algorithm.

[0132] 3. Recommendation Strategy: The trained RL model can recommend content based on users' real-time behavior data, secondary sentiment information, and interest features. It can also dynamically adjust the recommendation strategy based on users' feedback on the recommended content, users' primary sentiment information, and predicted user behavior patterns to generate personalized recommended content.

[0133] The following example illustrates in detail the content recommendation method provided in the embodiments of this application.

[0134] Figure 3 This is a flowchart illustrating a specific example of a content recommendation method provided in an embodiment of this application, such as... Figure 3 As shown, it includes the following steps:

[0135] Step 301: Obtain real-time and historical behavioral data of users from the operator's B domain data;

[0136] Step 302: Perform data preprocessing on real-time and historical behavioral data;

[0137] Step 303: Combine the BERT model, LDA model and GNN model to perform interest mining on the preprocessed historical behavior data in order to construct the user's interest graph;

[0138] Step 304: Combine RNN, CNN and RL models to perform context-aware content recommendation based on preprocessed real-time behavioral data and interest graphs.

[0139] This embodiment combines operator B-domain data with various advanced algorithms to construct a user interest model (i.e., interest graph) and perform context-aware recommendations, significantly improving the accuracy and user satisfaction of the recommendation system. The fusion of BERT, LDA, and GNN models makes the user interest model more comprehensive and accurate, while the combination of RNN, CNN, and RL models ensures the real-time nature and personalization of recommended content, thus forming an innovative and efficient recommendation system.

[0140] This application primarily applies to a user service recommendation system for telecom operators. It involves constructing user interest graphs using multi-algorithm fusion (BERT + LDA + GNN) by combining operator B-domain data, and performing context-aware recommendations using multi-algorithm fusion (RNN + CNN + RL). Compared to traditional content-based and collaborative filtering recommendation schemes, BERT's deep semantic analysis and LDA's topic modeling accurately extract users' interests and emotional states, making the recommendation results more aligned with users' actual needs. With the widespread adoption of 5G networks and the increasing demand for personalized services, operators face significant challenges and opportunities to improve user satisfaction and service quality. This application allows operators to more accurately understand user needs, provide more attentive services, thereby increasing user stickiness and satisfaction, and enhancing market competitiveness, thus demonstrating broad application prospects.

[0141] See Figure 4 , Figure 4 This is a schematic diagram of the structure of a content recommendation device provided in an embodiment of this application, as shown below. Figure 4 As shown, the content recommendation device 400 includes:

[0142] The first acquisition module 401 is used to acquire real-time and historical behavior data of users in the business domain at the current time point;

[0143] The generation module 402 is used to generate the user's interest map based on the historical behavior data, wherein the interest map is a representation of the user's interests in the business domain;

[0144] The recommendation module 403 is used to recommend content to the user based on the real-time behavior data and the interest graph.

[0145] Optionally, the generation module 402 is specifically used for:

[0146] Using the first model, semantic analysis is performed on the historical behavior data to obtain the user's points of interest;

[0147] Using the second model, the historical behavior data and the points of interest are used to perform topic modeling to obtain the user's interest topics;

[0148] Using the third model, a graph is constructed from the points of interest and the topics of interest to obtain the user's interest graph.

[0149] Optionally, the recommendation module 403 is specifically used for:

[0150] The fourth model is used to perform feature processing on the interest graph to obtain the user's interest features;

[0151] Based on the real-time behavioral data and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content.

[0152] The first recommended content is recommended to the user.

[0153] Optionally, the device further includes:

[0154] The second acquisition module is used to acquire the user's feedback information regarding the first recommended content;

[0155] The determination module is used to determine decision-driven information based on the feedback information and the user's first sentiment information, wherein the first sentiment information is obtained by semantic analysis of the feedback information using a first model;

[0156] The driving module is used to adjust the model parameters of the fifth model based on the decision-driving information.

[0157] Optionally, the device further includes:

[0158] The prediction module is used to predict the user's behavior pattern at the next time point by using the sixth model based on the historical behavior data.

[0159] The driving module is specifically used to adjust the model parameters of the fifth model based on the behavioral pattern and the decision-driving information, so that the content recommendation decision of the fifth model at the next time point conforms to the behavioral pattern.

[0160] Optionally, the device further includes:

[0161] The analysis module is used to perform sentiment analysis on the real-time behavior data to obtain the user's second sentiment information on the real-time behavior data;

[0162] The recommendation module 403 is specifically used to make content recommendation decisions based on the real-time behavior data, the second sentiment information and the interest features, using the fifth model to obtain the first recommended content.

[0163] The content recommendation device 400 can implement all the processes implemented in the above-described content recommendation method embodiments and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0164] See Figure 5 The figure shows a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 5 As shown, the electronic device 500 includes: a processor 501, a memory 502, a user interface 503, and a bus interface 504.

[0165] Processor 501 is used to read the program from memory 502 and execute the following procedures:

[0166] Acquire real-time and historical behavioral data of users in the business domain at the current point in time;

[0167] Based on the historical behavior data, an interest map of the user is generated, which represents the user's interests in the business domain.

[0168] Based on the real-time behavioral data and the interest graph, content recommendations are made to the user.

[0169] exist Figure 5 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 501 and memory represented by memory 502 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 504 provides an interface. For different user devices, user interface 503 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0170] The processor 501 is responsible for managing the bus architecture and general processing, while the memory 502 can store the data used by the processor 501 when performing operations.

[0171] In some embodiments, the processor 501 is further configured to:

[0172] Using the first model, semantic analysis is performed on the historical behavior data to obtain the user's points of interest;

[0173] Using the second model, the historical behavior data and the points of interest are used to perform topic modeling to obtain the user's interest topics;

[0174] Using the third model, a graph is constructed from the points of interest and the topics of interest to obtain the user's interest graph.

[0175] In some embodiments, the processor 501 is further configured to:

[0176] The fourth model is used to perform feature processing on the interest graph to obtain the user's interest features;

[0177] Based on the real-time behavioral data and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content.

[0178] The first recommended content is recommended to the user.

[0179] In some embodiments, the processor 501 is further configured to:

[0180] Obtain the user's feedback information regarding the first recommended content;

[0181] Based on the feedback information and the user's first emotional information, decision-driving information is determined, wherein the first emotional information is obtained by semantic analysis of the feedback information using a first model.

[0182] Based on the decision-driven information, the model parameters of the fifth model are adjusted.

[0183] In some embodiments, the processor 501 is further configured to:

[0184] Based on the historical behavior data, the sixth model is used to predict the user's behavior pattern at the next time point.

[0185] Based on the behavioral pattern and the decision-driven information, the model parameters of the fifth model are adjusted so that the content recommendation decision of the fifth model at the next time point conforms to the behavioral pattern.

[0186] In some embodiments, the processor 501 is further configured to:

[0187] Sentiment analysis is performed on the real-time behavior data to obtain the user's second sentiment information regarding the real-time behavior data;

[0188] Based on the real-time behavioral data, the second sentiment information, and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content.

[0189] Preferably, the present invention also provides an electronic device 500, including a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the computer program is executed by the processor 501, it implements the various processes of the above-mentioned recommended method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0190] This invention also provides a readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the recommended method embodiments described above and achieves the same technical effects. To avoid repetition, it will not be described again here. The readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0191] This application also provides a computer program product, including computer instructions. When these computer instructions are executed by a processor, they implement the various processes of the above-mentioned recommended method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0192] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0193] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0195] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0196] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0197] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0198] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A content recommendation method, characterized in that, The method includes: Acquire real-time and historical behavioral data of users in the business domain at the current point in time; Based on the historical behavior data, an interest map of the user is generated, which represents the user's interests in the business domain. Based on the real-time behavioral data and the interest graph, content recommendations are made to the user.

2. The method according to claim 1, characterized in that, The step of generating the user's interest graph based on the historical behavior data includes: Using the first model, semantic analysis is performed on the historical behavior data to obtain the user's points of interest; Using the second model, the historical behavior data and the points of interest are used to perform topic modeling to obtain the user's interest topics; Using the third model, a graph is constructed from the points of interest and the topics of interest to obtain the user's interest graph.

3. The method according to claim 1, characterized in that, The content recommendation for the user based on the real-time behavior data and the interest graph includes: The fourth model is used to perform feature processing on the interest graph to obtain the user's interest features; Based on the real-time behavioral data and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content. The first recommended content is recommended to the user.

4. The method according to claim 3, characterized in that, After recommending the first recommended content to the user, the method further includes: Obtain the user's feedback information regarding the first recommended content; Based on the feedback information and the user's first emotional information, decision-driving information is determined, wherein the first emotional information is obtained by semantic analysis of the feedback information using a first model. Based on the decision-driven information, the model parameters of the fifth model are adjusted.

5. The method according to claim 4, characterized in that, Before adjusting the model parameters of the fifth model based on the decision-driven information, the method further includes: Based on the historical behavior data, the sixth model is used to predict the user's behavior pattern at the next time point. The step of adjusting the model parameters of the fifth model based on the decision-driven information includes: Based on the behavioral pattern and the decision-driven information, the model parameters of the fifth model are adjusted so that the content recommendation decision of the fifth model at the next time point conforms to the behavioral pattern.

6. The method according to claim 3, characterized in that, Before obtaining the first recommended content by making content recommendation decisions using the fifth model based on the real-time behavior data and the interest features, the method further includes: Sentiment analysis is performed on the real-time behavior data to obtain the user's second sentiment information regarding the real-time behavior data; The first recommended content is obtained by using the fifth model to make content recommendation decisions based on the real-time behavior data and the interest features, including: Based on the real-time behavioral data, the second sentiment information, and the interest features, the fifth model is used to make content recommendation decisions to obtain the first recommended content.

7. A content recommendation device, characterized in that, The device includes: The first acquisition module is used to acquire real-time and historical behavior data of users in the business domain at the current point in time. The generation module is used to generate the user's interest graph based on the historical behavior data, wherein the interest graph represents the user's interests in the business domain; The recommendation module is used to recommend content to the user based on the real-time behavior data and the interest graph.

8. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the content recommendation method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the content recommendation method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the content recommendation method as described in any one of claims 1 to 6.

Images