A commodity push method and system based on user preference analysis

Abstract

The present invention discloses a commodity push method and system based on user preference analysis, which relates to the technical field of e-commerce. It includes collecting multi-source data, performing sentiment analysis on user comments to generate user portraits and constructing a commodity knowledge graph, using a fuzzy Kano model to evaluate the importance of features and generating feature-sentiment pairs; constructing a user-commodity interaction graph based on the user portrait and the feature-sentiment pairs, and calculating the context-aware similarity between commodities according to the spatio-temporal information in the user-commodity interaction graph. By using sentiment analysis technology and the fuzzy Kano model, the present invention not only improves the accuracy of user portraits, but also refines the importance evaluation of commodity features through feature-sentiment pairs. By constructing a user-commodity interaction graph and combining context-aware similarity calculation, it not only considers the real-time location and environmental changes of users, but also can dynamically adjust the recommended content, significantly improving the personalization degree of the recommendation system and user satisfaction.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and particularly to a commodity push method and system based on user preference analysis. Background Art

[0002] In the context of the rapid development of digitalization and e-commerce, personalized recommendation systems have become a key technology for attracting users and enhancing user experience. The core goal of a recommendation system is to accurately predict users' interests and preferences in order to recommend the most suitable commodities or services to users. Early recommendation systems mainly relied on simple collaborative filtering techniques, which predicted new commodities or services that users might be interested in by analyzing users' historical behavior data. With the progress of technology, recommendation systems have gradually incorporated more complex algorithms, such as matrix factorization, deep learning, and natural language processing, making the recommendation results more accurate and personalized.

[0003] However, existing recommendation technologies still have some deficiencies. Many systems still rely on explicit user feedback, such as ratings and reviews, while ignoring the potential value of implicit feedback and failing to fully utilize users' emotional tendencies and context information, resulting in the lack of personalization and relevance of recommendation results in some cases, affecting the effectiveness of the recommendation system and users' ultimate satisfaction. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned existing commodity push methods and systems based on user preference analysis, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that many systems still rely on explicit user feedback, such as ratings and reviews, while ignoring the potential value of implicit feedback and failing to fully utilize users' emotional tendencies and context information, resulting in the lack of personalization and relevance of recommendation results in some cases, affecting the effectiveness of the recommendation system and users' ultimate satisfaction.

[0006] To solve the above technical problems, the present invention provides the following technical solution: A commodity push method based on user preference analysis, which includes collecting multi-source data, performing sentiment analysis on user comments to generate user portraits and constructing a commodity knowledge graph, using a fuzzy Kano model to evaluate the importance of features and generating feature - sentiment pairs; constructing a user - commodity interaction graph based on the user portrait and the feature - sentiment pairs, and calculating the context-aware similarity between commodities according to the spatio-temporal information in the user - commodity interaction graph; generating a recommendation list according to the context-aware similarity between commodities.

[0007] As a preferred solution of the commodity push method based on user preference analysis described in the present invention, wherein: collecting multi-source data, performing sentiment analysis on user comments to generate user portraits and constructing a commodity knowledge graph refers to collecting users' behavioral data, commodity data, and context data including geographical location, timestamp, and device information from websites, mobile applications, and social media, cleaning the collected data, and using NLP tools to perform word segmentation, stop word removal, and part-of-speech tagging on user comments;

[0008] Select key features in user behavior, including access frequency, purchase frequency, and social interaction frequency, use one-hot encoding and TF-IDF methods to convert behavioral features into vector representations and identify feature words in comments, integrate behavioral data with geographical location, timestamp, and device information, and attach context tags to each user behavior record;

[0009] Use the BERT model for sentiment analysis, input the feature words into the BERT model to obtain the feature vectors of the text, input the feature vectors output by the BERT model into the classification layer composed of a fully connected layer and a Softmax activation function, and output the probability of each sentiment category;

[0010] Calculate the sentiment intensity using a linear function in another fully connected layer:

[0011]

[0012] In the formula, I is the sentiment intensity, σ is the Sigmoid function, V i and b i are the weights and biases of the sentiment intensity layer, and T is the transpose operation;

[0013] Calculate the sentiment score for each comment according to the sentiment category probability value and sentiment intensity:

[0014] S l =P(r l )×I(r l ),

[0015] In the formula, S l is the sentiment score of the l-th user comment, P(r l ) is the sentiment category probability value of the l-th user comment, and I(r l ) is the sentiment intensity of the l-th user comment;

[0016] Calculate the overall sentiment score of each user, integrating the sentiment information of all comments:

[0017]

[0018] In the formula, S totalis the overall emotional score of the user, and N is the number of comments of the user;

[0019] Allocate emotional labels to users according to the total emotional score:

[0020] If S total ≥ 0.75, the emotional label is "positive";

[0021] If 0.75 > S total > 0.25, the emotional label is "neutral";

[0022] If S total ≤ 0.25, the emotional label is "negative";

[0023] Combine the behavioral feature vector with the emotional score to form a user profile and use the K-means clustering algorithm to divide users into different groups;

[0024] Extract features from the collected commodity data, standardize them, and then convert the attribute information into a feature vector A. Use one-hot encoding to represent the category and brand, and standardize the price and rating;

[0025] Identify the relationships between commodities through co-occurrence analysis and collaborative filtering techniques, and calculate the relationship strength PMI(x,y) between commodities using point mutual information:

[0026]

[0027] In the formula, P(x,y) is the probability that commodities x and y appear simultaneously, and P(x) and P(y) are the probabilities that commodity x and commodity y appear alone respectively;

[0028] Create a node for each commodity. The node attributes include the commodity feature vector A and the user emotional label, and define the edges between commodities according to the PMI value and the user's behavior pattern;

[0029] When new commodities are added and commodity information changes, update, recalculate the PMI values of relevant commodities, and update the corresponding edge weights and node information in the commodity knowledge graph.

[0030] As a preferred solution of the commodity push method based on user preference analysis described in the present invention, wherein: the use of the fuzzy Kano model to evaluate the feature importance and generate feature-emotion pairs means designing a questionnaire containing the extracted commodity features, using a Likert scale to ask users about their satisfaction and expectation levels for each commodity feature, using an online survey tool to collect user responses, performing coding conversion on the user responses, and converting the Likert scale scores into the numerical inputs required by the fuzzy set;

[0031] For satisfaction and expectation levels, three fuzzy sets are defined, including a high fuzzy set, a medium fuzzy set, and a low fuzzy set. Each fuzzy set is represented using a triangular membership function:

[0032]

[0033] where δ(q) is the membership function, a', b', and c' are the breakpoints for high, medium, and low satisfaction, respectively, and q is the score value;

[0034] Using fuzzy Karnaugh logic, the features are classified according to the membership degrees of the fuzzy sets of user satisfaction and expectation levels, and the importance index of each product feature is calculated using fuzzy AND:

[0035] G t =min(μ sat (f), μ exp (f)),

[0036] where G t is the importance index of product feature t, μ sat (f) and μ exp (f) are the fuzzy membership degrees of satisfaction and expectation, respectively;

[0037] Generate a comprehensive importance report for each product feature based on the feature importance index and sentiment score;

[0038] Combine the sentiment score and importance index of the feature to generate a list of feature - sentiment pairs for each product. Each feature - sentiment pair includes the feature name, sentiment score, and importance index.

[0039] As a preferred embodiment of the product push method based on user preference analysis according to the present invention, wherein: constructing a user - product interaction graph based on user portraits and feature - sentiment pairs means combining the user's behavior data with feature - sentiment pairs, and for each layer of user - product interaction, annotating the relevant product features and the user's sentiment score and importance index for the product features;

[0040] Create a node for each independent user, and the node attributes include the user's identification information. Create a node for each independent product, and the node attributes include the product ID, category, and brand. Map each group of the user's behaviors to the edges of the user - product interaction graph and use the context information as the edge attributes. Store the user - product interaction graph using a graph database.

[0041] As a preferred embodiment of the product push method based on user preference analysis according to the present invention, wherein: calculating the context - aware similarity between products according to the spatio - temporal information in the user - product interaction graph means using the GraphSAGE model to learn the embedding representation of products from the interaction graph. The embedding vector generation formula is:

[0042]

[0043] Wherein, is the embedding vector of node v in the k-th layer, is the embedding vector of node u in the (k-1)-th layer, Q(v) is the set of neighbor nodes of node v, A is the aggregation function, and W (k) is the weight matrix, and B (k) is the bias term in the k-th layer, and μ is the ReLU activation function;

[0044] Combining the embedding vectors and context attributes of the products, calculate the context-aware similarity between products:

[0045]

[0046] Wherein, Sim og is the similarity between product o and product g, h o and h g are the embedding vectors of the products, γ is the weight factor of the context, and O(o,g) is the context similarity;

[0047] According to the calculated similarity, recommend context-related similar products to the user.

[0048] As a preferred solution of the product push method based on user preference analysis according to the present invention, wherein: the generating a recommendation list according to the context-aware similarity between products means that for each user, select M products from the products with the highest similarity to the user's historical interaction products as the recommendation candidate set, and filter out the products that do not conform to the current situation according to the user's current context to optimize the candidate set;

[0049] Construct feature vectors for each candidate product, use the GBDT model to predict the scores of the recommendation candidate set for each user, sort the recommendation candidate set in descending order according to the predicted scores, and select the top M products according to the sorting result to form the final recommendation list, and push the product content of interest to the user according to the recommendation list.

[0050] As a preferred solution of the product push method based on user preference analysis according to the present invention, wherein: the pushing the product content of interest to the user according to the recommendation list means defining the push frequency and time for instant push according to the user's online status and active time, monitoring and recording the user's reaction to the pushed products in real time, integrating the collected user feedback into the user profile, and updating the user preferences and behavior patterns in real time.

[0051] Another object of the present invention is to provide a product push system based on user preference analysis, which includes,

[0052] A data collection module for collecting user behavior data, commodity data, and context data from multiple channels and preprocessing the data.

[0053] A construction module for extracting key features from the preprocessed data to construct a user profile and constructing a commodity knowledge graph based on the analysis of commodity data.

[0054] An evaluation module for calculating the importance index of each commodity feature and generating a feature - sentiment pair by combining the sentiment score and importance index of the commodity feature.

[0055] An analysis module for constructing a user - commodity interaction graph based on the user profile and the feature - sentiment pair, using the GraphSAGE model to learn the embedded representation of the commodity from the interaction graph, and calculating the context - aware similarity between commodities.

[0056] A recommendation module for generating a preliminary recommendation list based on the similarity and using the GBDT model to perform personalized ranking on the list to finally generate a personalized recommendation list, pushing commodities according to the list and collecting user feedback.

[0057] A computer device comprising: a memory and a processor; the memory stores a computer program, and when the processor executes the computer program, the steps of the commodity pushing method based on user preference analysis are implemented.

[0058] A computer - readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the commodity pushing method based on user preference analysis are implemented.

[0059] The beneficial effects of the present invention are as follows: By using sentiment analysis technology and the fuzzy Kano model, the present invention not only improves the accuracy of the user profile, but also refines the importance evaluation of commodity features through the feature - sentiment pair. By constructing a user - commodity interaction graph and combining the calculation of context - aware similarity, it not only considers the user's real - time location and environmental changes, but also can dynamically adjust the recommended content, significantly improving the personalization degree of the recommendation system and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0061] Figure 1 It is a flow diagram of the commodity pushing method based on user preference analysis.

[0062] Figure 2Schematic diagram for the generation of feature - sentiment pairs.

[0063] Figure 3 Schematic diagram of the structure of a commodity push system based on user preference analysis. Specific implementation manners

[0064] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific implementation manners of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0065] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art may make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0066] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or selectively exclusive embodiments from other embodiments.

[0067] Embodiment 1. Refer to Figure 1 and Figure 2 , which is the first embodiment of the present invention. This embodiment provides a commodity push method based on user preference analysis. The commodity push method based on user preference analysis includes:

[0068] S1. Collect multi-source data, perform sentiment analysis on user comments to generate user portraits and construct a commodity knowledge graph, use a fuzzy Kano model to evaluate feature importance and generate feature - sentiment pairs;

[0069] Specifically, collecting multi-source data, performing sentiment analysis on user comments to generate user portraits and construct a commodity knowledge graph means collecting users' behavioral data, commodity data, and context data including geographical location, timestamp, and device information from websites, mobile applications, and social media, cleaning the collected data, and using NLP tools to perform word segmentation, stop word removal, and part-of-speech tagging on user comments;

[0070] Select key features in user behavior, including access frequency, purchase frequency, and social interaction frequency, use one-hot encoding and TF-IDF methods to convert behavioral features into vector representations and identify feature words in comments, including battery life and screen resolution;

[0071] Integrate behavioral data with geographical location, timestamp, and device information, and attach context tags to each user behavior record;

[0072] For sentiment analysis using the BERT model, the feature words are input into the BERT model to obtain the feature vectors of the text. The feature vectors output by the BERT model are input into a classification layer composed of a fully connected layer and a Softmax activation function to output the probabilities of each sentiment category:

[0073]

[0074] where P(y = c|z) is the probability of predicting as category c given the feature vector z, W c and b c are the weights and biases of category c, C is the total number of sentiment categories, W j is the weight matrix related to category j, b j is the bias related to category j, T is the transpose operation, and the calculation of weights and biases is achieved through initialization, definition of the loss function, backpropagation, and iterative application of the optimization algorithm;

[0075] The sentiment intensity is calculated using a linear function in another fully connected layer:

[0076]

[0077] where I is the sentiment intensity, with a value range of [0,1], σ is the Sigmoid function, V i and b i are the weights and biases of the sentiment intensity layer, and T is the transpose operation;

[0078] The sentiment score is calculated for each comment based on the sentiment category probability value and the sentiment intensity:

[0079] S l = P(r l ) × I(r l ),

[0080] where S l is the sentiment score of the l-th user comment, P(r l ) is the sentiment category probability value of the l-th user comment, and I(r l ) is the sentiment intensity of the l-th user comment (a real number between 0 and 1);

[0081] Calculate the overall sentiment score of each user, integrating the sentiment information of all comments:

[0082]

[0083] where S total is the overall sentiment score of the user, with a range of [0,1], and N is the number of comments of the user;

[0084] Assign emotional labels to users based on the total emotional score:

[0085] If S total ≥ 0.75, the emotional label is "positive";

[0086] If 0.75 > S total > 0.25, the emotional label is "neutral";

[0087] If S total ≤ 0.25, the emotional label is "negative";

[0088] The selection of 0.75 and 0.25 as thresholds is based on the idea of quartiles, where the data is divided into four equal parts, each part containing 25% of the data points. Positive emotions are usually considered to be comments with higher scores. Therefore, choosing data points above the 75th percentile as positive emotions can ensure that only truly positive comments are classified as positive. The threshold for negative emotions is set below the 25th percentile to ensure that comments classified as negative actually contain more negative emotions. This data-based quartile method can adapt to different datasets and emotional distributions, making the emotional classification closer to the actual data distribution. Such a division can better balance the quantities among different categories, avoiding too many or too few in a certain category, thereby improving the generalization ability and accuracy of the model;

[0089] Combine the behavioral feature vector with the emotional score to form a user profile and use the K-means clustering algorithm to divide users into different groups;

[0090] Extract features from the collected product data, including category, brand, price, and rating. After standardization, convert the attribute information into a feature vector A, use one-hot encoding to represent the category and brand, and standardize the price and rating;

[0091] Identify the relationships between products through co-occurrence analysis and collaborative filtering techniques, and calculate the relationship strength PMI(x,y) between products using point mutual information:

[0092]

[0093] In the formula, P(x,y) is the probability that products x and y appear simultaneously, and P(x) and P(y) are the probabilities that product x and product y appear alone, respectively;

[0094] Create a node for each product, and the node attributes include the product feature vector A and the user emotional label. Define the edges between products according to the PMI value and the user's behavior pattern, and the edge attributes reflect the relationship type and strength between products;

[0095] Update when new products are added or product information changes, recalculate the PMI values of relevant products, and update the corresponding edge weights and node information in the product knowledge graph.

[0096] By collecting user behavior data and context information from multiple channels such as websites, mobile apps, and social media, the platform can obtain a rich and diverse dataset that reflects users' real interests and behavior patterns. Using NLP tools to conduct sentiment analysis on user comments, the BERT model can accurately capture the complex emotions in user comments. When users browse products on the platform and leave comments, the system can analyze their sentiment in real time. If a user expresses strong positive emotions towards a certain type of product (such as electronic products) in the evaluation, the system will automatically adjust the user profile, record the preference for this type of product, and provide support for subsequent product recommendations. By combining the user's behavior characteristics (such as visit frequency, purchase frequency) with the sentiment analysis results, the system constructs an accurate user profile. At the same time, the product knowledge graph updates the product relationships and user preference changes in real time to ensure the timeliness of the recommendation system. The accuracy of the user profile directly affects the construction of the product knowledge graph, and the real-time updated knowledge graph in turn optimizes the user profile, forming a positive feedback between the two. When users continuously browse different categories of products (such as electronic devices, household items), the system will update the user profile in real time according to the browsing frequency and sentiment in the comments, identify the special preference for electronic devices of the user, and strengthen the relationship weight between the electronic device node and the user node in the knowledge graph to enhance the accuracy of subsequent recommendations. Using the K-means clustering algorithm to divide users into groups, and classifying users into different interest groups according to the sentiment scores and behavior characteristics. In this way, the recommendation system can identify and push personalized content that better matches users' preferences, significantly improving the relevance and accuracy of recommendations.

[0097] User group segmentation provides a more detailed user profile, enabling the recommendation system to make differentiated recommendations among different groups, thereby increasing the success rate of recommendations. The platform can identify user groups who like technology products and recommend the latest technology products and related accessories to them, or push home product promotion information to users who love home products. Group segmentation enables each user to receive product recommendations that match their interests, thus enhancing the user's purchase intention and the platform's sales. The dynamic update mechanism of the product knowledge graph enables the system to flexibly respond to market changes and fluctuations in user demand. Real-time data updates ensure that the recommendation results of the system are always based on the latest information, providing strong support for the enterprise's market decision-making. Dynamic adaptability ensures the real-time nature of the recommendation system, and real-time data provides an accurate basis for market decision-making. When new products are launched or the sales volume of a certain product surges, the system will automatically update the weights of relevant nodes and edges in the product knowledge graph, adjust the recommendation strategy, and give priority to recommending new products and popular products. Enterprises can use this information to adjust inventory, formulate promotion strategies, and improve market response speed and profitability.

[0098] Furthermore, using the fuzzy Kano model to evaluate feature importance and generate feature - sentiment pairs means designing a questionnaire that includes the extracted product features, asking users about their satisfaction and expectation levels for each product feature using a Likert scale (1 - 5 points), collecting user responses using an online survey tool, performing coding conversion on the user responses, and converting the Likert scale scores into the numerical inputs required for the fuzzy set;

[0099] For satisfaction and expectation levels, define three fuzzy sets, including a high fuzzy set, a medium fuzzy set, and a low fuzzy set, and represent each fuzzy set using a triangular membership function:

[0100]

[0101] In the formula, δ(q) is the membership function, with a value range of [0,1], a', b', c' are the cut-off points for high, medium, and low satisfaction, and q is the score value;

[0102] Using fuzzy Kano logic, classify the features according to the fuzzy set membership degrees of user satisfaction and expectation levels, and calculate the importance index of each product feature using fuzzy AND:

[0103] G t =min(μ sat (f),μ exp (f));

[0104] In the formula, G t is the importance index of product feature t, μ sat (f) and μ exp (f) are the fuzzy membership degrees of satisfaction and expectation respectively;

[0105] Generate a comprehensive importance report for each product feature based on the feature importance index and sentiment score;

[0106] Combine the sentiment score and importance index of the feature to generate a list of feature - sentiment pairs for each product. Each feature - sentiment pair includes the feature name, sentiment score, and importance index.

[0107] By designing a questionnaire that extracts product features and using a Likert scale to collect user satisfaction and expectation levels, the fuzzy Kano model can meticulously evaluate the importance of each product feature. This process combines users' subjective perceptions and objective data analysis to generate a more comprehensive feature importance index. Using fuzzy sets and triangular membership functions to represent user satisfaction and expectation, fuzzy logic is introduced into feature evaluation, significantly improving the distinguishability of feature importance. The membership function calculates the fuzzy membership degree at different satisfaction levels through a piecewise linear function, accurately capturing users' fuzzy perception of features. This analysis can not only identify the features that users consider crucial but also adjust the direction of product development and improvement based on users' feedback. For example, if "battery life" shows a high importance index in both the fuzzy membership degrees of satisfaction and expectation, then this feature should be a key focus for product improvement. Combining the sentiment score and importance index of the feature to generate feature - sentiment pairs further deepens the understanding of user preferences. Each feature - sentiment pair includes the feature name, sentiment score, and importance index, providing a detailed overview of users' emotions and feature preferences. The calculated feature importance index G t Integrates users' satisfaction and expectation to ensure that the importance of each product feature can be quantified and compared. This result is combined with the sentiment score to generate a list of feature - sentiment pairs that provides accurate emotional and performance analysis for each product. Through the analysis of feature - sentiment pairs, the system can generate a comprehensive importance report for each product feature. This report not only provides strong support for the enterprise's marketing strategy but also offers a clear direction for product optimization. For example, by analyzing the feature - sentiment pairs of a certain smartphone, it can be found that users have a positive emotion and a high importance evaluation of "screen resolution", thus suggesting emphasizing this feature in market promotion.

[0108] S2. Construct a user - product interaction graph based on the user profile and feature - sentiment pairs, and calculate the context - aware similarity between products according to the spatio - temporal information in the user - product interaction graph;

[0109] Specifically, constructing a user - product interaction graph based on the user profile and feature - sentiment pairs means combining the user's behavior data with the feature - sentiment pairs, annotating the relevant product features and user emotions for each piece of user behavior data, and annotating the relevant product features and the user's sentiment score and importance index for the emotion of each layer of user - product interaction;

[0110] Create a node for each independent user, and the node attributes include the identification information of the user. Create a node for each independent product, and the node attributes include the product ID, category, and brand. Map each group of user behaviors to the edges of the user-product interaction graph and use the context information as the edge attributes. Store the user-product interaction graph using a graph database.

[0111] Create a node for each independent user, and the node attributes include the user's identification information, behavior characteristics, interest tags, etc. These attributes can comprehensively depict the user's personal profile and behavior patterns. Create a node for each independent product, and the node attributes include the product ID, category, brand, and the sentiment score and feature importance obtained through the feature-sentiment pair. This information helps the system identify the value of the product from the user's perspective. Through the creation of user and product nodes, the system can accurately identify and describe each user and product. This detailed node information lays a solid foundation for subsequent data analysis and recommendation. The feature-sentiment pair information contained in the product node enables the recommendation system to consider the user's sentiment tendency and feature preference, thereby improving the personalization and accuracy of the recommendation. By converting user behaviors into edges in the graph, the system can comprehensively capture the complex interaction relationships between users and products. This multi-dimensional interaction modeling helps to reveal the deep needs and potential interests of users. By converting user behaviors into edges in the graph, the system can comprehensively capture the complex interaction relationships between users and products. This multi-dimensional interaction modeling helps to reveal the deep needs and potential interests of users. Utilize the user and product nodes and the edge relationships between them to construct a complete user-product interaction graph. This graph reflects all the interaction relationships between users and products, including direct interactions and indirect associations.

[0112] Furthermore, calculating the context-aware similarity between products according to the spatio-temporal information in the user-product interaction graph means using the GraphSAGE model to learn the embedding representation of products from the interaction graph. The embedding vector generation formula is:

[0113]

[0114] In the formula, is the embedding vector of node v in the k-th layer, is the embedding vector of node u in the (k - 1)-th layer, Q(v) is the set of neighbor nodes of node v, A is the aggregation function, which commonly includes mean, maximum, summation, etc. In this embodiment, the mean aggregation is preferably used, W (k) is the weight matrix, B (k) is the bias term in the k-th layer, and μ is the ReLU activation function;

[0115] Calculate the context-aware similarity between products by combining the embedding vectors and context attributes of the products:

[0116]

[0117] In the formula, Sim og is the similarity between product o and product g, h o and h g are the embedding vectors of the products, γ is the weight factor of the context. By determining the context factors that affect the user's purchase decision (such as time, location, device type, etc.), using statistical analysis methods (such as analysis of variance, chi-square test) to evaluate the influence degree of different context factors on user behavior, determine the influence of each context factor, provide a basis for setting the weight factor γ. Based on the analysis results, set the initial value of γ. For example, if it is found that the time factor has a significant impact on the user's purchase behavior, a relatively high γ value can be set accordingly. Design an A / B test or a multivariate test, randomly assign users to different experimental groups, and each group applies a different γ value. Compare the performance of the recommendation system under different γ values. The concerned metrics include click-through rate, conversion rate, user satisfaction, etc. Use metrics such as the validation error of the machine learning model, the precision and recall rate of information retrieval to evaluate the effects of different γ values. Adjust the value of γ according to the experimental results to find the optimal solution. Select the γ value that performs best in the experiment as the final value. C(o,g) is the context similarity;

[0118] Recommend context-related similar products to the user according to the calculated similarity.

[0119] The GraphSAGE model generates embedding vectors that can capture the complex relationships and features between products by sampling and aggregating neighbor node information layer by layer. This high-dimensional embedding representation provides accurate data support for similarity calculation and enhances the model's representation ability. GraphSAGE uses sampling technology to avoid the complexity of full-graph calculation in traditional graph neural networks and improves the computational efficiency of large-scale datasets. This feature enables the system to process a large amount of user and product data and maintain efficient real-time response. The embedding vectors are not only used for calculating product similarity but also provide a basis for the personalized recommendation system, directly affecting the final recommendation accuracy. The context-aware similarity combines product embedding and context attributes, and can be dynamically adjusted according to the user's real-time needs and preferences to ensure that the recommended products are highly relevant. This personalized recommendation improves user satisfaction and engagement. By combining context information (such as the user's time preference, geographical location), the recommendation system can identify and match the user's specific needs, provide more targeted product suggestions, and increase the purchase intention. The calculation of context-aware similarity depends on real-time context information, which forms a closed loop with the dynamic update of the interaction graph and the embedding representation learning of the GraphSAGE model, enabling the system to quickly respond and adjust the recommendation strategy.

[0120] S3. Generate a recommendation list based on the context-aware similarity between products;

[0121] Specifically, generating a recommendation list based on the context-aware similarity between products means that for each user, N products are selected from the products with the highest similarity to the user's historical interaction products as the recommendation candidate set, and the products that do not conform to the current situation are filtered according to the user's current context to optimize the candidate set;

[0122] Construct a feature vector for each candidate product, including product similarity, user historical behavior, context information, product attributes, and sentiment analysis scores. Use the GBDT model for sorting prediction, train the GBDT model using historical user interaction data, and apply cross-validation and grid search techniques to optimize the model parameters. Define the optimization goal as maximizing the click-through rate and user satisfaction;

[0123] Use the trained GBDT model to predict the scores of the recommendation candidate set for each user, sort the recommendation candidate set in descending order according to the predicted scores, and select the top N products according to the sorting results to form the final recommendation list;

[0124] Push the product content that the user is interested in to the user according to the recommendation list.

[0125] By analyzing the user's real-time context information (such as geographical location, browsing time, device type, etc.), relevant product recommendations highly related to the current situation are provided for the user. This context-awareness ability enables the recommendations to be not only based on the user's historical behavior but also responsive to their immediate environment and mood changes, greatly enhancing the relevance and attractiveness of the recommendations. By providing spatio-temporally relevant product recommendations, the system can significantly improve the user's shopping experience, reduce information overload, and increase the user's trust and satisfaction with the recommendation system. For example, when a user searches for gifts on the eve of a holiday, the system can recommend products that match the festive atmosphere, enhancing the willingness to purchase. By using cross-validation and grid search techniques, the present invention ensures the consistency and superiority of the GBDT model's performance on different data subsets, systematically finding the optimal model parameters. This method improves the model's generalization ability and reduces the risk of overfitting. The optimization of the GBDT model enables faster processing of user requests, improving the system's response ability to real-time data, making the user hardly feel the waiting, thus enhancing the overall user satisfaction. Dynamically adjusting the candidate set according to the user's current situation ensures that the recommended products not only match the user's personal preferences but also adapt to their current needs. This flexible candidate set adjustment strategy improves the acceptance rate and click-through rate of the recommendations. Using the GBDT model to perform personalized ranking on the candidate products, considering multi-dimensional features of the products such as similarity, user historical behavior scores, and context information, makes the recommendation results more accurate and targeted. By analyzing the recommendation effects and user feedback, the present invention can provide enterprises with insights into user preferences and behavior trends, assisting enterprises in making more informed decisions in aspects such as market promotion, inventory management, and new product development. The system can capture and analyze market dynamics in real-time, respond to changes in user behavior, and help enterprises quickly adjust market strategies to seize market opportunities.

[0126] Furthermore, pushing the product content that the user is interested in to the user according to the recommendation list means defining the push frequency and time, and making an immediate push according to the user's online status and active time. Real-time monitoring and recording of the user's reactions to the pushed products, including click-through rate, purchase conversion rate, and page stay time, are carried out, and a user feedback channel is provided. The collected user feedback is integrated into the user profile to update the user preferences and behavior patterns in real-time.

[0127] Through accurate online status detection, it is possible to push messages during the user's most active time, improving the exposure rate and click-through rate of the pushed content. The dynamic adjustment of the push frequency makes the push rhythm more in line with the user's habits, avoiding over-disturbing the user and enhancing the user experience. The system records the user's behavior after receiving the pushed content in real time, including click-through rate, purchase conversion rate, and page stay time. Using data analysis technology, the system evaluates the effectiveness of the pushed content and provides a user feedback channel through which users can submit opinions and suggestions on the pushed content, helping the system understand the users' real needs and satisfaction. By continuously integrating user feedback, the system can keep the user profile updated in real time, making the user profile more in line with the real user behavior.

[0128] Example 2, referring to Figure 3 , is the second embodiment of the present invention. This embodiment is different from the previous one and provides a commodity push system based on user preference analysis, which includes

[0129] A data collection module for collecting user behavior data, commodity data, and context data from multiple channels and preprocessing the data;

[0130] A construction module for extracting key features from the preprocessed data to construct a user profile and constructing a commodity knowledge graph based on the analysis of commodity data;

[0131] An evaluation module for calculating the importance index of each commodity feature and generating a feature-sentiment pair by combining the sentiment score and importance index of the commodity feature;

[0132] An analysis module for constructing a user-commodity interaction graph based on the user profile and the feature-sentiment pair and using the GraphSAGE model to learn the embedded representation of the commodity from the interaction graph, and calculating the context-aware similarity between commodities;

[0133] A recommendation module for generating a preliminary recommendation list based on the similarity and using the GBDT model to perform personalized sorting on the list to finally generate a personalized recommendation list, pushing commodities according to the list and collecting user feedback.

[0134] When the above-mentioned functions are implemented in the form of 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 the present invention, in essence, or the part that contributes to the prior art, or a part of this 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 for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0135] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0136] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), fiber optic devices, and portable compact disc read-only memories (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as necessary, and then storing it in a computer memory.

[0137] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques known in the art or a combination thereof can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

Claims

1. A commodity push method based on user preference analysis, characterized in that: including, collecting multi-source data, performing sentiment analysis on user comments, calculating sentiment scores, generating user portraits and constructing product knowledge graphs, using the fuzzy Kano model to evaluate feature importance, calculating the importance index of each product feature, and generating feature-sentiment pairs; constructing a user-product interaction graph based on the user portrait and feature-sentiment pairs, and calculating the context-aware similarity between products according to the spatio-temporal information in the user-product interaction graph; generating a recommendation list according to the context-aware similarity between products; combining the sentiment scores and importance indices of features to generate a feature-sentiment pair list for each product, where each feature-sentiment pair includes a feature name, a sentiment score, and an importance index; the calculating the context-aware similarity between products according to the spatio-temporal information in the user-product interaction graph refers to using the GraphSAGE model to learn the embedded representation of products from the interaction graph, and the embedded vector generation formula is: In the formula, is the embedding vector of node v in the k-th layer, is the embedding vector of node u in the (k - 1)-th layer, Q(v) is the set of neighbor nodes of node v, A is the aggregation function, W (k) is the weight matrix, B (k) is the bias term in the k-th layer, and μ is the ReLU activation function; combining the embedded vectors and context attributes of products to calculate the context-aware similarity between products: where Sim og is the similarity between product o and product g, h o and h g are the embedding vectors of the products, γ is the weight factor of the context, and O(o, g) is the context similarity; recommending context-related similar products to users according to the calculated similarity.

2. The method for pushing products based on user preference analysis according to claim 1, wherein: the collecting multi-source data, performing sentiment analysis on user comments, calculating sentiment scores, generating user portraits and constructing product knowledge graphs refers to collecting users' behavioral data, product data, and context data including geographical location, timestamp, and device information from websites, mobile applications, and social media, cleaning the collected data, and using NLP tools to perform word segmentation, stop word removal, and part-of-speech tagging on user comments; selecting key features in user behavior, including access frequency, purchase frequency, and social interaction frequency, converting the behavioral features into vector representations using one-hot encoding and TF-IDF methods, identifying feature words in comments, integrating the behavioral data with geographical location, timestamp, and device information, and attaching context tags to each user behavior record; performing sentiment analysis using the BERT model, inputting the feature words into the BERT model to obtain the feature vectors of the text, inputting the feature vectors output by the BERT model into a classification layer composed of a fully connected layer and a Softmax activation function, and outputting the probability of each sentiment category; calculating the sentiment intensity using a linear function in another fully connected layer: I = σ(V i T z + b i ), where I is the emotional intensity, σ is the Sigmoid function, V i and b i are the weights and biases of the emotional intensity layer, and T is the transpose operation; calculating the sentiment score for each comment according to the sentiment category probability value and sentiment intensity: S l = P(r l ) × I(r l ), where S l is the sentiment score of the l-th user comment, P(r l ) is the sentiment category probability value of the l-th user comment, and I(r l ) is the sentiment intensity of the l-th user comment; calculating the overall sentiment score of each user and integrating the sentiment information of all comments: Where S total is the overall sentiment score of the user, and N is the number of comments of the user; assigning sentiment labels to users according to the total sentiment score: If S total ≥ 0.75, then the sentiment label is "positive"; If 0.75 > S total > 0.25, then the sentiment label is "Neutral"; If S total ≤ 0.25, then the sentiment label is "negative"; combining the behavioral feature vectors with the sentiment scores to form user portraits and using the K-means clustering algorithm to divide users into different groups; extracting features from the collected product data, standardizing them, and converting the attribute information into a feature vector A, representing the category and brand using one-hot encoding, and standardizing the price and rating; identifying the relationships between products through co-occurrence analysis and collaborative filtering techniques, and calculating the relationship strength PMI(x,y) between products using point mutual information: where P(x,y) is the probability that products x and y appear simultaneously, and P(x) and P(y) are the probabilities that products x and y appear alone, respectively; Create a node for each product. The node attributes include the product feature vector A and the user sentiment label. Define the edges between products according to the PMI value and the user's behavior pattern; Update when new products are added or product information changes. Recalculate the PMI values of related products and update the corresponding edge weights and node information in the product knowledge graph.

3. The commodity push method based on user preference analysis according to claim 2, wherein: The use of the fuzzy Kano model to evaluate feature importance, calculate the importance index of each product feature and generate a feature - sentiment pair refers to designing a questionnaire containing the extracted product features, using a Likert scale to ask users about their satisfaction and expected levels for each product feature, using an online survey tool to collect user responses, performing coding conversion on the user responses, and converting the Likert scale scores into the numerical inputs required by the fuzzy set; For satisfaction and expected levels, define three fuzzy sets, including a high fuzzy set, a medium fuzzy set, and a low fuzzy set, and use triangular membership functions to represent each fuzzy set: In the formula, δ(q) is the membership function, a', b', c' are the cut-off points for high, medium, and low satisfaction, and q is the score value; Use fuzzy Kano logic to classify features according to the membership degrees of the fuzzy sets of user satisfaction and expected levels, and use fuzzy AND to calculate the importance index of each product feature: G t = min(μ sat (f), μ exp (f)), where G t is the importance index of the product feature t, and μ sat (f) and μ exp (f) are the fuzzy membership degrees of satisfaction and expectation respectively; Generate a comprehensive importance report for each product feature based on the feature importance index and the sentiment score; Combine the sentiment score and importance index of the feature to generate a list of feature - sentiment pairs for each product. Each feature - sentiment pair includes the feature name, sentiment score, and importance index.

4. The method for pushing products based on user preference analysis according to claim 3, wherein: The construction of the user - product interaction graph based on the user profile and feature - sentiment pairs means combining the user's behavior data with the feature - sentiment pairs, and for each layer of user - product interaction, annotating the relevant product features and the user's sentiment score and importance index for the product features; Create a node for each independent user. The node attributes include the user's identification information. Create a node for each independent product. The node attributes include the product ID, category, and brand. Map each group of the user's behaviors to the edges of the user - product interaction graph and use the context information as the edge attribute. Store the user - product interaction graph using a graph database.

5. The method for pushing products based on user preference analysis according to claim 4, characterized in that: The generation of the recommendation list according to the context-aware similarity between products means that for each user, select M products from the products with the highest similarity to the user's historical interaction products as the recommendation candidate set, filter out the products that do not conform to the current situation according to the user's current context, and optimize the candidate set; Construct a feature vector for each candidate product, use the GBDT model to predict the scores of the recommendation candidate set for each user, sort the recommendation candidate set in descending order according to the predicted scores, and select the top M products according to the sorting result to form the final recommendation list, and push the product content that the user is interested in to the user according to the recommendation list.

6. The commodity push method based on user preference analysis according to claim 5, wherein: The pushing of product content of interest to the user according to the recommended list means defining the pushing frequency and time, and performing instant pushing according to the user's online status and active time, monitoring and recording the user's reaction to the pushed products in real time, integrating the collected user feedback into the user profile, and updating the user preferences and behavior patterns in real time.

7. A product push system based on user preference analysis for the product push method based on user preference analysis according to any one of claims 1-6, characterized in that: Including, a data collection module, configured to collect user behavior data, product data, and context data from multiple channels and preprocess the data; a construction module, configured to extract key features from the preprocessed data to construct a user profile and construct a product knowledge graph based on the analysis of product data; an evaluation module, configured to calculate the importance index of each product feature and generate a feature-sentiment pair by combining the sentiment score and importance index of the product feature; an analysis module, configured to construct a user-product interaction graph according to the user profile and the feature-sentiment pair, and use the GraphSAGE model to learn the embedded representation of the product from the interaction graph, and calculate the context-aware similarity between products; a recommendation module, configured to generate a preliminary recommendation list according to the similarity and use the GBDT model to perform personalized sorting on the list to finally generate a personalized recommendation list, push products according to the list, and collect user feedback.

8. A computer device, comprising: a memory and a processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the product pushing method based on user preference analysis according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the product pushing method based on user preference analysis according to any one of claims 1 to 6 are implemented.

Images