Knowledge graph-based identification analysis system

Abstract

The invention relates to the technical field of data processing, and particularly discloses an identifier analysis system based on a knowledge graph, which is used for solving the problem of low accuracy of recommended commodities caused by lack of extraction of emotion information in user online comments in the prior art. According to the method, the user comments, the commodity names and the commodity scores are obtained from the user online comments, word segmentation processing and part-of-speech tagging are performed on the user comments, the emotion words are extracted by using the preset emotion dictionary, the emotion tendency of the user is judged based on the polarity of the emotion words, and the user shopping knowledge graph is constructed. Extracting emotions and entities in the user comments, and forming a user commodity recommendation model through semantic analysis and preference mode mining; according to the method, the accuracy and the individuation level of the recommendation system are improved, and the intelligence and the interpretability of the recommendation system are improved through sentiment analysis and multi-dimensional mining.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and more specifically, to an identification and resolution system based on a knowledge graph. Background Art

[0002] In existing recommendation systems, identification and resolution mainly involve understanding the entities behind user behaviors (such as browsing, clicking, purchasing, etc.), matching them with the existing entities in the knowledge graph, thereby constructing a more accurate user portrait and realizing interest expansion. Traditional recommendation algorithms ignore the information contained in written reviews in order to measure users' online ratings of goods and services, thus reducing the accuracy of personalized recommendations. However, users' comments express opinions and reflect implicit preferences and emotions regarding the functions of products or services. When users surf the Internet, they leave behavioral information through browsers and other applications. By mining online user data such as purchasing, browsing, and rating behaviors, user preferences are analyzed and applied to personalized recommendations for goods or services. However, this method does not consider users' online comments. It can be seen from the increasing amount of evaluation information about products or services on the Internet that existing users are eager to express their true opinions and preferences, and this evaluation information usually affects the selection decisions of other users. Users' comments usually contain their subjective opinions and reflect their preferences and emotional tendencies regarding the attributes of products or services. By analyzing users' online comments, not only can the characteristics, advantages, and disadvantages of products and services be discovered, but also users' preferences and concerns can be found. However, in the existing technology, there is a lack of extraction of emotional information in users' online comments, resulting in a low accuracy of recommended goods. To solve the above problems, a technical solution is provided herein. Summary of the Invention

[0003] To overcome the above-mentioned defects of the prior art, the present invention provides an identification and resolution system based on a knowledge graph, which is used to solve the problem in the prior art that there is a lack of extraction of emotional information in users' online comments, resulting in a low accuracy of recommended goods, so as to solve the problems raised in the above background art.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] An identification and resolution system based on a knowledge graph includes a data collection layer, a data parsing and processing layer, a knowledge graph construction and management layer, and a commodity recommendation layer; the data collection layer is used to obtain user comments, commodity names, and commodity ratings from users' online comments;

[0006] The data parsing and processing layer is used to extract user comments, perform word segmentation processing and part-of-speech tagging on the user comments, extract emotion words using a preset emotion dictionary, and judge the user's emotional tendency based on the polarity of the emotion words;

[0007] The knowledge graph construction management layer is used to construct the user's shopping knowledge graph;

[0008] The product recommendation layer is used to extract emotions and entities from user reviews, and through semantic analysis and preference pattern mining, form a user product recommendation model; the product recommendation layer includes a semantic analysis module and a user preference module; the semantic analysis module is connected to the user preference module;

[0009] The semantic analysis module is used to decompose user reviews, extract entities and emotions in the reviews, and match them to the user's shopping knowledge graph;

[0010] The user preference module is used to map the "product - feature - emotion" relationship into a user preference graph model, where nodes represent products, edges represent the user review sentiment polarity judgment value, obtain user reviews for the same type of products based on the user preference graph model, construct a user product recommendation model to analyze the user's preference degree for this type of product, and the formula of the user product recommendation model is:

[0011]

[0012] In the formula: p(u,f) is the preference degree of user u for product f, F is the total number of products of the same type, Sim(f,p f ) is the similarity between product f and the product p that the user has reviewed f between, and q xf is the user review sentiment polarity judgment value of product f.

[0013] As a further solution of the present invention, the knowledge graph construction management layer is used to construct the user's shopping knowledge graph; the entities in the user's shopping knowledge graph include user entities, product entities, brand entities, attribute entities, and category entities, and the relationships include user - product relationships, user - brand relationships, user - category relationships, product - brand relationships, product - attribute relationships, product - category relationships, brand - category relationships, and attribute - category relationships.

[0014] As a further solution of the present invention, the data parsing and processing layer includes an emotion polarity discrimination module, an emotion word classification module, and a user emotion polarity judgment module; the emotion polarity discrimination module is used to calculate the proximity of an emotion word to a reference word with positive or negative meaning in the emotion dictionary, and based on semantic similarity, discriminate the polarity of the emotion word; the emotion word classification module classifies emotion words without adverbs as first - type emotion words and emotion words with adverbs as second - type emotion words, and calculates their emotion indices according to the second - type emotion words; the emotion polarity judgment module is used to import the emotion indices of the first - structure emotion words, the second - structure emotion words, and the third - structure emotion words in the user review into the review emotion polarity judgment model to judge the user review emotion polarity.

[0015] As a further solution of the present invention, the emotion polarity discrimination module is used to calculate the proximity between the emotion word and the reference words with positive or negative meanings in the emotion dictionary, and to discriminate the polarity of the emotion word based on semantic similarity. The formula for semantic similarity analysis is:

[0016]

[0017] In the formula: SH(α) is the polarity discrimination coefficient of the emotion word α, n is the total number of commendatory words in the emotion dictionary, Sim(α, com i ) is the similarity between the emotion word α and the i-th commendatory word in the emotion dictionary, α is the emotion word to be recognized, com i is the i-th commendatory word in the emotion dictionary, Sim(α, der j ) is the similarity between the emotion word α and the j-th derogatory word in the emotion dictionary, der j is the total number of derogatory words in the emotion dictionary, β is the total number of derogatory words in the emotion dictionary, Dis(α, com i ) is the distance between the emotion word α and the i-th commendatory word in the emotion dictionary, Dis(α, der j ) is the distance between the emotion word α and the j-th derogatory word in the emotion dictionary, m is the total number of derogatory words in the emotion dictionary.

[0018] As a further solution of the present invention, the second emotion word includes the first structural emotion word, the second structural emotion word, and the third structural emotion word; the formula for calculating the emotion index of the first structural emotion word is:

[0019]

[0020] In the formula: S w1 is the emotion index of the k-th first structural emotion word, SH(α k1 ) is the polarity discrimination coefficient of the k-th first structural emotion word α k1 , α k1 is the k-th first structural emotion word, is the polarity coefficient of the positive adverb / negative adverb;

[0021] The formula for calculating the emotion index of the second structural emotion word is:

[0022]

[0023] In the formula: S w2 is the emotion index of the second structural emotion word, SH(α g2 ) is the polarity discrimination coefficient of the g-th second structural emotion word α g2 , α g2 is the g-th second structural emotion word, is the polarity coefficient of the positive adverb before the g-th second-structure emotion word;

[0024] The formula for calculating the sentiment index of the third-structure emotion word is:

[0025]

[0026] In the formula: S w3 is the sentiment index of the third-structure emotion word, γ is the weakening degree of the negative adverb, SH(α t3 ) is the polarity discrimination coefficient of the t-th third-structure emotion word α t3 , α t3 is the t-th third-structure emotion word, is the polarity coefficient of the positive adverb before the t-th third-structure emotion word.

[0027] As a further solution of the present invention, the emotion polarity judgment module is used to import the sentiment index of the first-structure emotion word, the sentiment index of the second-structure emotion word, and the sentiment index of the third-structure emotion word in the user comment into the comment sentiment polarity judgment model to judge the sentiment polarity of the user comment. The formula of the comment sentiment polarity judgment model is:

[0028] q x = lgS w1 + αS w2 +(1 - α)S w3 ;

[0029] In the formula: q x is the user comment sentiment polarity judgment value, S w1 is the sentiment index of the k-th first-structure emotion word, S w2 is the sentiment index of the second-structure emotion word, S w3 is the sentiment index of the third-structure emotion word;

[0030] Compare the user comment sentiment polarity judgment value with the preset user comment sentiment polarity judgment threshold. If the user comment sentiment polarity judgment value is greater than or equal to the preset user comment sentiment polarity judgment threshold, the user comment sentiment is positive; if the user comment sentiment polarity judgment value is less than the preset user comment sentiment polarity judgment threshold, the user comment sentiment is negative.

[0031] Technical effects and advantages of an identification and resolution system based on a knowledge graph according to the present invention: The present invention obtains user comments, product names, and product ratings from user online reviews, performs word segmentation and part-of-speech tagging on the user comments, extracts sentiment words using a preset sentiment dictionary, determines the sentiment tendency of the user based on the polarity of the sentiment words, constructs a user shopping knowledge graph, extracts the sentiment and entities in the user comments, and forms a user product recommendation model through semantic parsing and preference pattern mining; the present invention not only improves the accuracy and personalization level of the recommendation system, but also enhances the intelligence and interpretability of the recommendation system through sentiment analysis and multi-dimensional mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic structural diagram of an identification and resolution system based on a knowledge graph provided by the present invention;

[0033] Figure 2 It is a schematic diagram of the operation process of an identification and resolution system based on a knowledge graph provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] The following will clearly and completely describe the technical solutions in the present invention with reference to the accompanying drawings in the present invention. Obviously, the described technical solutions are only a part of the present invention, rather than all of them. All other technical solutions obtained by those of ordinary skill in the art based on the technical solutions in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0035] An identification and resolution system based on a knowledge graph includes a data acquisition layer, a data parsing and processing layer, a knowledge graph construction and management layer, and a product recommendation layer; the data acquisition layer is connected to the data parsing and processing layer, the data parsing and processing layer is connected to the knowledge graph construction and management layer, and the knowledge graph construction and management layer is connected to the product recommendation layer;

[0036] The data acquisition layer is used to obtain user comments, product names, and product ratings from user online reviews;

[0037] The data parsing and processing layer is used to extract user comments, perform word segmentation and part-of-speech tagging on the user comments, extract sentiment words using a preset sentiment dictionary, and determine the sentiment tendency of the user based on the polarity of the sentiment words;

[0038] The knowledge graph construction and management layer is used to construct a user shopping knowledge graph; the entities in the user shopping knowledge graph include user entities, product entities, brand entities, attribute entities, and category entities, and the relationships include user-product relationships, user-brand relationships, user-category relationships, product-brand relationships, product-attribute relationships, product-category relationships, brand-category relationships, and attribute-category relationships;

[0039] The product recommendation layer is used to extract the sentiment and entities in user reviews, and through semantic parsing and preference pattern mining, a user product recommendation model is formed.

[0040] For example, user A commented, "The battery life of this mobile phone is very strong, but the screen display is not good." The above comment is tokenized into: "this", "mobile phone", "battery life", "very strong", "screen display", "not good". Among them, "battery life" and "screen display" are product features, "very strong" and "not good" are sentiment words. "Very strong" is a positive sentiment word, and "not good" is a negative adverb. The entities in the user shopping knowledge graph constructed by them include: user entity: user A; product entity: mobile phone; attribute entities: battery life and screen display; category entity: electronic product; Based on the above entities, the following relationships are constructed: user-product relationship: (user A) purchases (mobile phone); product-attribute relationship: (mobile phone) has (battery life), (mobile phone) has (screen display); product-category relationship: (mobile phone) belongs to (electronic product); attribute-category relationship: (electronic product) has (battery life), (electronic product) has (screen display).

[0041] By extracting the product features and sentiment polarities in user reviews, not only the overall attitude of users towards products is analyzed, but also the emotional preferences for specific attributes (such as "battery life", "photo-taking effect") are delved into. Combining the preferences of users across different products, brands, and categories, the knowledge graph can dynamically build a user interest model and recommend products that best match the user's current needs; the constructed knowledge graph not only includes the relationship between products and users, but also the association relationships between brands, attributes, and categories, enabling the recommendation system to make recommendations based on different dimensions; based on the sentiment analysis in user reviews, the recommendation system can clearly explain the reasons for recommendations, enhancing users' trust in the recommended products; the knowledge graph has the ability of dynamic update and self-learning. The system will continuously improve the knowledge graph with new user reviews and optimize the recommendation effect. If a user expresses a preference for a certain brand or attribute in multiple reviews, the system will gradually increase the recommendation weight for such products, achieving personalized evolution; through semantic parsing and preference pattern mining, the system can not only identify the directly expressed needs of users, but also mine potential needs; it not only improves the accuracy and personalization level of the recommendation system, but also enhances the intelligence and interpretability of the recommendation system through sentiment analysis and multi-dimensional mining, significantly enhancing the user experience, increasing the platform conversion rate, and playing an important role in fields such as e-commerce, social networking, and entertainment.

[0042] The data parsing and processing layer includes a sentiment polarity discrimination module, a sentiment word classification module, and a user sentiment polarity judgment module; the sentiment polarity discrimination module is connected to the sentiment word classification module, and the sentiment word classification module is connected to the user sentiment polarity judgment module;

[0043] The emotional polarity discrimination module is used to calculate the proximity of an emotional word to a reference word with positive or negative meaning in the emotional dictionary, and to determine the polarity of the emotional word based on semantic similarity. The formula for semantic similarity analysis is as follows:

[0044]

[0045] In the formula: SH(α) is the polarity discrimination coefficient of the emotional word α, n is the total number of positive words in the emotional dictionary, Sim(α, com i ) is the similarity between the emotional word α and the i-th positive word in the emotional dictionary, α is the emotional word to be recognized, com i is the i-th positive word in the emotional dictionary, Sim(α, der j ) is the similarity between the emotional word α and the j-th negative word in the emotional dictionary, der j is the total number of negative words in the emotional dictionary, β is the total number of negative words in the emotional dictionary, Dis(α, com i ) is the distance between the emotional word α and the i-th positive word in the emotional dictionary, Dis(α, der j ) is the distance between the emotional word α and the j-th negative word in the emotional dictionary, and m is the total number of negative words in the emotional dictionary.

[0046] The proximity between the emotional word to be recognized and the positive words (positive-direction words) and negative words (negative-direction words) in the emotional dictionary is calculated using semantic similarity, thereby determining the polarity of the emotional word. By calculating the similarity between the emotional word to be recognized and the positive / negative words, rather than relying solely on dictionary matching or simple keyword rules, misjudgments caused by an incomplete vocabulary are avoided; the parameter β in the formula can be dynamically adjusted to adapt to the sensitivity requirements of emotional polarity in different fields; by comparing the distances to positive and negative emotional words, misjudgments caused by semantic ambiguity are reduced. For example, "ruthless" may express both negative emotions (vicious) and positive emotions (decisive), and the system determines the emotional tendency by calculating the distances to words with different polarities; by calculating semantic similarity, not only the problem of insufficient dictionary coverage in traditional sentiment analysis methods is solved, but also the emotional polarity of new words and complex expressions can be judged in real time.

[0047] The emotional word classification module classifies emotional words without adverbs as first emotional words, and emotional words with adverbs as second emotional words, and calculates their emotional indices based on the second emotional words; the second emotional words include first-structure emotional words, second-structure emotional words, and third-structure emotional words; the structure of the first-structure emotional word is positive adverb / negative adverb + first emotional word; the structure of the second-structure emotional word is positive adverb + negative adverb + first emotional word; the structure of the third-structure emotional word is negative adverb + positive adverb + first emotional word;

[0048] The formula for calculating the sentiment index of the first - structure emotion words is as follows:

[0049]

[0050] In the formula: S w1 is the sentiment index of the k - th first - structure emotion word, SH(α k1 ) is the polarity discrimination coefficient of the k - th first - structure emotion word α k1 , α k1 is the k - th first - structure emotion word, is the polarity coefficient of positive / negative adverbs;

[0051] The formula for calculating the sentiment index of the second - structure emotion words is as follows:

[0052]

[0053] In the formula: S w2 is the sentiment index of the second - structure emotion word, SH(α g2 ) is the polarity discrimination coefficient of the g - th second - structure emotion word α g2 , α g2 is the g - th second - structure emotion word, is the polarity coefficient of the positive adverb before the g - th second - structure emotion word;

[0054] The formula for calculating the sentiment index of the third - structure emotion words is as follows:

[0055]

[0056] In the formula: S w3 is the sentiment index of the third - structure emotion word, γ is the weakening degree of the negative adverb, SH(α t3 ) is the polarity discrimination coefficient of the t - th third - structure emotion word α t3 , α t3 is the t - th third - structure emotion word, is the polarity coefficient of the positive adverb before the t - th third - structure emotion word.

[0057] By classifying emotion words and calculating the sentiment index in combination with adverbial structures, the sentiment analysis becomes more delicate and accurate. In actual emotional expressions, adverbs such as "very", "slightly", "not very" etc. have a significant impact on the sentiment polarity of emotion words. This module adds adverbs to the formula, which can effectively distinguish subtle differences such as "very good" and "a bit good", avoiding simple and crude sentiment analysis; in the calculation of the third-structure emotion words, a weakening parameter is added, effectively preventing the excessive weakening of positive sentiment by negative adverbs, making the weakening amplitude of negative adverbs adjustable flexibly to adapt to the needs of different fields and contexts, enhancing the robustness and flexibility of the system; introducing the sentiment polarity coefficient of adverbs into the calculation to form an accurate sentiment index, enabling the system to output specific sentiment scores for easy sorting and quantitative analysis, rather than simply "positive, negative, neutral" labels; the sentiment dictionary relied on by the module can be dynamically expanded and adjusted flexibly for different industries and fields, ensuring the adaptability of the model to new words and new expression ways, reducing the cold start problem, not only improving the accuracy and robustness of the system, but also enhancing the ability to understand complex texts and empowering various sentiment analysis tasks.

[0058] Among them, positive adverbs include very, especially, extremely, highly, quite, very much, extremely, very, relatively, rather, quite, slightly, a bit, somewhat, a little, somewhat, rather; negative adverbs include not at all, not in the least, not in the slightest, not very, not really, not enough, a bit, not much, rather, hardly, almost not; the summary table of the polarity coefficients of positive adverbs and negative adverbs is shown in Table 1:

[0059]

[0060] Table 1 Summary table of the polarity coefficients of positive adverbs and negative adverbs

[0061] The emotion polarity judgment module is used to import the sentiment index of the first-structure emotion words, the sentiment index of the second-structure emotion words, and the sentiment index of the third-structure emotion words in the user comment into the comment sentiment polarity judgment model to judge the sentiment polarity of the user comment. The formula of the comment sentiment polarity judgment model is:

[0062] q x =lgS w1 +αS w2 +(1-α)S w3 ;

[0063] In the formula: q x is the user comment sentiment polarity judgment value, S w1 is the sentiment index of the kth first-structure emotion word, S w2 is the sentiment index of the second-structure emotion word, S w3 is the sentiment index of the third-structure emotion word;

[0064] Compare the user comment sentiment polarity judgment value with the preset user comment sentiment polarity judgment threshold. If the user comment sentiment polarity judgment value is greater than or equal to the preset user comment sentiment polarity judgment threshold, the user comment sentiment is positive; if the user comment sentiment polarity judgment value is less than the preset user comment sentiment polarity judgment threshold, the user comment sentiment is negative.

[0065] By comprehensively integrating the sentiment indices of emotion words at different levels, a sentiment polarity judgment model is constructed to more comprehensively evaluate the sentiment tendency of user comments. The first - structure emotion words (such as "very good"), the second - structure emotion words (such as "specially dissatisfied"), and the third - structure emotion words (such as "not very bad") have different degrees of importance in sentiment expression. This formula, through the weighted combination of these three types of emotion words, makes sentiment analysis more comprehensive and three - dimensional. By integrating the sentiment intensities at different levels, it can effectively avoid the one - sidedness of a single emotion word in overall sentiment judgment; the first - structure emotion words, through logarithmic calculation, play a more significant amplifying role in extremely positive or negative emotions, ensuring that key sentiment features take precedence in influencing the overall emotion judgment. The second and third - structure emotion words correct the polarity through linear weighting, reflecting the fine - tuning effect of slight modification or ironic expression on sentiment; the logarithmic function can compress the exponential growth of extreme emotion words (such as "super great" or "extremely bad"), making the system more stable and avoiding the misjudgment of sentiment polarity caused by individual strongly emotional words. Extreme negative words will not cause the imbalance of the entire sentiment judgment due to excessive values, enhancing the robustness of the model; by calculating the user comment sentiment polarity judgment value, the system quantifies sentiment analysis into specific numerical values, facilitating the setting of reasonable thresholds for sentiment classification; this formula can sensitively capture complex sentiment modifications (such as combinations of negation + affirmation or affirmation + negation), thereby enhancing the system's judgment ability for ironic, euphemistic expressions, etc.; by introducing the indices of three types of emotion words and a flexible weighting mechanism, the accuracy and robustness of sentiment analysis are significantly improved, especially in dealing with complex and multi - level sentiment expressions, enabling the system to be more in line with the actual sentiment when facing real user comments, and improving the user experience and recommendation quality.

[0066] The product recommendation layer includes a semantic parsing module and a user preference module; the semantic parsing module is connected to the user preference module;

[0067] The semantic parsing module is used to decompose user comments, extract entities and emotions in the comments, and match them to the user shopping knowledge graph;

[0068] The user preference module is used to map the "product - feature - sentiment" relationship into a user preference graph model, where nodes represent products and edges represent the user comment sentiment polarity judgment value. Based on the user preference graph model, comments of the user on the same type of products are obtained, and a user product recommendation model is constructed to analyze the user's preference degree for this type of product. The formula of the user product recommendation model is:

[0069]

[0070] Where: p(u, f) is the preference degree of user u for product f, F is the total number of products of the same type, Sim(f, p f ) is the similarity between product f and the products p that the user has commented on f ; q xf is the user comment sentiment polarity judgment value of product f;

[0071] Obtain the preference degree of the user for the product, and compare the preference degree of the user for the product with the preset preference degree threshold of the user for the product. If the preference degree of the user for the product is greater than or equal to the preset preference degree threshold of the user for the product, continue to recommend products of this type; if the preference degree of the user for the product is less than the preset preference degree threshold of the user for the product, stop recommending products of this type.

[0072] The product recommendation layer combines semantic parsing and the user preference graph model, extracts emotions and product features from user comments, quantifies the preference degrees of users for different products, and realizes personalized recommendation. By calculating the sentiment polarity and product similarity of the products that the user has commented on, it can more truly reflect the user's preference for specific attributes of the product; for new products or new users, the similarity between products can be used for recommendation, that is, even if the user has not directly commented on this product, it can also be recommended through other similar products; after quantifying the preference degree, it can be judged whether to continue to recommend products by setting a threshold, so as to avoid recommending products that do not meet the user's needs, reduce recommendation redundancy and user loss; not only analyze the user's purchase records, but also deeply mine the sentiment polarity in the user's comments to ensure that the recommendation results are interpretable. After each user publishes a new comment, the system automatically updates the user preference graph model to ensure that the recommendation results always reflect the user's latest interests and needs. When the user's preference degree for a certain type of product is lower than the threshold, the system will stop recommending the same type of products, reduce unnecessary recommendation behaviors, and avoid user disgust.

[0073] In the embodiment of the present invention, user comments, product names, and product scores are obtained from user online comments, the user comments are segmented and part-of-speech tagged, emotion words are extracted using a preset emotion dictionary, the emotional tendency of the user is judged based on the emotion word polarity, and a user shopping knowledge graph is constructed to extract emotions and entities in the user comments. Through semantic parsing and preference pattern mining, a user product recommendation model is formed; the present invention not only improves the accuracy and personalization level of the recommendation system, but also enhances the intelligence and interpretability of the recommendation system through sentiment analysis and multi-dimensional mining.

[0074] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

[0075] Finally: The above description is only the preferred solution of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An identification and resolution system based on a knowledge graph, comprising a data acquisition layer, a data parsing and processing layer, a knowledge graph construction and management layer, and a commodity recommendation layer; characterized in that, The data collection layer is used to obtain user comments, product names, and product ratings from user online reviews; The data parsing and processing layer is used to extract user comments, perform word segmentation and part-of-speech tagging on user comments, extract emotion words using a preset emotion dictionary, and judge the user's emotional tendency based on the polarity of emotion words; The knowledge graph construction and management layer is used to construct a user shopping knowledge graph; The product recommendation layer is used to extract emotions and entities in user comments, and form a user product recommendation model through semantic parsing and preference pattern mining; the product recommendation layer includes a semantic parsing module and a user preference module; The semantic parsing module is connected to the user preference module; The semantic parsing module is used to decompose user comments, extract entities and emotions in the comments, and match them to the user shopping knowledge graph; The user preference module is used to map the "product - feature - emotion" relationship into a user preference graph model, where nodes represent products and edges represent the emotional polarity judgment values of user comments. Based on the user preference graph model, obtain user comments on the same type of products, construct a user product recommendation model to analyze the user's preference degree for this type of product. The formula for the user product recommendation model is: Where: p(u,f) represents the preference degree of user u for product f, F is the total number of products of the same type, Sim(f,p f ) represents the similarity between product f and the products p f that the user has commented on, and q xf is the judgment value of the user comment sentiment polarity of product f.

2. The identification and resolution system based on a knowledge graph according to claim 1, wherein The knowledge graph construction and management layer is used to construct a user shopping knowledge graph; the entities in the user shopping knowledge graph include user entities, product entities, brand entities, attribute entities, and category entities, and the relationships include user - product relationship, user - brand relationship, user - category relationship, product - brand relationship, product - attribute relationship, product - category relationship, brand - category relationship, and attribute - category relationship.

3. The identification and resolution system based on a knowledge graph according to claim 1, wherein The data parsing and processing layer includes an emotion polarity discrimination module, an emotion word classification module, and a user emotion polarity judgment module; the emotion polarity discrimination module is used to calculate the proximity of an emotion word to a reference word with positive or negative meaning in the emotion dictionary, and judge the polarity of the emotion word based on semantic similarity; The emotion word classification module classifies emotion words without adverbs as first - type emotion words and emotion words with adverbs as second - type emotion words, and calculates its emotion index based on the second - type emotion words; the emotion polarity judgment module is used to import the emotion indexes of the first - structure emotion words, second - structure emotion words, and third - structure emotion words in the user comments into the comment emotion polarity judgment model to judge the emotional polarity of the user comments.

4. The identification and resolution system based on the knowledge graph according to claim 3, characterized in that, The emotion polarity discrimination module is used to calculate the proximity of an emotion word to a reference word with positive or negative meaning in the emotion dictionary, and judge the polarity of the emotion word based on semantic similarity. The formula for semantic similarity analysis is: Where: SH(α) is the polarity discrimination coefficient of the emotion word α, n is the total number of positive words in the emotion dictionary, Sim(α, com i ) is the similarity between the emotion word α and the i-th positive word in the emotion dictionary, α is the emotion word to be recognized, com i is the i-th positive word in the emotion dictionary, Sim(α, der j ) is the similarity between the emotion word α and the j-th negative word in the emotion dictionary, der j is the total number of negative words in the emotion dictionary, β is the total number of negative words in the emotion dictionary, Dis(α, com i ) is the distance between the emotion word α and the i-th positive word in the emotion dictionary, Dis(α, der j ) is the distance between the emotion word α and the j-th negative word in the emotion dictionary, m is the total number of negative words in the emotion dictionary.

5. The identification and resolution system based on a knowledge graph according to claim 3, wherein The second - type emotion words include first - structure emotion words, second - structure emotion words, and third - structure emotion words; the formula for calculating the emotion index of the first - structure emotion words is: Where: S w1 is the emotional index of the k-th first structural emotion word, and SH(α k1 ) is the polarity discrimination coefficient of the k-th first structural emotion word α k1 , α k1 is the k-th first structural emotion word, is the polarity coefficient of positive / negative adverbs; The formula for calculating the emotion index of the second - structure emotion words is: Where: S w2 is the emotional index of the second structural emotion word, and SH(α g2 ) is the polarity discrimination coefficient of the g-th second structural emotion word α g2 , α g2 is the g-th second structural emotion word, is the polarity coefficient of the positive adverb before the g-th second structural emotion word; The formula for calculating the emotion index of the third - structure emotion words is: Where: S w3 is the emotional index of the third-structured emotion word, γ is the weakening degree of the negative adverb, SH(α t3 ) is the polarity discrimination coefficient of the t-th third-structured emotion word α t3 , α t3 is the t-th third-structured emotion word, is the polarity coefficient of the positive adverb before the t-th third-structured emotion word.

6. The identification and resolution system based on a knowledge graph according to claim 5, wherein The emotion polarity judgment module is used to import the emotion indexes of the first - structure emotion words, second - structure emotion words, and third - structure emotion words in the user comments into the comment emotion polarity judgment model to judge the emotional polarity of the user comments. The formula for the emotion polarity judgment model is: q x = lgS w1 + αS w2 +(1 - α)S w3 ; Where: q x is the user comment sentiment polarity judgment value, S w1 is the sentiment index of the k-th first-structure emotion word, S w2 is the sentiment index of the second-structure emotion word, S w3 is the sentiment index of the third-structure emotion word; Compare the user comment sentiment polarity judgment value with the preset user comment sentiment polarity judgment threshold. If the user comment sentiment polarity judgment value is greater than or equal to the preset user comment sentiment polarity judgment threshold, the user comment sentiment is positive; If the user comment sentiment polarity judgment value is less than the preset user comment sentiment polarity judgment threshold, the user comment sentiment is negative.

Images