Personalized recognition recommendation system and method based on knowledge graph

Abstract

The invention discloses a personalized recognition recommendation system and method based on a knowledge graph, and relates to the field of personalized recommendation systems.The method comprises the steps that historical behavior data of a user is collected and preprocessed, a cross-domain knowledge graph is constructed based on the preprocessed historical behavior data, a user comprehensive portrait is generated, and the user comprehensive portrait is further analyzed; and constructing a causal relationship model according to the user comprehensive portrait and the behavior pattern analysis result, outputting a causal relationship between the user behaviors, generating a personalized recommendation list for the user based on the causal relationship between the user behaviors, and monitoring the user behaviors in real time. According to the method, entity and relation description with rich semantics is given to the user portrait, the personalized level of recommendation is remarkably improved, potential interest points of the user can be recognized, and a personalized recommendation result is provided.

Description

Technical Field

[0001] The present invention relates to the field of personalized recommendation, and in particular to a personalized recognition recommendation system and method based on knowledge graph. Background Art

[0002] With the rapid development of information technology and the Internet, personalized recommendation methods have gradually become an important means for various platforms to improve user experience. Traditional recommendation algorithms mainly rely on collaborative filtering, content-based recommendation, matrix decomposition and other technologies. These methods generate personalized recommendations by analyzing the user's historical behavior and the characteristics of the products and content themselves.

[0003] However, the existing recommendation methods based on knowledge graphs still have some shortcomings in the construction and application process. First, the construction of knowledge graphs is limited to a single field and fails to fully capture the interactive behavior of users in multiple fields, resulting in a one-sided user portrait and failure to fully reflect the multi-dimensional interests of users. Second, existing knowledge graphs are often static and cannot be dynamically adjusted as user behavior changes, resulting in recommendation results lagging behind the real-time needs of users. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a personalized recognition recommendation system and method based on knowledge graph to solve the problem that the knowledge graph construction is limited to a single field and fails to fully capture the interactive behavior of users in multiple fields, resulting in a one-sided user portrait and unable to fully reflect the multi-dimensional interests of users. Secondly, the existing knowledge graphs are often static and cannot be dynamically adjusted as user behavior changes, resulting in the problem that the recommendation results lag behind the real-time needs of users.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a personalized recognition and recommendation method based on a knowledge graph, which includes collecting historical behavior data of users and preprocessing it;

[0008] Build a cross-domain knowledge graph based on pre-processed historical behavior data, generate a comprehensive user profile, and further analyze the comprehensive user profile to obtain behavior pattern analysis results;

[0009] Build a causal relationship model based on comprehensive user portraits and behavior pattern analysis results, and output the causal relationship between user behaviors;

[0010] Generate personalized recommendation lists for users based on the causal relationship between user behaviors;

[0011] Monitor user behavior in real time and use anomaly detection algorithms to analyze the differences between the user's latest behavior data and their historical behavior data;

[0012] Adjust the personalized recommendation list based on the difference results, and optimize the cross-domain knowledge graph and causal relationship model based on the user's behavior.

[0013] As a preferred solution of the personalized recognition and recommendation method based on knowledge graph described in the present invention, the user's historical behavior data is collected and preprocessed, specifically including the following steps:

[0014] Collect user clicks, browsing, purchases, searches, social interactions, and health data from multiple data sources;

[0015] Clean and remove duplicate data from the collection;

[0016] Use natural language processing technology to process unstructured data;

[0017] All data were normalized.

[0018] As a preferred solution of the personalized identification and recommendation method based on knowledge graph described in the present invention, wherein: a cross-domain knowledge graph is constructed based on the pre-processed historical behavior data, a comprehensive user portrait is generated, and the comprehensive user portrait is further analyzed to obtain a behavior pattern analysis result, which specifically includes the following steps:

[0019] Use entity recognition techniques to identify entities from preprocessed data and identify explicit and implicit relationships between entities;

[0020] Build an independent knowledge graph for each field, with the identified entities as nodes and the relationships between entities as edges;

[0021] Integrate the knowledge graphs of each field into a unified cross-domain knowledge graph through entity alignment and relationship mapping technology;

[0022] Based on the user nodes in the cross-domain knowledge graph, identify the relationship between the user nodes and other nodes, and obtain the user's behavior data;

[0023] Based on the user's behavior data, the relationship strength between the user and entity nodes in different fields is analyzed. Recursive neural networks and multi-layer graph embedding technology are introduced to recursively generate new graph levels after each user operation, capture the evolution of user behavior, and generate a dynamic comprehensive portrait of the user.

[0024] Extract behavioral features in different fields based on dynamic comprehensive portraits;

[0025] Analyze user behaviors in different fields through the Apriori algorithm and find out the correlation between these behaviors;

[0026] The behavioral characteristics of users in different fields and the associations between behaviors are combined and analyzed to obtain behavioral analysis results.

[0027] As a preferred solution of the personalized identification and recommendation method based on knowledge graph described in the present invention, a causal relationship model is constructed according to the comprehensive user portrait and behavior pattern analysis results, and the causal relationship between user behaviors is output, which specifically includes the following steps:

[0028] Extract key behavioral features related to potential causal relationships from comprehensive user profiles;

[0029] Use key behavioral features as nodes of the Bayesian network;

[0030] Determine the causal relationship between nodes through data-driven learning methods, and regard the causal relationship between nodes as the edge of the Bayesian network;

[0031] Use maximum likelihood estimation to calculate historical behavior data, obtain the conditional probability table of each node and its parent node, and complete the construction of the Bayesian network;

[0032] Use Bayesian networks to generate causal relationships between user behaviors based on comprehensive user portraits and behavior pattern analysis results.

[0033] As a preferred solution of the personalized identification and recommendation method based on knowledge graph described in the present invention, a personalized recommendation list is generated for the user based on the causal relationship between user behaviors, specifically including the following steps:

[0034] Extract features representing user behavior from the user behavior pattern analysis results to form a user behavior feature vector;

[0035] Calculate based on the causal relationship model and the user's behavioral feature vector to obtain the possibility of the user purchasing the product under the current behavior pattern;

[0036] Based on the causal relationship between user behaviors, the user's future needs are obtained and a priority-ordered personalized recommendation list is generated. The expression is:

[0037]

[0038] Among them, Q i Indicates the priority of recommended product i, A represents the possibility of purchasing product i under the user's current behavior mode. u represents the behavioral feature vector of user u, V i represents the weight of product i, V j represents the weight of product j, C jrepresents the actual interaction behavior of the user purchasing product j, C i represents the actual interaction behavior of the user to purchase product i, and B represents the total number of products in the current product pool.

[0039] As a preferred solution of the personalized recognition and recommendation method based on knowledge graph described in the present invention, the user behavior is monitored in real time, and the difference between the user's latest behavior data and its historical behavior data is analyzed using an anomaly detection algorithm, which specifically includes the following steps:

[0040] Set data difference thresholds based on historical behavior data;

[0041] Define a time window for real-time behavior;

[0042] Use the time series difference measurement algorithm to calculate the difference between real-time behavior data and historical behavior data, and compare it with the difference threshold;

[0043] When the difference value is greater than the difference threshold, these differences are analyzed and abnormal behaviors are marked, and the causes of the abnormal behaviors are further analyzed.

[0044] As a preferred solution of the personalized recognition and recommendation method based on knowledge graph described in the present invention, the personalized recommendation list is adjusted according to the difference results, and the cross-domain knowledge graph and causal relationship model are optimized according to the user's behavior, which specifically includes the following steps:

[0045] Based on the difference detection results, identify the difference types between user behavior and historical behavior;

[0046] Adjust the personalized recommendation list based on dynamically updated user behavior profiles and difference types;

[0047] Adjust and update cross-domain knowledge graphs and causal models based on real-time behavioral data.

[0048] In a second aspect, the present invention provides a personalized recognition recommendation system based on knowledge graph, comprising:

[0049] Data collection module, which collects users' historical behavior data as the basic data for subsequent analysis;

[0050] The preprocessing module cleans, denoises, formats and normalizes the collected user behavior data;

[0051] The knowledge graph construction module builds a cross-domain knowledge graph based on pre-processed historical behavior data to capture entities and their relationships;

[0052] User portrait module combines user behavior data with knowledge graph to generate a comprehensive portrait of the user;

[0053] The user behavior analysis module conducts in-depth analysis of user behavior patterns based on comprehensive user portraits, extracting the rules and preference patterns of user behavior;

[0054] The causal model building module builds a causal relationship model between user behaviors based on the comprehensive profile of users and the analysis results of their behavior patterns, and identifies the causal chain behind the behaviors;

[0055] The personalized recommendation module generates a personalized recommendation list for users based on the causal relationship model of user behavior to match their current needs and interests;

[0056] The monitoring and anomaly detection module monitors the latest behavior data of users in real time and uses anomaly detection algorithms to analyze the differences between the latest data and historical behavior data;

[0057] List adjustment module, dynamically adjusts personalized recommendation lists based on real-time behavior and anomaly detection results;

[0058] The knowledge graph and model optimization module continuously optimizes the domain knowledge graph and causal relationship model based on the user's latest behavior data.

[0059] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the personalized recognition and recommendation method based on the knowledge graph as described in the first aspect of the present invention.

[0060] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the personalized recognition and recommendation method based on the knowledge graph as described in the first aspect of the present invention.

[0061] The beneficial effects of the present invention are as follows: by constructing a cross-domain knowledge graph based on data from multiple fields and generating a comprehensive portrait of the user, a comprehensive understanding of the user's multi-dimensional and multi-domain behavior is achieved. The behavioral characteristics of users in different fields are integrated and analyzed, and semantically rich entity and relationship descriptions are given to the user portrait, which significantly improves the personalization level of the recommendation. By dynamically adjusting the recommendation list, the recommendation results can be updated in time when the user behavior changes, thereby ensuring the real-time and accuracy of the recommendation. Through the introduction of the causal relationship model, it can not only meet the current needs of users, but also identify the user's potential points of interest, provide personalized recommendation results, greatly improve the user experience and increase the rationality and accuracy of the recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0063] Figure 1 This is a flowchart of the personalized recognition and recommendation method based on knowledge graph in Example 1.

[0064] Figure 2 This is a system diagram of the personalized recognition recommendation system based on knowledge graph in Example 1. DETAILED DESCRIPTION

[0065] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0066] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0067] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0068] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a personalized recognition recommendation method based on knowledge graph, comprising the following steps:

[0069] S1. Through multi-channel (such as e-commerce platforms, social media, health applications, financial fields, etc.) and multi-dimensional data collection, it can cover various user behaviors and ensure the comprehensiveness and diversity of data.

[0070] S1.1. The collected historical behavior data includes but is not limited to click data, browsing data, purchase data, search data, social interaction data, health data, etc.

[0071] Click data refers to the specific objects (such as products, articles, videos, etc.) that users click on on the platform, reflecting the users' immediate interests.

[0072] Browsing data refers to the time a user stays on a page and the content they browse, and is used to analyze the user's interest depth and preferences.

[0073] Purchase data refers to the user's purchase history, including information such as purchased items, time, and amount, which directly reflects the user's consumption habits and needs.

[0074] Search data refers to users’ search behavior on the platform, revealing users’ proactive needs and emerging interests.

[0075] Social interaction data refers to the interactive behaviors of users on social platforms (such as likes, comments, shares, etc.), which are used to establish users' social relationship networks and analyze users' social influence and its potential impact on behavior.

[0076] Health data refers to the collection of users' health data (such as physical examination records, exercise data, etc.), especially on health or medical platforms, to help provide personalized health advice and recommendations.

[0077] Further explanation: By collecting user behavior data from multiple data sources, we can obtain comprehensive and three-dimensional user behavior information, not limited to data of a specific dimension, which greatly enriches the dimension of user portraits. This provides comprehensive data support for the subsequent cross-domain knowledge graph construction and user behavior analysis, and can more accurately understand the diverse needs and interests of users, and improve the personalization and accuracy of recommendations.

[0078] S1.2. Clean and remove duplicate data from the collected data.

[0079] S1.3. Among the historical behavior data of users, unstructured data (such as text, comments, search terms, product descriptions, etc.) accounts for a large proportion. In order to make full use of this data, natural language processing technology is introduced to analyze and process it.

[0080] The processing steps are text analysis, entity recognition and sentiment analysis.

[0081] Text analysis refers to the operations such as word segmentation, part-of-speech tagging, and sentiment analysis on text data such as user comments and descriptions to extract useful information from the data, such as the user's specific needs and emotional tendencies.

[0082] Entity recognition refers to the use of natural language processing technology to identify key entities in text (such as products, brands, places, etc.), thereby converting unstructured data into structured data to facilitate subsequent knowledge graph construction.

[0083] Sentiment analysis refers to the analysis of user comments and feedback to identify user attitudes toward a product or service, thereby better understanding user preferences.

[0084] It is further explained that natural language processing technology can effectively mine deep information in unstructured data, making up for the shortcomings of structured data. This processing process improves the utilization rate of data, and can more accurately identify users' interests, emotional tendencies and other characteristics, thereby providing a richer basis for personalized recommendations. At the same time, natural language processing technology converts unstructured data into structured data, providing more semantic information for the construction of knowledge graphs, improving the richness of knowledge graphs and the accuracy of recommendations.

[0085] S1.4. In order to ensure that data from different data sources can be effectively compared and analyzed under the same framework, all data were standardized.

[0086] S2. Extract key entities from the preprocessed historical behavior data through entity recognition technology. The entities can be users, products, content, places, brands, etc., depending on the application scenario.

[0087] Entity recognition technology relies on natural language processing algorithms and automatically identifies important entities in the data in combination with contextual information. This includes not only explicit entities (such as product names, user IDs, etc.), but also implicit entities (such as user interests, emotional tendencies, etc.). It can also identify explicit relationships (such as direct purchases, browsing, etc.) and implicit relationships (such as potential interest associations, emotional resonance, etc.) between these entities through context.

[0088] S2.1. After identifying entities and their relationships, build independent knowledge graphs for each field (such as e-commerce, social networking, health, etc.). Each graph consists of entities (nodes) and relationships (edges) between entities.

[0089] For example, in the e-commerce field, entities such as goods, brands, and users establish relationships through browsing, commenting, and purchasing; in the social field, interactions between users and content, and between users and social friends form relationships. The purpose of independently building a domain knowledge graph is to ensure that the semantic information of each domain can be fully mined and expressed, and to avoid confusion between data in different domains.

[0090] Further explanation: By building an independent knowledge graph for each field, we can capture the behavioral characteristics of users in different fields in detail and avoid misunderstandings and deviations caused by simple data mixing. This fine-grained division can generate more accurate and context-related recommendations for each field, effectively improving the accuracy of the recommendation results.

[0091] S2.2. Integrate knowledge graphs from different fields into a unified cross-domain knowledge graph through entity alignment and relationship mapping techniques. Entity alignment technology identifies the same or similar entities (such as users, brands, etc.) in different fields and maps them to the same node. Relationship mapping technology identifies the commonalities of relationships in different fields (such as the connection between purchase and browsing behavior) and standardizes them into a unified relationship.

[0092] Further explanation: By integrating independent domain graphs into a unified cross-domain knowledge graph, a global perspective analysis of user behavior can be achieved. This can not only analyze user behavior in a single domain, but also understand the linkage effect of user behavior in different domains. Cross-domain graph integration greatly enhances semantic understanding capabilities, making recommendation results more context-aware and providing richer and more comprehensive recommendations.

[0093] S2.3. In a unified cross-domain knowledge graph, users exist as core nodes. By analyzing the relationship between user nodes and other entity nodes, user behavior data is extracted. These relationships can be direct (such as a user purchased a certain product) or indirect (such as a user's friend recommended a certain product). Not only does it focus on the relationship between users and a single entity, but it also analyzes the complex interactions between users and multiple entities.

[0094] It is further explained that by extracting user behavior data from cross-domain knowledge graphs, it is possible to accurately capture the user's interactions in multiple fields. This comprehensive and detailed behavior data provides important support for generating a comprehensive profile of the user, and can make recommendations based on the user's global behavior, further improving the relevance of recommendations and user satisfaction.

[0095] S2.4. Calculate the relationship strength between users and entities in different fields. The expression is:

[0096]

[0097] in, Represents user u and entity e i The strength of the relationship within time t, N represents the number of interaction types between the user and the entity (such as clicks, views, etc.), and w k represents the weight of the k-th type of interaction behavior, f k (u,e i ,t') represents the interaction between user u and entity e at time t' i The frequency of the k-th type of interaction behavior, α k It represents the attenuation coefficient of the k-th behavior, which controls the decay speed of the interactive behavior over time. t represents the current time point, and t' represents the past time point.

[0098] S2.4.1. Based on the knowledge graphs in different fields, users interact with entity nodes in multiple fields. These interaction relationships are normalized and the comprehensive relationship measurement between users and entities is calculated. The expression is:

[0099]

[0100] Among them, R u (e i ) represents user u and entity e i The comprehensive relationship measure of M represents the number of fields, β j represents the weight of field j, reflecting the importance of the field in the overall relationship;

[0101] S2.4.2. According to entity e i The multi-dimensional attribute features construct the feature function g(e i ), which is used to reflect the impact of different aspects of an entity on the user's comprehensive profile;

[0102] Combining recursive neural networks and multi-layer graph embedding technology, a dynamic comprehensive profile of the user is recursively generated based on the data updated after each user behavior. The expression is:

[0103]

[0104] in, represents the dynamic comprehensive portrait generated by user u after the n+1th operation, L represents the number of entities in the field involved by the user, and γ i Indicates that the user is in entity e i The weight of the behavior in the field, Indicates that the user interacts with entity e during the nth operation i The comprehensive relationship measure, g(e i ) represents the influence of entity features on user portrait generation, λ represents the time decay parameter, which controls the influence weight of recent behavior on user portrait, T(e i ) represents the user and entity e i The latest interaction time, de i Represents entity e i The variable being integrated is used to represent the continuous cumulative contribution to the entity.

[0105] Dynamic comprehensive portrait The value range of is [0,∞). The larger the value, the stronger the user's interest is, and the more likely the user is to make recommendations in these areas.

[0106] It is further explained that by introducing the dynamic comprehensive portrait generated by recursive neural network and multi-layer graph embedding technology, we can have a comprehensive understanding of the user's interests, needs, consumption habits, etc., and can also capture and analyze changes in user behavior in real time to ensure the dynamic and timeliness of the recommendation results.

[0107] S2.5. Extract users’ behavioral characteristics in different fields based on dynamic comprehensive portraits, including field preferences, behavior frequency and behavior patterns.

[0108] The domain preference refers to the distribution of users' interests in multiple domains. For example, users may have more purchasing behaviors in the e-commerce domain and more interactive behaviors in the social domain.

[0109] The behavior frequency refers to the frequency of a user's behavior in a specific field, for example, in the health field, the frequency of a user's steps and heart rate data being recorded every day.

[0110] The behavior pattern refers to the fact that the user's behavior pattern may show similar trends in different fields. For example, the user's product browsing and purchasing behavior on the e-commerce platform may be related to their sharing behavior on social media.

[0111] S2.6. Extracting user behavior characteristics in different fields based on dynamic comprehensive portraits, defining user behaviors in different fields as items. For example, user browsing behavior in the e-commerce field, like behavior in the social field, and more than 5,000 steps in the health field, etc.

[0112] S2.6.1. Use the Apriori algorithm to find frequently occurring behavior combinations. For example, users browsing products in the e-commerce field and users liking in the social field may often occur at the same time.

[0113] The Apriori algorithm is a classic association rule mining algorithm, which is often used to analyze frequent item sets and association rules between item sets in large data sets. Its core idea is to iteratively discover frequent item sets and then generate association rules based on these frequent item sets.

[0114] S2.6.2. Generate association rules between behaviors based on frequent item sets. For example, the rule It means that after browsing a product, users usually like it on social platforms.

[0115] To further illustrate, through the analysis of the Apriori algorithm, the potential correlation between the user's behavior in different fields can be identified. This correlation analysis not only helps to understand the user's behavior pattern, but can also be used to optimize the recommendation function. For example, by identifying the correlation between the user's purchase behavior in the e-commerce field and the sharing behavior in the social field, when the user shares a certain type of product, related products can be recommended for purchase.

[0116] S2.7 combines the user's behavioral characteristics and association rules for analysis to obtain behavioral patterns such as cross-domain behavior transfer, behavior prediction, and personalized recommendation optimization.

[0117] The cross-domain behavior transfer refers to identifying the behavior transfer patterns of users in different domains. For example, through association rules It is speculated that users’ behavior in the e-commerce field may affect their behavior in the social field.

[0118] The behavior prediction refers to the user's current behavior characteristics and association rules, which can predict the user's future behavior. For example, if a user shares a certain type of product on a social platform, it is inferred that the user may buy similar products on an e-commerce platform.

[0119] The personalized recommendation optimization refers to optimizing the recommendation function by identifying the user's behavior patterns and association rules to provide users with more personalized recommendations. For example, if a user frequently browses a certain type of product and likes similar products on a social platform, this type of product can be recommended first.

[0120] Further explanation: Apriori algorithm can be used to discover potential correlations between user behaviors in different fields. This cross-field behavior analysis can combine user behaviors in one field to predict potential needs in other fields. By combining user behavior characteristics with association rules, user behavior patterns can be mined.

[0121] S3. Based on comprehensive portraits, analyze users’ behaviors such as clicks, browsing, purchases, and searches, and identify key behavioral features that affect subsequent behaviors. For example, if a user frequently searches for a certain type of product recently, it may indicate that he or she is about to purchase it.

[0122] Key behavioral features are those that are significantly associated with user behavioral decisions or subsequent behavioral changes.

[0123] For example, in purchasing behavior, a user's purchase of a certain type of product may be the result of other behaviors (such as browsing, clicking) or trigger subsequent behaviors (such as evaluation, repeat purchase).

[0124] Click behavior, the behavior of users clicking on a product on an e-commerce platform, may be affected by previous search behavior and may also affect subsequent purchasing decisions.

[0125] Search behavior: Users’ search behavior may reveal potential interests or needs and is often a precursor to purchasing behavior.

[0126] S3.1. The key behavioral features extracted from the user's comprehensive portrait are used as nodes in the Bayesian network.

[0127] Each node represents a user behavior feature, such as browsing product A, searching for product B, purchasing product C, etc. The directed edges between nodes represent causal relationships, that is, the occurrence of a behavior (node) may affect subsequent behaviors. For example, a user's click on product A may lead to the subsequent purchase of product A.

[0128] S3.2. After determining the nodes, add or delete edges gradually through greedy search, and evaluate the impact of each modification on the structure score by incrementally calculating BIC. In each candidate network structure, use BIC to calculate the score of the structure, and select the network structure with the highest score as the final Bayesian network structure.

[0129] The BIC refers to the Bayesian Information Criterion, which is a widely used scoring criterion whose goal is to penalize overly complex models while ensuring that the model fits the data.

[0130] The final Bayesian network structure directly reflects the dependencies and potential causal relationships between user behavior features.

[0131] S3.2.1. The final Bayesian network structure is used to determine which behavioral features (nodes) have causal relationships, and to determine the direction of the causal relationship (i.e., which behavior may lead to the occurrence of another behavior).

[0132] For example, the resulting network structure may contain the following causal relationships:

[0133] User searches for product A—>User browses product A

[0134] This edge indicates that after searching for a product, a user may browse the product's detail page, which is a reasonable causal relationship.

[0135] User browses product A -> User purchases product A

[0136] This edge indicates that after a user browses a product, he or she may further purchase the product, which is another reasonable causal relationship.

[0137] S3.2.2. The structure of the network is determined, and the conditional probability table of each node is calculated based on historical data.

[0138] The conditional probability table represents the probability of a node (behavior) occurring given a parent node.

[0139] The maximum likelihood estimation is used to calculate the historical behavior data, and the expression is:

[0140]

[0141] Among them, P(X=x|Pa(X)=pa) represents the conditional probability that node X takes the value x under the condition that the value of the parent node is pa, which represents the possibility of the current node occurring when certain conditions of the parent node occur. P represents probability, which refers to the possibility of an event occurring. X represents a node in the network. In a Bayesian network, a node usually represents a certain behavioral feature of a user. For example, X may represent whether a user purchases product A. X=x represents that node X takes a specific value x. For example, if X represents whether a user purchases product A, then X=yes, indicating that the user purchases product A, and X=no, indicating that the user does not purchase product A. Pa(X) represents the parent node set of node X. Parent nodes are those nodes that directly affect the current node. Pa(X)=pa represents that the parent node set Pa(X) takes a specific value combination pa. For example, if Pa(X) contains whether a user browses product A, then Pa(X)=yes, indicating that the user browses product A. Count represents the number of times a certain event occurs in historical data.

[0142] For example, X indicates whether the user purchased product A. X = yes means the user purchased product A, and X = no means the user did not purchase product A. Pa(X) indicates whether the user browsed product A. Pa(X) = yes means the user browsed product A, and Pa(X) = no means the user did not browse product A.

[0143] Through historical data, we get the counts: Count(X=yes, Pa(X)=yes)=58, which means that the number of times the user browsed and purchased product A is 58, and Count(Pa(X)=pa)=100, which means that the total number of times the user browsed product A is 100 (including the cases of purchasing and not purchasing after browsing the product);

[0144]

[0145] This means that when a user browses product A, the probability of purchasing product A is 58%.

[0146] Further explanation: by identifying the causal relationship between user behaviors, it is possible to infer the user's possible future behavior from the user's current behavior, thereby providing more accurate support for recommendations. The causal relationship in the Bayesian network makes the recommendation results more interpretable. Through the causal chain, it can explain why a certain product or content is recommended, thereby enhancing user trust.

[0147] S4, extracted behavioral feature vector A u Including but not limited to browsing frequency, click behavior, purchase history, search habits and dwell time.

[0148] The browsing frequency refers to the frequency with which users browse different types of products;

[0149] The click behavior refers to the number of times a user clicks on a certain type of product or advertisement;

[0150] The purchase history refers to the commodities and their categories that the user has purchased before;

[0151] The search habits refer to the keywords that users search for on the platform;

[0152] The dwell time refers to the dwell time of a user on a certain category of product page.

[0153] S4.1. Output the causal relationship between user behaviors based on the causal relationship model to obtain the possibility of users purchasing a certain product under the current behavior pattern;

[0154] For example, product C i is a product that the user may purchase, then the user purchases product C under the current behavior mode. i The possibility can be expressed as

[0155] S4.2. Generate a priority-ordered personalized recommendation list, which is expressed as:

[0156]

[0157] Among them, Q i Indicates the priority of recommended product i, A represents the possibility of purchasing product i under the user's current behavior mode. u represents the behavioral feature vector of user u, V i represents the weight of product i, V j represents the weight of product j, C j represents the actual interaction behavior of the user purchasing product j, C i represents the actual interaction behavior of the user to purchase product i, and B represents the total number of products in the current product pool.

[0158] It is further explained that by analyzing the user's behavior patterns and extracting behavior features, the user's future needs can be accurately predicted, thereby generating an accurate personalized recommendation list. Compared with the traditional rule-based recommendation method, the recommendation method based on the causal relationship of user behavior can better capture the changes in user interests and improve the relevance of recommendations.

[0159] S5. Extract the user's normal behavior range from historical data and set a difference threshold.

[0160] S5.1. In order to detect changes in real-time behavior, a short fixed window is defined to capture the user's behavior data within a certain period of time.

[0161] S5.2. The user's behavior data is regarded as time series data, that is, the user's behavior is recorded in chronological order within a certain period of time. In order to measure the deviation between real-time behavior and historical behavior, the Euclidean distance is used to calculate the difference between real-time behavior data and historical behavior data. The expression is:

[0162]

[0163] Where d represents the Euclidean distance between two time series, that is, the difference between the user's real-time behavior data and historical behavior data, and h a represents the ath data point of real-time behavior data, y a represents the ath data point of historical behavior data, and z represents the length of the time series.

[0164] S5.2.1. After obtaining the difference value, compare the difference value with the previously set difference threshold.

[0165] If the difference value is less than or equal to the difference threshold, it is considered that the current user's behavior is within the normal range and no abnormality occurs.

[0166] If the difference value is greater than the difference threshold, it indicates that the user's real-time behavior is significantly different from the historical behavior, which may be a potential abnormal behavior.

[0167] S5.2.2. Further analyze the causes of abnormal behavior.

[0168] For example, if there is a sudden and significant increase in user purchasing behavior, it may be due to the impact of a promotion.

[0169] If a user's login location suddenly changes, it could be a sign that their account has been compromised.

[0170] Further explanation: By defining time windows and using time series difference measurement algorithms, user behavior can be monitored in real time and abnormal differences from historical behavior data can be quickly detected. This real-time nature allows for immediate response to abnormal behavior, reducing potential risks or losses. By setting a difference threshold and combining it with anomaly detection algorithms, it is possible to effectively distinguish between normal behavior fluctuations and true abnormal behavior. Especially when combined with historical behavior data, accurate detection can be performed based on the user's past behavior patterns, reducing the possibility of false positives and false negatives.

[0171] S6, classify the differences into frequency differences, category preference differences, spatiotemporal behavior differences, and device behavior differences.

[0172] The frequency difference refers to a change in the user's operating frequency;

[0173] For example, if a user suddenly increases the number of times they browse or purchase a product, this may reflect a change in user interests or the influence of external stimuli (such as promotional activities).

[0174] The category preference difference refers to a change in the user's preference for certain product categories.

[0175] For example, a user previously browsed mainly electronic products, but recently began to browse clothing products frequently. This difference in category preference may indicate that the user's needs have changed.

[0176] The temporal and spatial behavior difference refers to a significant change in the user's access time or location.

[0177] For example, if a user usually shops during the day but has recently started frequenting late at night, or if the user's login location suddenly changes, these could be signs that the account has been compromised.

[0178] The device behavior difference refers to a change in the user's operating device.

[0179] For example, if a user switches from a mobile device to a desktop device, or switches from a commonly used device to a new device, this may indicate that the user's usage scenario has changed and may even involve security issues.

[0180] S6.1. Based on the updated user profile, calculate the user's preference values ​​for different products, reorder the products in the recommendation list, and consider the impact of the difference type.

[0181] For example, when a user’s category preference is detected to have changed significantly, products related to the new category will be recommended first.

[0182] Further explanation: By dynamically updating user portraits and adjusting recommendation lists in real time, the accuracy of recommendations can be significantly improved, avoiding outdated products or irrelevant content recommendations. Users can see recommended products that are highly relevant to their current needs and interests on the platform, thereby improving user satisfaction and stickiness.

[0183] S6.2. Update the relationships in the knowledge graph based on the user’s real-time behavior.

[0184] When the user's preference changes, the weight of the relationship between the user and nodes such as product category and brand will be adjusted. For example, if the user has recently frequently purchased products of a certain brand, the connection between the user and the brand node will be strengthened.

[0185] If it is detected that users have increased their preference for certain products or categories, the relationship weights between these products and other related products in the knowledge graph are adjusted. This update can help introduce more related products when recommending, thereby improving the diversity and relevance of recommendations.

[0186] S6.3. When the user's real-time behavior changes significantly, update the causal relationship model.

[0187] Continuously update user behavior patterns through real-time behavior data.

[0188] For example, if a user has recently started to frequently purchase items in a new category, the causal relationship between items in this category and other items can be relearned and the recommendation strategy can be adjusted.

[0189] Through causal modeling, we can identify which behaviors have the greatest impact on users’ future purchasing decisions and adjust the priority of recommendations accordingly.

[0190] Further explanation: By updating the cross-domain knowledge graph, we can more comprehensively understand user preferences and needs, and improve the diversity and relevance of product recommendations. The update of the causal model can better capture the causal logic behind user behavior, thereby improving the rationality and accuracy of the recommendation strategy. By updating the knowledge graph and causal model, we can more flexibly recommend some long-tail products (i.e. products that users may be interested in but do not browse often), thereby increasing the overall sales of the platform.

[0191] This embodiment also provides a personalized recognition recommendation system based on knowledge graph, including:

[0192] Data collection module, which collects users' historical behavior data as the basic data for subsequent analysis;

[0193] The preprocessing module cleans, denoises, formats and normalizes the collected user behavior data;

[0194] The knowledge graph construction module builds a cross-domain knowledge graph based on pre-processed historical behavior data to capture entities and their relationships;

[0195] User portrait module combines user behavior data with knowledge graph to generate a comprehensive portrait of the user;

[0196] The user behavior analysis module conducts in-depth analysis of user behavior patterns based on comprehensive user portraits, extracting the rules and preference patterns of user behavior;

[0197] The causal model building module builds a causal relationship model between user behaviors based on the comprehensive profile of users and the analysis results of their behavior patterns, and identifies the causal chain behind the behaviors;

[0198] The personalized recommendation module generates a personalized recommendation list for users based on the causal relationship model of user behavior to match their current needs and interests;

[0199] The monitoring and anomaly detection module monitors the latest behavior data of users in real time and uses anomaly detection algorithms to analyze the differences between the latest data and historical behavior data;

[0200] List adjustment module, dynamically adjusts personalized recommendation lists based on real-time behavior and anomaly detection results;

[0201] The knowledge graph and model optimization module continuously optimizes the domain knowledge graph and causal relationship model based on the user's latest behavior data.

[0202] This embodiment also provides a computer device, which is suitable for the personalized recognition and recommendation method based on the knowledge graph, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the personalized recognition and recommendation method based on the knowledge graph proposed in the above embodiment.

[0203] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0204] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the personalized recognition and recommendation method based on the knowledge graph proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0205] In summary, the present invention achieves a comprehensive understanding of the multi-dimensional and multi-domain behaviors of users by: constructing a cross-domain knowledge graph based on data from multiple fields and generating a comprehensive portrait of the user. Integrating and analyzing the behavioral characteristics of users in different fields, giving semantically rich entity and relationship descriptions to user portraits, significantly improving the personalization level of the recommendation function. By dynamically adjusting the recommendation list, the recommendation results can be updated in time when the user behavior changes, thereby ensuring the real-time and accuracy of the recommendation. Through the introduction of the causal relationship model, it can not only meet the current needs of users, but also identify the user's potential points of interest, provide personalized recommendation results, greatly improve the user experience and increase the rationality and accuracy of the recommendation.

[0206] Example 2, referring to Table 1, is the second example of the present invention. In order to further verify the technical solution of the present invention, experimental simulation data of a personalized recognition and recommendation method based on a knowledge graph is provided.

[0207] In order to demonstrate the advantages of the personalized recognition recommendation method based on the knowledge graph, this embodiment compares this method with the existing recommendation method. The existing recommendation method uses the collaborative filtering algorithm (Collaborative Filtering), which is one of the most widely used recommendation algorithms. The collaborative filtering algorithm analyzes the historical behavior data of a large number of users and uses the behavior patterns of similar users (such as purchases, browsing, ratings, etc.) to predict the content that the current user may be interested in. Collaborative filtering algorithms are mainly divided into two categories: user-based collaborative filtering (User-based Collaborative Filtering) and item-based collaborative filtering (Item-based Collaborative Filtering). In this experiment, the existing recommendation method uses user-based collaborative filtering, that is, by calculating the similarity between users, finding other users with similar behaviors to the target user, and then generating a recommendation list for the target user based on the behaviors of these similar users.

[0208] The data in this example comes from an e-commerce platform, which contains historical behavior data of 1,000 users in different fields, including click, browse, purchase and search records. The data collection spans 6 months and covers multiple fields, such as electronic products, clothing, daily necessities, etc. The existing recommendation method uses collaborative filtering algorithm to generate recommendation lists for users, while the recommendation method based on knowledge graph adopts a recommendation method that combines cross-domain knowledge graph with causal relationship model.

[0209] The implementation steps are as follows:

[0210] Data collection and preprocessing: Collect historical behavior data of users on e-commerce platforms, including clicks, browsing, purchases, searches, etc. The data is cleaned, deduplicated, and formatted, and unstructured data is standardized using natural language processing technology.

[0211] Construction of cross-domain knowledge graph: Based on the pre-processed data, entity recognition technology is used to identify entities in different fields from the user's behavior data, such as the products purchased by the user, the pages browsed, etc. By identifying the explicit and implicit relationships between entities, the knowledge graph within the field is constructed, and then the knowledge graphs of multiple fields are integrated into a unified cross-domain knowledge graph through entity alignment and relationship mapping.

[0212] User comprehensive portrait generation: Based on cross-domain knowledge graphs, recursive neural networks and multi-layer graph embedding technology are used to generate dynamic comprehensive portraits of users. Whenever a user generates a new behavior, the user's portrait is recursively updated to capture the evolutionary characteristics of the user's behavior.

[0213] Behavior pattern analysis and causal relationship model construction: Use the Apriori algorithm to analyze users' cross-domain behavior patterns and identify the correlation between behaviors. Then, based on the comprehensive user portrait and behavior pattern analysis results, build a causal relationship model to identify the potential causal relationship between user behaviors. Use the Bayesian network to build a causal relationship chain, calculate the conditional probability of each node based on the data-driven method, and finally generate a causal relationship model of user behavior.

[0214] Personalized recommendation list generation: Based on the causal relationship model and behavior pattern analysis results, a personalized recommendation list is generated for the user. Each recommended product is prioritized according to the user behavior feature vector and product weight to ensure that the recommendation list can more accurately reflect the user's potential needs.

[0215] Difference detection and real-time adjustment: Define a real-time behavior window (1 week) and use the time series difference measurement algorithm to calculate the difference between the user's real-time behavior data and historical behavior data. Based on the difference results, dynamically adjust the recommendation list and optimize the cross-domain knowledge graph and causal relationship model.

[0216] The details are shown in Table 1 below:

[0217] Table 1 Comparison of personalized recommendation methods

[0218]

[0219] By comparing the existing collaborative filtering recommendation method with the personalized recommendation method based on knowledge graph, the data in the table clearly shows the significant differences between the two.

[0220] First, from the perspective of user clicks, the recommendation method based on the knowledge graph can better capture the user's interests. The number of user clicks increased from 135 times in the collaborative filtering method to 180 times, with a difference value of 45, far exceeding the set difference threshold of 30. This shows that the recommendation method based on the knowledge graph has a higher recommendation relevance and can recommend products that better meet user needs, thereby attracting user clicks.

[0221] The increase in the number of user purchases was even more significant, from 18 times in the collaborative filtering method to 42 times, with a difference of 24, exceeding the threshold of 20. This result shows that the recommendation method based on the knowledge graph can not only attract user clicks, but also effectively promote conversions and help users find products that better meet their needs. The formation of this difference is due to the fact that this method identifies the deep-seated needs behind user behavior through a causal relationship model, while the collaborative filtering method only relies on the similarity between users and ignores the causal relationship between individual user behaviors.

[0222] The user stay time also increased from 23 minutes to 30 minutes, with a difference of 7 minutes, close to the threshold of 10 minutes. Although the difference in stay time is relatively small, it shows that the recommendation method based on knowledge graph can provide users with more attractive recommended content and make users spend more time on the platform. This result further proves the advantage of this method in accurately grasping the user's interests.

[0223] In terms of user satisfaction with recommendation lists, the recommendation method based on knowledge graphs performed particularly well, with user satisfaction increasing from 4.0 to 4.9, with a difference of 0.9, exceeding the threshold of 0.8. This reflects that users' recognition of recommendation lists has significantly increased, and recommendation lists based on knowledge graphs can more accurately reflect users' current needs and interests, while the recommendation results of collaborative filtering methods are relatively generalized.

[0224] In terms of the recognition rate of abnormal user behavior, the recognition rate of the recommendation method based on the knowledge graph reached 94%, while the collaborative filtering method was only 60%, with a difference value of 34, far exceeding the threshold of 15. This shows that the knowledge graph recommendation method performs better in real-time monitoring of user behavior and detecting anomalies. By analyzing the differences in user behavior, the recommendation strategy can be adjusted in time to ensure the accuracy and real-time nature of the recommendation results, while the collaborative filtering method is slow in responding to behavioral changes and is difficult to respond quickly.

[0225] Overall, the personalized recommendation method based on knowledge graph shows obvious advantages. First, in terms of the accuracy of personalized recommendation, this method achieves a recommendation accuracy of 92%, while the recommendation accuracy of the collaborative filtering method is only about 70%. Secondly, the knowledge graph method can generate dynamic user portraits, capture changes in user behavior in real time, and predict users' future needs through causal models, thereby generating more targeted recommendation lists. In contrast, the collaborative filtering method relies on the similarities between users and cannot fully explore the deep correlation of individual user behaviors, resulting in low generalization and accuracy of recommended content.

[0226] In addition, the knowledge graph method is particularly outstanding in abnormal behavior identification and real-time adjustment. By using the time series difference measurement algorithm, it can accurately identify changes in user behavior, and timely adjust the recommendation list and optimize the knowledge graph. This dynamic adjustment capability greatly enhances the flexibility of the recommendation method in dealing with complex user behaviors, while the collaborative filtering method lacks this real-time adjustment mechanism and is difficult to deal with immediate changes in user behavior.

[0227] In summary, the personalized recognition and recommendation method based on knowledge graph significantly improves the recommendation accuracy, user satisfaction and abnormal behavior recognition capabilities through more accurate user portraits, behavior pattern analysis and causal relationship models, especially in response to changes in user behavior, showing higher flexibility and response speed. These innovative and novel effects have been fully verified in the experimental data, proving the uniqueness and advantages of this method in existing recommendation technologies.

[0228] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A personalized recognition and recommendation method based on knowledge graph, characterized by: include, Collect users’ historical behavior data and pre-process it; Build a cross-domain knowledge graph based on pre-processed historical behavior data, generate a comprehensive user profile, and further analyze the comprehensive user profile to obtain behavior pattern analysis results; Build a causal relationship model based on comprehensive user portraits and behavior pattern analysis results, and output the causal relationship between user behaviors; Generate personalized recommendation lists for users based on the causal relationship between user behaviors; Monitor user behavior in real time and use anomaly detection algorithms to analyze the differences between the user's latest behavior data and their historical behavior data; Adjust the personalized recommendation list based on the difference results, and optimize the cross-domain knowledge graph and causal relationship model based on the user's behavior.

2. The personalized recognition and recommendation method based on knowledge graph according to claim 1, characterized in that: Collect the user's historical behavior data and pre-process it, which includes the following steps: Collect user clicks, browsing, purchases, searches, social interactions, and health data from multiple data sources; Clean and remove duplicate data from the collection; Use natural language processing technology to process unstructured data; All data were normalized.

3. The personalized recognition and recommendation method based on knowledge graph according to claim 2, characterized in that: Based on the pre-processed historical behavior data, a cross-domain knowledge graph is constructed to generate a comprehensive user portrait. The comprehensive user portrait is further analyzed to obtain the behavior pattern analysis results. The specific steps include: Use entity recognition techniques to identify entities from preprocessed data and identify explicit and implicit relationships between entities; Build an independent knowledge graph for each field, with the identified entities as nodes and the relationships between entities as edges; Integrate the knowledge graphs of each field into a unified cross-domain knowledge graph through entity alignment and relationship mapping technology; Based on the user nodes in the cross-domain knowledge graph, identify the relationship between the user nodes and other nodes, and obtain the user's behavior data; Based on the user's behavior data, the relationship strength between the user and entity nodes in different fields is analyzed. Recursive neural networks and multi-layer graph embedding technology are introduced to recursively generate new graph levels after each user operation, capture the evolution of user behavior, and generate a dynamic comprehensive portrait of the user. Extract behavioral features in different fields based on dynamic comprehensive portraits; Analyze user behaviors in different fields through the Apriori algorithm and find out the correlation between these behaviors; The behavioral characteristics of users in different fields and the associations between behaviors are combined and analyzed to obtain behavioral analysis results.

4. The personalized recognition and recommendation method based on knowledge graph according to claim 3, characterized in that: A causal relationship model is constructed based on the comprehensive user portrait and behavior pattern analysis results to output the causal relationship between user behaviors. The specific steps include: Extract key behavioral features related to potential causal relationships from comprehensive user profiles; Use key behavioral features as nodes of the Bayesian network; Determine the causal relationship between nodes through data-driven learning methods, and regard the causal relationship between nodes as the edge of the Bayesian network; Use maximum likelihood estimation to calculate historical behavior data, obtain the conditional probability table of each node and its parent node, and complete the construction of the Bayesian network; Use Bayesian networks to generate causal relationships between user behaviors based on comprehensive user portraits and behavior pattern analysis results.

5. The personalized recognition and recommendation method based on knowledge graph according to claim 4, characterized in that: Generate a personalized recommendation list for users based on the causal relationship between user behaviors, specifically including the following steps: Extract features representing user behavior from the user behavior pattern analysis results to form a user behavior feature vector; Calculate based on the causal relationship model and the user's behavioral feature vector to obtain the possibility of the user purchasing the product under the current behavior pattern; Based on the causal relationship between user behaviors, the user's future needs are obtained and a personalized recommendation list with priority order is generated. The expression is: Among them, Q i Indicates the priority of recommended product i, A represents the possibility of purchasing product i under the user's current behavior mode. u Represents the behavioral feature vector of user U, V i represents the weight of commodity I, V j represents the weight of product j, C j represents the actual interaction behavior of the user purchasing product j, C i represents the actual interaction behavior of the user to purchase product i, and B represents the total number of products in the current product pool.

6. The personalized recognition and recommendation method based on knowledge graph according to claim 5, characterized in that: Monitor user behavior in real time and use anomaly detection algorithms to analyze the differences between the user's latest behavior data and their historical behavior data. The specific steps include the following: Set data difference thresholds based on historical behavior data; Define time windows for real-time behavior; Use the time series difference measurement algorithm to calculate the difference between real-time behavior data and historical behavior data, and compare it with the difference threshold; When the difference value is greater than the difference threshold, these differences are analyzed and abnormal behaviors are marked, and the causes of the abnormal behaviors are further analyzed.

7. The personalized recognition and recommendation method based on knowledge graph according to claim 6, characterized in that: Adjust the personalized recommendation list based on the difference results, and optimize the cross-domain knowledge graph and causal relationship model based on the user's behavior. The specific steps include: Based on the difference detection results, identify the difference types between user behavior and historical behavior; Adjust the personalized recommendation list based on dynamically updated user behavior profiles and difference types; Update cross-domain knowledge graphs and causal models based on real-time behavioral data.

8. A personalized recognition and recommendation system based on a knowledge graph, based on the personalized recognition and recommendation method based on a knowledge graph according to any one of claims 1 to 7, characterized in that: include, Data collection module, which collects users' historical behavior data as the basic data for subsequent analysis; The preprocessing module cleans, denoises, formats and normalizes the collected user behavior data; The knowledge graph construction module builds a cross-domain knowledge graph based on pre-processed historical behavior data to capture entities and their relationships; User portrait module combines user behavior data with knowledge graph to generate a comprehensive portrait of the user; The user behavior analysis module conducts in-depth analysis of user behavior patterns based on comprehensive user portraits, extracting the rules and preference patterns of user behavior; The causal model building module builds a causal relationship model between user behaviors based on the comprehensive profile of users and the analysis results of their behavior patterns, and identifies the causal chain behind the behaviors; The personalized recommendation module generates a personalized recommendation list for users based on the causal relationship model of user behavior to match their current needs and interests; The monitoring and anomaly detection module monitors the latest behavior data of users in real time and uses anomaly detection algorithms to analyze the differences between the latest data and historical behavior data; List adjustment module, dynamically adjusts personalized recommendation lists based on real-time behavior and anomaly detection results; The knowledge graph and model optimization module continuously optimizes the domain knowledge graph and causal relationship model based on the user's latest behavior data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the personalized recognition and recommendation method based on the knowledge graph are implemented in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the personalized recognition and recommendation method based on knowledge graph described in any one of claims 1 to 7 are implemented.

Images