Context semantic recognition method and system based on user behaviors

Abstract

The invention relates to the technical field of natural language processing, and discloses a context semantic recognition method and system based on user behaviors, and the method comprises the steps: collecting the multi-source behavior data of a target user, and extracting the semantic features of the user behaviors; constructing a semantic association graph of the user behaviors, detecting semantic understanding defects of the user behaviors, and setting a defect correction mechanism of the semantic understanding defects; semantic modes of user behaviors in different scenes are identified, and real intentions of a target user in different scenes are analyzed; constructing a causal relationship graph of the user behaviors, analyzing driving factors of the user behaviors, and generating behavior intention paraphrases of the target user; and in combination with the semantic association map, the defect correction mechanism and the behavior intention paraphrasing, executing context semantic recognition processing of the user behavior to obtain a semantic recognition result. According to the method, the implicit intention of the user behavior can be understood, and the accuracy of context semantic recognition is improved.

Description

TECHNICAL FIELD

[0001] The application relates to a context semantic recognition method and system based on user behavior, and belongs to the technical field of natural language processing. BACKGROUND

[0002] The context semantic based on user behavior refers to the real meaning and intention expressed by a user inferred in combination with user behavior and various background information related to the behavior, for example, a user searches for a nearby hotel at 10 o'clock at night, in combination with time and location information, the system can infer that the user may have the demand for overnight accommodation and hopes to find a suitable hotel near the current location. In the current digital era, Internet technology continues to rapidly innovate and profoundly changes people's life and work patterns. With the wide popularization of front-line technologies such as the Internet of Things, mobile Internet, and big data, the data size in the global range is rapidly expanding at an exponential rate. Behind this massive amount of data, the deep understanding of user real intention and context semantic becomes a key challenge.

[0003] The traditional context semantic recognition method mainly adopts a keyword matching method, which directly searches for a keyword input by a user in text data, and then determines semantics according to a pre-set rule or a word library. Although this method can return a result in a short time and meet the instant demand of a user, the method can only retrieve information according to the literal matching of a keyword, and can not understand the specific meaning and implied intention of a word in different context semantic contexts. For example, a user searches for 'lose weight', which can mean that the user wants to understand a weight loss method or wants to purchase a weight loss related product, but the keyword matching method can hardly accurately determine the specific demand of the user from a single 'lose weight' keyword.

[0004] Therefore, a solution is urgently needed to understand the implied intention of user behavior and improve the accuracy of context semantic recognition. SUMMARY

[0005] The application provides a context semantic recognition method and system based on user behavior, which mainly aims to understand the implied intention of user behavior and improve the accuracy of context semantic recognition.

[0006] To achieve the above-mentioned purpose, the application provides a context semantic recognition method based on user behavior, which comprises the following steps: acquiring a target user and corresponding user behavior of the target user, collecting multi-source behavior data of the target user, wherein the multi-source behavior data comprises operation behavior data, interaction behavior data, external environment data and multi-modal data, extracting a continuous behavior sequence of the target user from the multi-source behavior data, and identifying external environment data of the user behavior; simulate a context scenario of the user behavior based on the continuous behavior sequence and the external environment data, and extract semantic features of the user behavior according to the context scenario; identify entity objects and correlation relationships in the multi-source behavior data, construct a semantic correlation graph of the user behavior according to the entity objects and the correlation relationships, detect semantic understanding defects of the user behavior based on the semantic correlation graph, and set a defect correction mechanism of the semantic understanding defects; extract corrected semantics of the user behavior based on the defect correction mechanism, identify semantic modes of the user behavior in different scenarios according to the semantic features and the corrected semantics, and analyze real intentions of the target user in the different scenarios based on the semantic modes; construct a causal relationship graph of the user behavior based on the semantic correlation graph, analyze driving factors of the user behavior according to the causal relationship graph, and generate behavior intention interpretation of the target user based on the driving factors and the real intentions; perform context semantic recognition processing of the user behavior in combination with the semantic correlation graph, the defect correction mechanism, and the behavior intention interpretation, and obtain a semantic recognition result.

[0007] Optionally, the simulating the context scenario of the user behavior based on the continuous behavior sequence and the external environment data comprises: identifying a target user corresponding to the continuous behavior sequence, and analyzing a behavior mode of the target user based on the continuous behavior sequence; extracting high-frequency behavior, periodic behavior, and associated behavior of the target user from the behavior mode; analyzing a behavior motive of the target user according to the high-frequency behavior, the periodic behavior, and the associated behavior; extracting external environment features of the user behavior based on the external environment data; simulating the context scenario of the user behavior according to the external environment features and the behavior motive.

[0008] Optionally, the extracting the semantic features of the user behavior according to the context scenario comprises: identifying behavior semantic features, motive semantic features, and external environment semantic features of the user behavior according to the context scenario; generating perceptual semantic features of the context scenario in combination with the behavior semantic features, the motive semantic features, and the external environment semantic features; collecting user feedback data in the context scenario; verify a reliability degree of the perception semantic feature based on the user feedback data; perform feature optimization processing of the perception semantic feature according to the reliability degree, to obtain a feature optimization result; extract a semantic feature of the user behavior from the perception semantic feature when the feature optimization result meets a preset effect.

[0009] Optionally, the constructing the semantic association graph of the user behavior according to the entity object and the association relationship comprises: constructing a basic relationship network of the user behavior according to the entity object and the association relationship; identifying a direct relationship between the entity objects based on the basic relationship network; analyzing an indirect relationship between the entity objects according to the direct relationship; generating a relationship expansion network of the user behavior based on the direct relationship and the indirect relationship; setting a semantic hierarchy of the entity object and the association relationship according to the relationship expansion network; constructing the semantic association graph of the user behavior in combination with the relationship expansion network and the semantic hierarchy.

[0010] Optionally, the detecting the semantic understanding defect of the user behavior based on the semantic association graph comprises: identifying a potential semantic of the user behavior based on the semantic association graph; collecting explicit semantic content corresponding to the user behavior, and calculating semantic similarity between the explicit semantic content and the potential semantic by using the following formula: ; wherein, the semantic similarity between the explicit semantic content and the potential semantic, a vector representation of the explicit semantic content, a vector representation of the potential semantic, N represents a vector dimension corresponding to the explicit semantic content and the potential semantic, and j represents a vector dimension index corresponding to the explicit semantic content and the potential semantic, a value of the jth dimension of the explicit semantic content vector, a value of the jth dimension of the potential semantic vector; identifying a semantic deviation performance of the user behavior according to the semantic similarity, and locating an initial inducing position of the semantic deviation performance; identifying a behavior context of the semantic deviation performance based on the initial inducing position; analyzing a potential inducing cause of the semantic deviation performance according to the behavior context; detect a semantic understanding defect of the user behavior based on the potential inducing cause and the initial inducing position.

[0011] Optionally, the defect correction mechanism of the semantic understanding defect is set based on the semantic association graph, including: a conflict semantic corresponding to the semantic understanding defect is extracted; a semantic relationship and a context of the conflict semantic are identified based on the semantic association graph; a defect formation cause of the conflict semantic is analyzed according to the semantic relationship and the context; a defect type of the conflict semantic is identified, and a defect repair method of the semantic understanding defect is set in combination of the defect type and the defect formation cause; the defect correction mechanism of the semantic understanding defect is set based on the defect repair method.

[0012] Optionally, the semantic pattern of the user behavior in different scenarios is identified according to the semantic feature and the corrected semantic, including: a behavior intention of the user behavior is analyzed according to the semantic feature and the corrected semantic; context information of the user behavior is extracted based on the behavior intention; a scenario intention of the user behavior is generated according to the context information and the behavior intention; behavior performance data corresponding to the user behavior is collected, and a behavior sequence of the behavior performance data is extracted; a general behavior pattern of the user behavior is analyzed based on the behavior sequence; a semantic pattern of the user behavior in different scenarios is identified according to the general behavior pattern and the scenario intention.

[0013] Optionally, the causal relationship graph of the user behavior is constructed based on the semantic association graph, including: behavior performance, context and behavior target are extracted from the semantic association graph; semantic association among the behavior performance, the behavior target and the context is analyzed; association data among the behavior performance, the behavior target and the context is collected according to the semantic association; causal relationship among the user behaviors is extracted from the association data; the causal relationship graph of the user behavior is constructed based on the behavior performance, the behavior target, the context and the causal relationship.

[0014] Optionally, the generating of the behavior intention interpretation of the target user based on the driving factor and the real intention comprises: performing a factor classification processing of the driving factor to obtain a classified driving factor; identifying a direct behavior trigger point and an internal behavior motivation of the target user according to the classified driving factor; performing intention layering on the real intention based on the direct behavior trigger point and the internal behavior motivation to obtain a surface intention and a deep intention; identifying a layered behavior mode of the target user under the surface intention and the deep intention; extracting a key behavior feature of the target user based on the layered behavior mode; identifying a personalized behavior feature of the target user from the key behavior feature; generating the behavior intention interpretation of the target user in combination with the surface intention, the deep intention and the personalized behavior feature.

[0015] To solve the above problems, the application further provides a context semantic recognition system based on user behavior, which comprises: a data collection module for acquiring a target user and corresponding user behavior, collecting multi-source behavior data of the target user, wherein the multi-source behavior data comprises operation behavior data, interaction behavior data, external environment data and multi-modal data, extracting a continuous behavior sequence of the target user from the multi-source behavior data, and identifying external environment data of the user behavior; a semantic feature extraction module for simulating a context scene of the user behavior based on the continuous behavior sequence and the external environment data, and extracting semantic features of the user behavior according to the context scene; a semantic defect correction module for identifying entity objects and associated relationships in the multi-source behavior data, constructing a semantic association graph of the user behavior according to the entity objects and the associated relationships, detecting semantic understanding defects of the user behavior based on the semantic association graph, and setting a defect correction mechanism for the semantic understanding defects; an intention recognition module for extracting corrected semantics of the user behavior based on the defect correction mechanism, identifying semantic modes of the user behavior in different scenes according to the semantic features and the corrected semantics, and analyzing real intentions of the target user in the different scenes based on the semantic modes; an intention interpretation module for constructing a causal relationship graph of the user behavior based on the semantic association graph, analyzing driving factors of the user behavior according to the causal relationship graph, and generating a behavior intention interpretation of the target user based on the driving factors and the real intention. a context semantic recognition module, configured to perform context semantic recognition processing on the user behavior in combination with the semantic association graph, the defect correction mechanism and the behavior intention interpretation, to obtain a semantic recognition result.

[0016] Compared with the problems described in the background art, the embodiments of the present application can comprehensively understand the behavior patterns and preferences of the target user by collecting multi-source behavior data of the target user, and can provide the time sequence of the user behavior and the correlation between behaviors by extracting the continuous behavior sequence of the target user from the multi-source behavior data and identifying the external environment data of the user behavior, thereby helping the system to restore the context scenario in which the user behavior occurs; further, the embodiments of the present application can more accurately understand the intention and demand behind the user behavior by simulating the context scenario of the user behavior based on the continuous behavior sequence and the external environment data, helping researchers and developers to analyze user behavior from a more macro perspective, helping to extract semantic features of the user behavior, and thereby more accurately understanding the true intention behind the user behavior; the embodiments of the present application can more comprehensively capture semantic information of user behavior by constructing a semantic association graph of the user behavior according to the entity object and the association relationship, helping the system to more deeply understand user intention; further, the embodiments of the present application can detect semantic understanding defects of the user behavior based on the semantic association graph, discover misunderstanding or omission of the system to the user behavior, thereby setting a defect correction mechanism for the semantic understanding defects, correcting the user's wrong expression, and thereby improving the accuracy of semantic understanding; the embodiments of the present application can more accurately understand the intention and meaning behind the user behavior by identifying the semantic patterns of the user behavior in different scenarios according to the semantic features and the corrected semantics, and can provide customized services for the user according to the user's personalized semantic patterns; further, the embodiments of the present application can deeply understand the user's demand in a specific scenario by analyzing the real intention of the target user in the different scenarios based on the semantic patterns, thereby improving user experience; the embodiments of the present application can associate user behavior with context information by constructing a causal relationship graph of the user behavior based on the semantic association graph, enhance the understanding of behavior semantics, predict future user behavior, and analyze the driving factors of the user behavior according to the causal relationship graph, which can help to determine the root motivation driving the user behavior and predict the future behavior trend of the user; further, the embodiments of the present application can help the system to more accurately understand the semantics of the user input text or behavior by generating the behavior intention interpretation of the target user based on the driving factors and the real intention, thereby improving the accuracy of semantic recognition;Finally, by combining the semantic association graph, the defect correction mechanism, and the behavioral intent interpretation, this embodiment of the invention performs contextual semantic recognition processing of the user behavior to obtain semantic recognition results. This captures the relationships between words and the contextual context, deeply uncovering the implicit motivations behind user behavior, avoiding misjudgments caused by simply relying on literal keyword matching, and improving the accuracy of semantic understanding. Combined with the defect correction mechanism, it can effectively handle noise, ambiguity, or errors in the input, improving the robustness of semantic recognition, thereby providing users with more accurate and intelligent semantic recognition services. Therefore, the contextual semantic recognition method and system based on user behavior provided by this embodiment of the invention can understand the implicit intent of user behavior and improve the accuracy of contextual semantic recognition. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a contextual semantic recognition method based on user behavior, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a module for implementing a contextual semantic recognition method based on user behavior, provided in an embodiment of the present invention.

[0018] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] This application provides a contextual semantic recognition method based on user behavior. The executing entity of this user behavior-based contextual semantic recognition method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the user behavior-based contextual semantic recognition method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0021] Example 1: Reference Figure 1 The diagram shown is a flowchart illustrating a contextual semantic recognition method based on user behavior according to an embodiment of the present invention. In this embodiment, the contextual semantic recognition method based on user behavior includes: S1, acquire a target user and corresponding user behavior of the target user, collect multi-source behavior data of the target user, wherein the multi-source behavior data includes operation behavior data, interaction behavior data, external environment data and multi-modal data, extract continuous behavior sequence of the target user from the multi-source behavior data, and identify external environment data of the user behavior.

[0022] The embodiment of the application can provide rich data support for context semantic recognition by acquiring a target user and corresponding user behavior of the target user, wherein the target user refers to a user group or individual that is focused on and analyzed in context semantic recognition, and the user behavior refers to various behavior actions exhibited by the target user in a specific scenario, which can be explicit (such as clicking, purchasing, inputting text, etc.) or implicit (such as dwell time, browsing path, expression change, etc.).

[0023] Further, the embodiment of the application can comprehensively understand the behavior mode and preference of the user by collecting multi-source behavior data of the target user, wherein the multi-source behavior data refers to a collection of user behavior data from multiple different sources or types, including operation behavior data (such as clicking, purchasing, inputting text, etc.), interaction behavior data (such as voice instructions, gesture recognition results), external environment data (including time, location, device state information when the user behavior occurs, etc.) and multi-modal data (such as voice, video, image, text, etc.).

[0024] The embodiment of the application can provide the time sequence of the user behavior and the correlation between behaviors by extracting continuous behavior sequence of the target user from the multi-source behavior data and identifying external environment data of the user behavior, thereby helping the system restore the context scenario in which the user behavior occurs, wherein the continuous behavior sequence refers to a collection of a series of behaviors in a certain time range in the order of occurrence, such as the user logs in the platform at 9:00 am, searches for “sports shoes” at 9:02, and browses the first five product pages in the search results from 9:03 to 9:08, and the external environment data refers to external conditions and background information when the user behavior occurs, such as time, location, device state, etc.

[0025] In an optional embodiment of the application, the following formula is used to extract the continuous behavior sequence of the target user from the multi-source behavior data: ; Wherein, represents the continuous behavior sequence of the target user, represents the i-th data point in the multi-source behavior data, A represents a collection of multi-source behavior data, t represents a time stamp for extracting the continuous behavior sequence, and T represents a time range of the continuous behavior sequence to be extracted near the time stamp t. Representing data points of the time stamp.

[0026] It needs to be explained that when extracting a continuous behavior sequence , t-T represents pushing T time units from the reference time stamp t, for example, if T is 5 minutes, then t-T is 5 minutes before t, all data points whose time stamp timestamp( ) satisfies t-T≤timestamp( )≤t will be included in the sequence .

[0027] S2, based on the continuous behavior sequence and the external environment data, simulating the context scenario of the user behavior, according to the context scenario, extracting the semantic features of the user behavior.

[0028] The embodiment of the application can more accurately understand the intention and demand behind the user behavior by simulating the context scenario of the user behavior based on the continuous behavior sequence and the external environment data, which helps researchers and developers to analyze user behavior from a more macro perspective, and the context scenario refers to the comprehensive description of the specific environment, situation and background when the user behavior occurs, including but not limited to the time, place, environmental conditions, state of the user.

[0029] As an embodiment of the application, the simulation of the context scenario of the user behavior based on the continuous behavior sequence and the external environment data includes: identifying the target user corresponding to the continuous behavior sequence, and analyzing the behavior pattern of the target user based on the continuous behavior sequence; extracting the high-frequency behavior, periodic behavior and associated behavior of the target user from the behavior pattern; analyzing the behavior motivation of the target user according to the high-frequency behavior, the periodic behavior and the associated behavior; extracting the external environment features of the user behavior based on the external environment data; and simulating the context scenario of the user behavior according to the external environment features and the behavior motivation.

[0030] The behavior mode refers to a behavior characteristic and law of a user in a certain time range, the high-frequency behavior refers to a behavior frequently appearing in a continuous behavior sequence, for example, the user opens a music application and plays music every night, the periodic behavior refers to a behavior repeatedly appearing in a certain time interval, for example, the user checks work emails every Monday to Friday in the morning, the associated behavior refers to behaviors associated with each other in a behavior sequence, for example, after purchasing an electronic product, the user usually searches for related accessories (such as a mobile phone case, earphones, etc.), the behavior motivation refers to an internal reason or demand driving the user to perform a certain behavior, for example, the behavior motivation of the user searching for a fitness course may be to maintain health or lose weight, and the external environment feature refers to external conditions and background information when the user behavior occurs, including time characteristics, location characteristics, environmental conditions, device states, and the like.

[0031] Optionally, the behavior mode analysis of the target user based on the continuous behavior sequence can be implemented by using an Apriori algorithm, the high-frequency behavior of the target user extracted from the behavior mode can be obtained by using a frequent pattern mining algorithm, such as an FP-Growth algorithm, the associated behavior of the target user extracted from the behavior mode can be implemented by using an association rule mining algorithm, such as an Apriori algorithm, and the behavior motivation analysis of the target user according to the high-frequency behavior, the periodic behavior and the associated behavior can be determined by using a Transformer deep learning model.

[0032] Further, according to the context scene, the semantic features of the user behavior can be extracted, which can help the system to more accurately understand the real intention behind the user behavior, the semantic features refer to features extracted from the user behavior and the context scene in which the user behavior is located, which can reflect the meaning and intention of the user behavior, including but not limited to semantic information of the user behavior, time characteristics, environmental characteristics, user state characteristics, interaction object characteristics, and behavior mode characteristics.

[0033] As an embodiment of the present application, according to the context scene, the semantic features of the user behavior are extracted, including: according to the context scene, the behavior semantic features, motivation semantic features and external environment semantic features of the user behavior are identified; the perception semantic features of the context scene are generated by combining the behavior semantic features, the motivation semantic features and the external environment semantic features; user feedback data in the context scene is collected; the reliability of the perception semantic features is verified based on the user feedback data; according to the reliability, the feature optimization processing of the perception semantic features is performed to obtain a feature optimization result; when the feature optimization result meets a preset effect, the semantic features of the user behavior are extracted from the perception semantic features.

[0034] The behavior semantic feature refers to a feature extracted from a user behavior itself and capable of reflecting a behavior meaning, including but not limited to a behavior frequency, a behavior type and a behavior content, the motivation semantic feature refers to a feature inferred from a user behavior and capable of reflecting a motivation behind the user behavior, for example, a motivation of a user to purchase a fitness equipment can be "exercising at home", the external environment semantic feature refers to semantic information extracted from external environment data, such as a rest time and a weather condition, the perception semantic feature refers to a context perception semantic feature capable of reflecting a user behavior and generated by combining the behavior semantic feature, the motivation semantic feature and the external environment semantic feature, for example, a user searches "relaxing music" (the behavior semantic feature) at 10 o'clock at night (the external environment feature), a motivation can be "relaxing mood and preparing for sleep" (the motivation semantic feature), and a perception semantic feature generated by the system in combination with the information can be "relaxing before sleep", the user feedback data refers to feedback information of a user to a service or recommended content provided by the system, for example, the user gives a high score to the recommended relaxing music, the reliability degree refers to accuracy and credibility of the perception semantic feature, the feature optimization processing refers to a process of adjusting and optimizing the perception semantic feature according to the user feedback data, the feature optimization result refers to the perception semantic feature obtained after the optimization processing, and the preset effect refers to a target effect for judging whether the feature optimization result meets a requirement.

[0035] Optionally, the motivation semantic feature recognition of the user behavior according to the context scene can be realized by using a psychological theory, the perception semantic feature generation of the context scene in combination with the behavior semantic feature, the motivation semantic feature and the external environment semantic feature can be realized by using a generation model, such as a generative adversarial network (GAN) model, and the reliability degree verification of the perception semantic feature based on the user feedback data can be determined by using a logistic regression model.

[0036] S3, an entity object and an association relationship in the multi-source behavior data are recognized, a semantic association graph of the user behavior is constructed according to the entity object and the association relationship, a semantic understanding defect of the user behavior is detected based on the semantic association graph, and a defect correction mechanism of the semantic understanding defect is set.

[0037] By recognizing the entity object and the association relationship in the multi-source behavior data, the embodiment of the application can construct a knowledge graph of the user behavior, and provide a structured background knowledge for an intelligent system. The entity object refers to an object or concept with independent meaning and recognizable in the multi-source behavior data, including but not limited to a user, a commodity, an event and a concept. The association relationship refers to a logical or semantic connection between the entity objects, such as a relationship between a user and a commodity and a relationship between commodities.

[0038] Optionally, the entity object recognition in the multi-source behavior data can be implemented by using a dictionary matching method, and the association relationship recognition in the multi-source behavior data can be obtained by using a neural network model, such as an LSTM neural network model.

[0039] Further, the embodiment of the present application can more comprehensively capture semantic information of user behavior by constructing a semantic association graph of the user behavior according to the entity objects and the association relationships, and help the system to more deeply understand the user intent, and the semantic association graph refers to a structured knowledge graph for representing entity objects and semantic relationships therebetween.

[0040] As an embodiment of the present application, the constructing a semantic association graph of the user behavior according to the entity objects and the association relationships comprises: constructing a basic relationship network of the user behavior according to the entity objects and the association relationships; identifying a direct relationship between the entity objects based on the basic relationship network; analyzing an indirect relationship between the entity objects according to the direct relationship; generating a relationship expansion network of the user behavior based on the direct relationship and the indirect relationship; setting a semantic hierarchy of the entity objects and the association relationships according to the relationship expansion network; and constructing the semantic association graph of the user behavior in combination with the relationship expansion network and the semantic hierarchy.

[0041] The basic relationship network refers to a network formed by taking entity objects as nodes and taking the most basic and directly observable association relationships therebetween as connecting edges, the direct relationship refers to a relationship between two entity objects directly connected in the basic relationship network, for example, in an e-commerce scenario, a user directly clicks on a certain product detail page, and there is a direct relationship of “click” between the two entity objects of “user” and “product detail page”, the indirect relationship refers to a contact indirectly established through other entity objects and direct relationships, for example, user A purchases product X, product X is provided by merchant Y, and merchant Y also sells product Z, so there is an indirect relationship between user A and product Z, the relationship expansion network refers to a more comprehensive network formed by comprehensively integrating the direct relationship and the indirect relationship on the basis of the basic relationship network, and the semantic hierarchy refers to abstracting low-level and specific entities and relationships into high-level and more general semantic categories, for example, classifying “purchase” as “consumption behavior”.

[0042] Optionally, the constructing of the basic relation network of the user behavior according to the entity objects and the association relations can be implemented by using a NetworkX library, the direct relation identification between the entity objects based on the basic relation network can be acquired by using an edge traversal algorithm of the NetworkX library, the indirect relation analysis between the entity objects according to the direct relation can be determined by using a reasoning tool, such as an Apache Jena reasoning tool, and the semantic hierarchy setting of the entity objects and the association relations according to the relation expansion network can be implemented by using a clustering algorithm, such as a hierarchical clustering algorithm.

[0043] By detecting the semantic understanding defects of the user behavior based on the semantic association graph, the embodiment of the present application can find the misunderstanding or omission of the system for the user behavior, thereby improving the accuracy of semantic understanding. The semantic understanding defects refer to the insufficient, wrong or biased cases in the semantic understanding of the user behavior, such as the neglect of the causal relationship between "browsing" and "purchasing".

[0044] As an embodiment of the present application, the detecting of the semantic understanding defects of the user behavior based on the semantic association graph comprises: identifying the potential semantics of the user behavior based on the semantic association graph; collecting the explicit semantic content corresponding to the user behavior, and calculating the semantic similarity between the explicit semantic content and the potential semantics; identifying the semantic bias performance of the user behavior according to the semantic similarity, and locating the initial evoking position of the semantic bias performance; identifying the behavior context of the semantic bias performance based on the initial evoking position; analyzing the potential evoking reason of the semantic bias performance according to the behavior context; and detecting the semantic understanding defects of the user behavior based on the potential evoking reason and the initial evoking position.

[0045] The potential semantics refers to the semantic information implied in the user behavior and not directly expressed, for example, when the user searches for "apple", the potential semantics may be to understand the nutritional value of the apple fruit, the purchase channel, or the product information related to the Apple company, and the explicit semantic content refers to the semantic information explicitly expressed in the user behavior, which is usually the surface meaning carried by the text, operation, etc. directly input, selected or displayed by the user, the semantic similarity refers to the closeness of the explicit semantic content and the potential semantics in the semantic level, the semantic deviation performance refers to the specific manifestation of the difference between the explicit semantic content and the potential semantics observed according to the calculation result of the semantic similarity, for example, the semantic deviation performance may be that the explicit semantic content of the user focuses more on the appearance description of the apple, and the potential semantics is mainly about the edible value of the apple, the initial inducing position refers to the specific point that first induces the semantic deviation to appear in the user behavior process, which can be a specific character input by the user, an operation performed, a time point or a specific behavior step, etc., the behavior context refers to the specific environment and background information when the user behavior related to the semantic deviation occurs, including the operation history of the user, the current interface display, the related context text, the time, the place and other factors, and the potential inducing reason refers to the deep and internal reason for the generation of the semantic deviation, for example, the word "apple" is simply understood as the product of the Apple company, and the meaning as a fruit is ignored.

[0046] Optionally, the potential semantic recognition of the user behavior based on the semantic association graph can be obtained through a graph neural network, such as a GAT graph neural network, the semantic deviation performance recognition of the user behavior according to the semantic similarity can be realized by using an anomaly detection algorithm, such as a LOF (Local Outlier Factor) algorithm, the behavior context recognition of the semantic deviation performance based on the initial inducing position can be determined through a sequence model, such as a recurrent neural network model, and the potential inducing reason analysis of the semantic deviation performance according to the behavior context can be realized by using a counterfactual analysis method.

[0047] In an optional embodiment of the application, the semantic similarity of the explicit semantic content and the potential semantics is calculated by using the following formula: ; Wherein, represents the semantic similarity of the explicit semantic content and the potential semantics, represents the vector representation of the explicit semantic content, represents the vector representation of the potential semantics, N represents the vector dimension corresponding to the explicit semantic content and the potential semantics, and j represents the vector dimension index corresponding to the explicit semantic content and the potential semantics, a value representing a jth dimension of an explicit semantic content vector, a value representing a jth dimension of a latent semantic vector.

[0048] Further, the embodiment of the present application can correct the user's error expression, improve the accuracy of semantic recognition, and optimize the context understanding by setting the defect correction mechanism of the semantic understanding defect based on the semantic association graph.

[0049] As an embodiment of the present application, the setting of the defect correction mechanism of the semantic understanding defect based on the semantic association graph comprises: extracting the conflict semantics corresponding to the semantic understanding defect; identifying the semantic relationship and the context of the conflict semantics based on the semantic association graph; analyzing the defect formation reason of the conflict semantics according to the semantic relationship and the context; identifying the defect type of the conflict semantics, and setting the defect repair mode of the semantic understanding defect in combination with the defect type and the defect formation reason; and setting the defect correction mechanism of the semantic understanding defect based on the defect repair mode.

[0050] The conflict semantics refers to the part corresponding to the semantic understanding defect which is contradictory or inconsistent with the context logic, for example, the conflict semantics in the sentence "I walk in the quiet city" is "quiet city", the semantic relationship refers to the association relationship of the conflict semantics in the semantic association graph, such as the hyponym relationship, synonym relationship, cause-effect relationship, etc., the context refers to the specific environment and background of the conflict semantics, the defect formation reason refers to various factors leading to the semantic understanding defect, such as the error understanding of some rare words or polysemous words, the defect type refers to the classification of the semantic understanding defect, and common types include vocabulary error, syntax error, semantic ambiguity, semantic contradiction, and context inconsistency, etc., and the defect repair mode refers to the repair strategy formulated according to the defect type and the defect formation reason, for example, for syntax error, the sentence structure needs to be adjusted, and the missing components need to be supplemented to make the sentence comply with the grammar rules; for semantic ambiguity, the ambiguity can be eliminated by clearly indicating the object, supplementing the modifier, etc.

[0051] Optionally, the semantic relationship identification of the conflict semantics based on the semantic association graph can be implemented by using dependency syntax analysis, the context context identification of the conflict semantics can be obtained by using the attention mechanism of a deep learning model, the defect formation reason analysis of the conflict semantics according to the semantic relationship and the context context can be implemented by using a comparative analysis method, for example, comparing the correct semantic understanding and the defective semantic understanding, analyzing the difference between the two, and thus finding out the reason for the defect formation, the defect type identification of the conflict semantics can be obtained by using a classification model, such as a decision tree model, and the defect repair method setting of the semantic understanding defect in combination with the defect type and the defect formation reason can be implemented by using a rule engine, such as replacing the wrong vocabulary with the correct vocabulary for vocabulary errors, and formulating a grammar correction rule to adjust the sentence structure, supplement or delete necessary components for syntax errors.

[0052] S4, based on the defect correction mechanism, extract the corrected semantics of the user behavior, identify the semantic mode of the user behavior in different scenes according to the semantic features and the corrected semantics, and analyze the real intention of the target user in the different scenes based on the semantic mode.

[0053] The embodiment of the present application can help to analyze the commonness and difference of user behavior in different scenes by extracting the corrected semantics of the user behavior based on the defect correction mechanism, and then summarize the corresponding semantic mode. The corrected semantics refers to the accurate semantic information obtained by adjusting, perfecting and clarifying the inaccurate, ambiguous, ambiguous or incomplete semantics in the user behavior, for example, the user says "I want that", which is not clear in semantics, and it is not known what the user wants specifically. However, if the context is combined, such as mentioning some objects before or using other information in the dialogue, the defect correction mechanism is used to determine that "that" said by the user refers to "the red cup placed on the table", and "the red cup placed on the table" is the corrected semantics.

[0054] Further, the embodiment of the present application can more accurately understand the intention and meaning behind the user behavior by identifying the semantic mode of the user behavior in different scenes according to the semantic features and the corrected semantics, and providing customized services for the user according to the personalized semantic mode of the user. The semantic mode refers to a patternized expression formed by abstracting and generalizing the semantic information implied by the user behavior in different scenes.

[0055] As an embodiment of the present application, the identifying the semantic mode of the user behavior in different scenarios according to the semantic features and the revised semantics comprises: analyzing the behavior intention of the user behavior according to the semantic features and the revised semantics; extracting the context information of the user behavior based on the behavior intention; generating the scenario intention of the user behavior according to the context information and the behavior intention; collecting the behavior performance data corresponding to the user behavior and extracting the behavior sequence of the behavior performance data; analyzing the general behavior mode of the user behavior based on the behavior sequence; and identifying the semantic mode of the user behavior in different scenarios according to the general behavior mode and the scenario intention.

[0056] The behavior intention refers to the internal purpose or motivation of the user when performing a certain behavior, the context information refers to the environment or background information of the user behavior, including time, place, scene, device, etc., the scenario intention refers to the specific performance of the behavior intention in a specific context, for example, the behavior intention is "social interaction", the context is "travel scenario", and the scenario intention is "sharing travel experience", the behavior performance data refers to various data that can reflect the user behavior, including operation records, event logs, sensor data, etc. generated in the process of user interaction with the system or application program, the behavior sequence refers to an ordered set of a series of user behaviors arranged in chronological order extracted from the behavior performance data, and the general behavior mode refers to a universal and representative behavior mode obtained by analyzing and abstracting the behavior sequence. For example, the general mode in the shopping scenario is browse → compare → decide → buy, and the general mode in the social scenario is post → interact → share.

[0057] Optionally, the behavior intention analysis of the user behavior according to the semantic features and the revised semantics can be implemented by using a pre-trained language model, such as a BERT language model, the scenario intention generation of the user behavior according to the context information and the behavior intention can be obtained by using an Easy Rules tool, and the general behavior mode analysis of the user behavior based on the behavior sequence can be implemented by using a Scikit-learn tool.

[0058] The embodiment of the present application can deeply understand the demand of the user in a specific scene, improve the user experience, and analyze the real intention of the target user in different scenes based on the semantic pattern, wherein the real intention refers to the purpose, motivation or expectation that the target user really wants to achieve when performing the behavior in a specific scene, for example, the user posts a scenic photo on social media, the real intention may be to share life, show his own photography works, or want to obtain the attention and praise of others, and the different scenes refer to various different backgrounds, conditions or context environments when the user behavior occurs, such as environmental scene, application scene, device scene, etc.

[0059] Optionally, the analysis of the real intention of the target user in different scenes based on the semantic pattern can be implemented by using an inference model, such as rule-based reasoning (RBR).

[0060] S5, a causal relationship graph of the user behavior is constructed based on the semantic association graph, the driving factors of the user behavior are analyzed according to the causal relationship graph, and the behavior intention interpretation of the target user is generated based on the driving factors and the real intention.

[0061] The embodiment of the present application can associate the user behavior with the context information, enhance the understanding of the behavior semantics, predict the future behavior of the user, and optimize the recommendation system or the user experience by constructing the causal relationship graph of the user behavior based on the semantic association graph, wherein the causal relationship graph refers to a graph structure describing the causal relationship between the elements such as user behavior, context and intention.

[0062] As an embodiment of the present application, constructing the causal relationship graph of the user behavior based on the semantic association graph comprises: extracting behavior performance, context and behavior target from the semantic association graph; analyzing the semantic association between the behavior performance, the behavior target and the context; collecting the associated data between the behavior performance, the behavior target and the context according to the semantic association; extracting the causal relationship between the user behaviors from the associated data; and constructing the causal relationship graph of the user behavior based on the behavior performance, the behavior target, the context and the causal relationship.

[0063] The behavior performance refers to specific behaviors exhibited by the user in an actual scene, such as browsing goods, adding to a shopping cart, placing an order, etc., the context refers to the environment or background information in which the user behavior occurs, including time, place, scene, device, etc., the behavior target refers to the motivation behind the user behavior, which is the driving force of the user behavior, the semantic association refers to the semantic relationship between the user behavior, the context and the behavior target, the associated data refers to the structured data between the user behavior, the context and the behavior intention, and the causal relationship refers to the relationship between the reasons and results existing between the user behaviors, that is, a behavior performance or a context factor is the reason for achieving another behavior or a behavior target.

[0064] Optionally, the extraction of the behavior performance, the context and the behavior target from the semantic association graph can be realized by using a graph traversal algorithm, the semantic association analysis between the behavior performance, the behavior target and the context can be obtained by using a graph embedding tool, such as a PyTorch Geometric tool, and the extraction of the causal relationship between the user behaviors from the associated data can be realized by using a causal inference algorithm, such as a do-calculus algorithm of Pearl.

[0065] Further, the embodiments of the present application can help to determine the root motivation driving the user behavior and predict the future behavior trend of the user by analyzing the driving factors of the user behavior according to the causal relationship graph, the driving factors refer to various factors that promote the user to produce a certain behavior, such as safety needs, social needs, self-actualization needs, etc.

[0066] Optionally, the driving factor analysis of the user behavior according to the causal relationship graph can be determined by using a logistic regression model.

[0067] The embodiments of the present application can help the system to more accurately understand the semantics of the text or behavior input by the user and improve the accuracy of semantic recognition by generating the behavior intention interpretation of the target user based on the driving factors and the true intention, the behavior intention interpretation refers to the conversion of the behavior, driving factors and intention of the user into a semantic representation that can be understood by a computer system or an analyst.

[0068] As an embodiment of the present application, the generating of the behavior intention interpretation of the target user based on the driving factor and the real intention comprises: performing factor classification processing of the driving factor to obtain classified driving factors; identifying a direct behavior trigger point and an internal behavior motivation of the target user according to the classified driving factors; performing intention layering on the real intention based on the direct behavior trigger point and the internal behavior motivation to obtain a surface intention and a deep intention; identifying a layered behavior pattern of the target user under the surface intention and the deep intention; extracting key behavior characteristics of the target user based on the layered behavior pattern; identifying personalized behavior characteristics of the target user from the key behavior characteristics; and generating the behavior intention interpretation of the target user in combination with the surface intention, the deep intention and the personalized behavior characteristics.

[0069] The factor classification processing refers to the process of classifying factors driving user behavior, the classified driving factors refer to different driving factor categories after classification processing, such as external environment, internal demand, emotional state, etc., the direct behavior trigger point refers to the direct trigger reason of user behavior, which is usually external environment or explicit demand, the internal behavior motivation refers to the internal reason of user behavior, which is usually implicit demand, emotional state or value, the deep intention refers to the internal motivation or long-term goal behind user behavior, such as the deep intention of a user to buy a mobile phone may be to meet the demand for efficient communication and information processing in work, or to improve his image and status in the social circle, the surface intention refers to the direct manifestation or short-term goal of user behavior, for example, a user searches for a mobile phone on an e-commerce platform, the surface intention may be to want to know the relevant information of the mobile phone or to buy the mobile phone, the layered behavior pattern refers to the behavior pattern of the user under different intention levels (surface intention and deep intention), the key behavior characteristics refer to the representative characteristics of user behavior, which can reflect the behavior pattern and intention of the user, such as the frequency, duration and sequence of behavior, etc., and the personalized behavior characteristics refer to the behavior characteristics of the user that can reflect the unique personality, preference and habit of the target user, such as historical behavior, interest preference, etc.

[0070] Optionally, the factor classification processing of the driving factor can be implemented by using a K-means clustering algorithm, the internal behavioral motivation recognition of the target user according to the classified driving factor can be obtained by a psychological model such as Maslow's hierarchy of needs model, the intent layering of the real intent based on the direct behavioral trigger point and the internal behavioral motivation can be implemented by using a layering model such as a Neo4j graph database, the corresponding layered behavioral pattern recognition of the target user according to the surface intent and the deep intent can be implemented by using a sequence pattern mining algorithm such as an SPMF algorithm, and the personalized behavioral characteristics of the target user can be obtained by using a collaborative filtering algorithm such as a Surprise algorithm.

[0071] S6, in combination with the semantic association graph, the defect correction mechanism and the behavior intent interpretation, performing context semantic recognition processing of the user behavior to obtain a semantic recognition result.

[0072] By combining the semantic association graph, the defect correction mechanism and the behavior intent interpretation, the embodiment of the present application performs context semantic recognition processing of the user behavior to obtain a semantic recognition result, which can capture the association relationship and context between words, deeply mine the implicit motivation behind the user behavior, avoid misjudgment caused by simply relying on literal matching of keywords, improve the accuracy of semantic understanding, combine the defect correction mechanism, effectively process noise, ambiguity or errors in the input, and improve the robustness of semantic recognition, thereby providing more accurate and intelligent semantic recognition services for users. The context semantic recognition processing refers to the process of accurately understanding the real meaning and implicit intent of user input by analyzing the context information (such as context, background, user historical behavior, etc.) of the user input text, combining the semantic association graph, the defect correction mechanism and the behavior intent interpretation. The semantic recognition result refers to the final understanding conclusion obtained after analyzing the real meaning and implicit intent of the user input text through context semantic recognition processing. For example, a user inputs “new sports shoes for men, light” in the search bar, then browses multiple brand men's sports shoes product pages, also checks some user reviews, then clicks the “add to cart” button of a certain brand of sports shoes, but does not immediately settle the account, and searches for “sports socks”. After context semantic recognition, the system concludes that the user's current main intent is to buy new men's light sports shoes, and has a purchase intention for a certain brand of sports shoes and has added them to the cart. At the same time, the user may have the need to purchase matching sports socks.

[0073] Compared with the problems described in the background art, the embodiments of the present application can comprehensively understand the behavior patterns and preferences of the target user by collecting multi-source behavior data of the target user, and can provide the time sequence of the user behavior and the correlation between behaviors by extracting the continuous behavior sequence of the target user from the multi-source behavior data and identifying the external environment data of the user behavior, thereby helping the system to restore the context scenario in which the user behavior occurs; further, the embodiments of the present application can more accurately understand the intention and demand behind the user behavior by simulating the context scenario of the user behavior based on the continuous behavior sequence and the external environment data, helping researchers and developers to analyze user behavior from a more macro perspective, helping to extract semantic features of the user behavior, and thereby more accurately understanding the true intention behind the user behavior; the embodiments of the present application can more comprehensively capture semantic information of user behavior by constructing a semantic association graph of the user behavior according to the entity object and the association relationship, helping the system to more deeply understand user intention; further, the embodiments of the present application can detect semantic understanding defects of the user behavior based on the semantic association graph, discover misunderstanding or omission of the system to the user behavior, thereby setting a defect correction mechanism for the semantic understanding defects, correcting the user's wrong expression, and thereby improving the accuracy of semantic understanding; the embodiments of the present application can more accurately understand the intention and meaning behind the user behavior by identifying the semantic patterns of the user behavior in different scenarios according to the semantic features and the corrected semantics, and can provide customized services for the user according to the user's personalized semantic patterns; further, the embodiments of the present application can deeply understand the user's demand in a specific scenario by analyzing the real intention of the target user in the different scenarios based on the semantic patterns, thereby improving user experience; the embodiments of the present application can associate user behavior with context information by constructing a causal relationship graph of the user behavior based on the semantic association graph, enhance the understanding of behavior semantics, predict future user behavior, and analyze the driving factors of the user behavior according to the causal relationship graph, which can help to determine the root motivation driving the user behavior and predict the future behavior trend of the user; further, the embodiments of the present application can help the system to more accurately understand the semantics of the user input text or behavior by generating the behavior intention interpretation of the target user based on the driving factors and the real intention, thereby improving the accuracy of semantic recognition;Finally, by combining the semantic association graph, the defect correction mechanism, and the behavioral intent interpretation, this embodiment of the invention performs contextual semantic recognition processing of the user behavior to obtain semantic recognition results. This captures the relationships between words and the contextual context, deeply uncovering the implicit motivations behind user behavior, avoiding misjudgments caused by simply relying on literal keyword matching, and improving the accuracy of semantic understanding. Combined with the defect correction mechanism, it can effectively handle noise, ambiguity, or errors in the input, improving the robustness of semantic recognition, thereby providing users with more accurate and intelligent semantic recognition services. Therefore, the contextual semantic recognition method and system based on user behavior provided by this embodiment of the invention can understand the implicit intent of user behavior and improve the accuracy of contextual semantic recognition.

[0074] Example 2: like Figure 2 The diagram shown is a functional block diagram of a contextual semantic recognition system based on user behavior according to the present invention.

[0075] The contextual semantic recognition system 200 based on user behavior described in this invention can be installed in an electronic device. Depending on the functions implemented, the contextual semantic recognition system based on user behavior may include a data collection module 201, a semantic feature extraction module 202, a semantic defect correction module 203, an intent recognition module 204, an intent interpretation module 205, and a contextual semantic recognition module 206. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0076] In this embodiment of the invention, the functions of each module / unit are as follows: The data collection module 201 is used to acquire target users and their corresponding user behaviors, collect multi-source behavior data of the target users, wherein the multi-source behavior data includes operation behavior data, interaction behavior data, external environment data and multimodal data, extract the continuous behavior sequence of the target users from the multi-source behavior data, and identify the external environment data of the user behaviors; The semantic feature extraction module 202 is used to simulate the context of the user behavior based on the continuous behavior sequence and the external environment data, and extract the semantic features of the user behavior according to the context. The semantic defect correction module 203 is used to identify entity objects and relationships in the multi-source behavior data, construct a semantic association graph of the user behavior based on the entity objects and relationships, detect semantic understanding defects of the user behavior based on the semantic association graph, and set a defect correction mechanism for the semantic understanding defects. The intention recognition module 204 is configured to extract a correction semantic of the user behavior based on the defect correction mechanism, recognize a semantic mode of the user behavior in different scenes according to the semantic feature and the correction semantic, analyze a real intention of the target user in the different scenes based on the semantic mode, and analyze the real intention of the target user in the different scenes based on the semantic mode. The intention explanation module 205 is configured to construct a causal relationship graph of the user behavior based on the semantic association graph, analyze a driving factor of the user behavior according to the causal relationship graph, and generate a behavior intention interpretation of the target user based on the driving factor and the real intention. The context semantic recognition module 206 is configured to perform a context semantic recognition process of the user behavior in combination with the semantic association graph, the defect correction mechanism, and the behavior intention interpretation, and obtain a semantic recognition result.

[0077] In detail, the modules in the context semantic recognition system 200 based on user behavior in the embodiments of the present application are used in the same technical means as the context semantic recognition method based on user behavior in the above-mentioned Figure 1 embodiments, and can produce the same technical effects, which will not be described here.

[0078] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application 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 application can be modified or replaced by equivalents without departing from the spirit and scope of the present application.

Claims

1. A contextual semantic recognition method based on user behavior, characterized in that, The method includes: Acquire target users and their corresponding user behaviors, collect multi-source behavioral data of the target users, wherein the multi-source behavioral data includes operation behavior data, interaction behavior data, external environment data and multimodal data, extract the continuous behavior sequence of the target users from the multi-source behavioral data, and identify the external environment data of the user behaviors; Based on the continuous behavior sequence and the external environment data, the context of the user behavior is simulated, and the semantic features of the user behavior are extracted according to the context. Identify entity objects and relationships in the multi-source behavioral data; construct a semantic association graph of user behavior based on the entity objects and relationships; detect semantic understanding defects of user behavior based on the semantic association graph; and set a defect correction mechanism for the semantic understanding defects. Based on the defect correction mechanism, the correction semantics of the user behavior are extracted. According to the semantic features and the correction semantics, the semantic patterns of the user behavior in different scenarios are identified. Based on the semantic patterns, the true intentions of the target user in the different scenarios are analyzed. Based on the semantic association graph, a causal relationship graph of the user behavior is constructed. According to the causal relationship graph, the driving factors of the user behavior are analyzed. Based on the driving factors and the true intention, the behavioral intention interpretation of the target user is generated. By combining the semantic association graph, the defect correction mechanism, and the behavioral intent interpretation, the contextual semantic recognition processing of the user behavior is performed to obtain the semantic recognition result.

2. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The contextual scenario for simulating the user behavior based on the continuous behavior sequence and the external environment data includes: Identify the target user corresponding to the continuous behavior sequence, and analyze the behavior pattern of the target user based on the continuous behavior sequence; The high-frequency behaviors, periodic behaviors, and related behaviors of the target user are extracted from the behavioral patterns. The behavioral motivations of the target user are analyzed based on the high-frequency behavior, the periodic behavior, and the associated behavior. Based on the external environment data, extract the external environment features of the user behavior; Based on the external environmental characteristics and the behavioral motivation, simulate the context of the user behavior.

3. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The step of extracting semantic features of the user behavior based on the context includes: Based on the context, the behavioral semantic features, motivational semantic features, and external environment semantic features of the user behavior are identified. By combining the behavioral semantic features, the motivational semantic features, and the external environment semantic features, the perceptual semantic features of the context scene are generated. Collect user feedback data in the aforementioned context; Based on the user feedback data, the reliability of the perceived semantic features is verified; Based on the reliability level, perform feature optimization processing on the perceived semantic features to obtain the feature optimization result; When the feature optimization result meets the preset effect, the semantic features of the user behavior are extracted from the perceived semantic features.

4. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The step of constructing a semantic association graph of user behavior based on the entity object and the association relationship includes: Based on the entity objects and the relationships, construct the basic relationship network of the user behavior; Based on the aforementioned basic relationship network, the direct relationships between the entity objects are identified; Based on the direct relationships, analyze the indirect relationships between the entity objects; Based on the direct and indirect relationships, a relational extension network for the user behavior is generated; Based on the relationship extension network, the semantic hierarchy of the entity objects and the association relationships is set; By combining the relationship extension network and the semantic hierarchy, a semantic association graph of the user behavior is constructed.

5. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The method of detecting semantic understanding defects in user behavior based on the semantic association graph includes: Based on the semantic association graph, the potential semantics of the user behavior are identified; Collect the explicit semantic content corresponding to the user behavior, and calculate the semantic similarity between the explicit semantic content and the latent semantic content using the following formula: ； in, This represents the semantic similarity between explicit semantic content and latent semantic content. Vector representation of explicit semantic content. The vector representation of latent semantics, where N represents the vector dimension corresponding to the explicit semantic content and the latent semantics, and j represents the index of the vector dimension corresponding to the explicit semantic content and the latent semantics. This represents the value of the j-th dimension of the explicit semantic content vector. This represents the value of the j-th dimension of the latent semantic vector; Based on the semantic similarity, identify the semantic deviation manifestations of the user behavior and locate the initial triggering location of the semantic deviation manifestations; Based on the initial triggering location, identify the behavioral context in which the semantic deviation manifests; Based on the behavioral context, analyze the potential causes of the semantic deviation. Based on the potential triggering cause and the initial triggering location, semantic understanding defects in the user behavior are detected.

6. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The mechanism for correcting semantic understanding deficiencies based on the semantic association graph includes: Extract the conflicting semantics corresponding to the aforementioned semantic understanding defects; Based on the semantic association graph, the semantic relationships and context of the conflict semantics are identified; Based on the semantic relationships and the context, analyze the reasons for the defects in the conflict semantics; Identify the defect type of the conflict semantics, and combine the defect type and the cause of the defect to set a defect repair method for the semantic understanding defect; Based on the aforementioned defect repair method, a defect correction mechanism for the semantic understanding defect is established.

7. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The step of identifying semantic patterns of user behavior in different scenarios based on the semantic features and the modified semantics includes: Based on the semantic features and the modified semantics, analyze the behavioral intent of the user's behavior; Based on the stated behavioral intent, extract the contextual information of the user's behavior; Based on the context information and the behavioral intent, a contextualized intent for the user behavior is generated; Collect the behavioral performance data corresponding to the user behavior, and extract the behavioral sequence of the behavioral performance data; Based on the behavioral sequence, analyze the general behavioral patterns of the user behavior; Based on the general behavior pattern and the contextual intent, identify the semantic patterns of the user behavior in different scenarios.

8. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The step of constructing a causal relationship graph of user behavior based on the semantic association graph includes: The behavioral performance, context, and behavioral target are extracted from the semantic association graph; Analyze the semantic relationships between the behavioral performance, the behavioral goal, and the context; Based on the semantic association, collect the association data between the behavioral performance, the behavioral goal, and the context; Extract the causal relationships between the user behaviors from the associated data; Based on the behavioral performance, the behavioral goal, the context, and the causal relationship, a causal relationship graph of the user behavior is constructed.

9. The contextual semantic recognition method based on user behavior as described in claim 1, characterized in that, The process of generating a behavioral intent interpretation for the target user based on the driving factors and the true intent includes: Perform factor classification processing on the driving factors to obtain classified driving factors; Based on the classification drivers, identify the direct behavioral triggers and intrinsic behavioral motivations of the target user; Based on the direct behavioral triggers and the underlying behavioral motivations, the true intentions are layered to obtain surface intentions and deep intentions. Identify the hierarchical behavioral patterns of the target user under the surface intent and the deep intent; Based on the hierarchical behavior pattern, the key behavioral features of the target user are extracted; Identify the personalized behavioral characteristics of the target user from the key behavioral characteristics; By combining the surface intent, the deep intent, and the personalized behavioral characteristics, a behavioral intent interpretation of the target user is generated.

10. A contextual semantic recognition system based on user behavior, characterized in that, The system includes: The data collection module is used to acquire target users and their corresponding user behaviors, collect multi-source behavioral data of the target users, wherein the multi-source behavioral data includes operation behavior data, interaction behavior data, external environment data and multimodal data, extract the continuous behavior sequence of the target users from the multi-source behavioral data, and identify the external environment data of the user behaviors; The semantic feature extraction module is used to simulate the context of the user behavior based on the continuous behavior sequence and the external environment data, and extract the semantic features of the user behavior according to the context. The semantic defect correction module is used to identify entity objects and relationships in the multi-source behavioral data, construct a semantic association graph of the user behavior based on the entity objects and relationships, detect semantic understanding defects of the user behavior based on the semantic association graph, and set a defect correction mechanism for the semantic understanding defects. The intent recognition module is used to extract the corrected semantics of the user behavior based on the defect correction mechanism, identify the semantic patterns of the user behavior in different scenarios based on the semantic features and the corrected semantics, and analyze the true intent of the target user in the different scenarios based on the semantic patterns. The intent interpretation module is used to construct a causal relationship graph of the user behavior based on the semantic association graph, analyze the driving factors of the user behavior according to the causal relationship graph, and generate the behavioral intent interpretation of the target user based on the driving factors and the true intent. The context semantic recognition module is used to combine the semantic association graph, the defect correction mechanism, and the behavioral intent interpretation to perform context semantic recognition processing of the user behavior and obtain semantic recognition results.

Images