Intelligent language emotion recognition method and system

Abstract

The invention relates to the related field of natural language recognition and judgment, and discloses an intelligent language emotion recognition method and system, which are used in the understanding of a big data model on the dialogue content of a user, and are used for evaluating and obtaining the dialogue content and the viewpoint site of each speaking object in the dialogue through the dialogue understanding based on the natural language. Therefore, matching evaluation can be carried out on the speaking content of the speaker according to the viewpoint stand so as to judge whether the speaking is a conventional expression conforming to the stand or a reverse expression with emotion filling.

Description

Technical Field

[0001] The present invention relates to the field of natural language recognition and judgment, and specifically to an intelligent language emotion recognition method and system. Background Art

[0002] In daily opinion exchanges and debates with multiple parties, the emotional expression of language is diverse. In different scenarios and emotions, the same language content, even with a similar tone of expression, may convey different or even opposite meanings. For example, when two people are debating, although one user has a different view, due to the fruitless debate for some time, they may use a positive expression to recognize the other party's view. In this case, the expression may have an implicit ironic meaning.

[0003] In the prior art, natural language processing and natural language emotion processing are both applicable to scenarios with emotional and expression coherence, and cannot effectively and accurately judge this type of suddenly changing language expression with reverse meaning, which may lead to a large deviation in the semantic analysis result. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent language emotion recognition method and system to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] An intelligent language emotion recognition method includes:

[0007] Splitting and recording voice communication data according to the speaking object, and performing natural language analysis on the voice communication data to obtain its basic semantic content. The voice communication data includes multiple voice segments, and each voice segment independently corresponds to the basic semantic content;

[0008] Performing natural semantic understanding on the basic semantic content, generating several stance keywords, and performing stance statistics on the speaking object to obtain a user stance preference grid, which is used to represent multiple stance keywords of the object and the corresponding statistical fitting rate;

[0009] Obtaining the current user's speech data, identifying and judging the corresponding speaking object based on voice feature recognition, and extracting the stance of the user's speech data to obtain the stance expression to be evaluated;

[0010] Compare the to-be-evaluated stance expression with the user stance preference grid based on the speaker. If it does not match the user stance preference grid, the current user speech data contains a reverse expression meaning, and the reverse expression meaning is used to represent an emotional stance expression where the speech content is mutually exclusive with the user's actual expression.

[0011] As a further solution of the present invention: The stance keywords include self-stance keywords and other-party stance keywords. The step of statistically analyzing the stance of the speaker to obtain the user stance preference grid specifically includes:

[0012] Based on a natural language understanding model, perform semantic association analysis on multiple consecutive voice communication data, and then judge the evaluation object corresponding to the stance keyword, and establish an association relationship;

[0013] Match the evaluation object with the speaker of the current voice communication data. If they match, it is determined as a self-stance keyword. If they do not match, it is determined as an other-party stance keyword, and the evaluation object is set as the stance evaluation object;

[0014] Assign a reference coefficient for statistics according to the nature of the stance keyword, and perform statistics on the stance keyword after the reference coefficient assignment based on the belonging object. The self-stance keyword is used to represent self-stance evaluation, the other-party stance keyword is used to represent the stance evaluation of non-self, and the reference coefficient assignment of the self-stance keyword is higher than that of the other-party stance keyword.

[0015] As a still further solution of the present invention: It further includes a stance inheritance step:

[0016] Record the historical user preference grid of the speaker, establish a stance information library of the speaker, and each user preference grid in the stance information library is associated with the user preference grid of the dialogue object in the corresponding communication;

[0017] Match the stance information library based on the stance keyword and the corresponding evaluation object of the current voice communication data, and retrieve the matching degree of the corresponding stance keyword. Select a rated number of user preference grids corresponding to the stance keywords as the to-be-screened stances in the order of decreasing matching degree;

[0018] Obtain the stance keyword of the second person through the voice communication data of the current conversation, and compare and retrieve the associated user preference grid of the to-be-screened stance through this stance keyword, so as to screen out the current similar stance association record based on the matching degree;

[0019] Inherit the user preference grid of the similar stance association record to limit the range of the stance preference situation of the speaker in the current communication.

[0020] As a further aspect of the present invention: it further includes a determination step based on event effects:

[0021] Perform semantic association analysis on multiple consecutive voice communication data, and judge the main event corresponding to the clear stance and independent events. The independent events are used to represent the subsidiary sub-events based on the main event;

[0022] Obtain the expected result of the main event, and evaluate the independent event based on the expected result. If the independent event contradicts the expected result, it indicates that the current independent event is a negative event relative to the main event, and perform a negative mark on the independent event;

[0023] If the evaluation expression of the current stance keyword contradicts the meaning of the negative mark, the current user speech data contains reverse expression meaning.

[0024] As a further aspect of the present invention: it further includes a judgment step of emotional expression inertia:

[0025] Perform natural language emotional expression analysis based on the user's historical voice communication data to obtain the emotional expression inertia of the user in language use. The emotional expression inertia is used to represent the user's language emotion use preference in normal speech expressions;

[0026] Perform natural language emotional expression analysis on the current voice communication data. If the emotional characteristics do not conform to the emotional expression inertia, the current user speech data contains reverse expression meaning.

[0027] An embodiment of the present invention aims to provide an intelligent language emotion recognition system, including:

[0028] A speech understanding module, which is used to split and record voice communication data according to the speaking object, and perform natural language analysis on the voice communication data to obtain its basic semantic content. The voice communication data includes multiple voice segments, and each voice segment independently corresponds to the basic semantic content;

[0029] A stance splitting module, which is used to perform natural semantic understanding on the basic semantic content, generate several stance keywords, and perform stance statistics on the speaking object to obtain a user stance preference grid. The user stance preference grid is used to represent multiple stance keywords of the object and the corresponding statistical fitting rate;

[0030] A stance positioning module, which is used to obtain the current user speech data, identify the corresponding speaking object based on voice feature recognition, and perform stance extraction on the user speech data to obtain the stance expression to be evaluated;

[0031] An emotional evaluation module, which is used to compare the to-be-evaluated position expression based on the user position preference grid of the speaker object. If it does not match the user position preference grid, the current user speech data contains a reverse expression meaning, and the reverse expression meaning is used to represent an emotional position expression way in which the speech content is mutually exclusive with the user's actual expression.

[0032] As a further solution of the present invention: the position keywords include self-position keywords and other-party position keywords, and the emotional evaluation module includes:

[0033] An association evaluation unit, which is used to perform semantic association analysis on continuous multiple voice communication data based on a natural language understanding model, and then judge the evaluation object corresponding to the position keyword and establish an association relationship;

[0034] A belonging judgment unit, which is used to match the evaluation object with the speaker object of the current voice communication data. If they match, it is determined as a self-position keyword. If they do not match, it is determined as an other-party position keyword, and the evaluation object is set as the position evaluation object;

[0035] A differential assignment unit, which is used to assign a reference coefficient in statistics to the position keyword according to its belonging property, and perform statistics on the position keyword after the reference coefficient assignment based on the belonging object. The self-position keyword is used to represent self-position evaluation, and the other-party position keyword is used to represent the position evaluation of non-self. The reference coefficient assignment of the self-position keyword is higher than that of the other-party position keyword.

[0036] As a still further solution of the present invention: it further includes a position inheritance module:

[0037] A feature recording unit, which is used to record the historical user preference grid of the speaker object and establish a position information library of the speaker object. Each user preference grid in the position information library is associated with the user preference grid of the dialogue object in the corresponding communication;

[0038] A retrieval matching unit, which is used to perform matching on the position information library based on the position keyword and the corresponding evaluation object of the current voice communication data, and retrieve the matching degree of the corresponding position keyword, and select a rated number of user preference grids corresponding to the position keyword as the to-be-screened positions in the order of decreasing matching degree;

[0039] A cross-retrieval unit, which is used to obtain the position keyword of the second person through the voice communication data of the current dialogue, and perform comparison and retrieval on the associated user preference grid of the to-be-screened position through this position keyword, so as to screen out the current similar position association record based on the matching degree;

[0040] A cross-validation unit, configured to inherit user preference grids for the similar stance association records, so as to limit the scope of the stance preference situation of the speaker in the current communication.

[0041] As a further aspect of the present invention: it further includes an effect judgment module:

[0042] An event localization unit, configured to perform semantic association analysis based on a plurality of consecutive voice communication data, and judge the main event and independent events corresponding to the clear stance, where the independent events are used to represent the subsidiary sub-events based on the main event;

[0043] An expectation judgment unit, configured to obtain the expected result of the main event, and evaluate the independent event based on the expected result. If the independent event contradicts the expected result, it indicates that the current independent event is a negative event relative to the main event, and a negative mark is made for the independent event;

[0044] An emotion evaluation unit, configured to determine that the current user speech data contains reverse expression meaning if the evaluation expression of the current stance keyword contradicts the meaning of the negative mark.

[0045] As a further aspect of the present invention: it further includes an expression inertia evaluation module:

[0046] An expression feature analysis unit, configured to perform natural language emotion expression analysis based on the user's historical voice communication data, and obtain the emotion expression inertia of the user in language use, where the emotion expression inertia is used to represent the language emotion use preference of the user in normal speech expressions;

[0047] An expression emotion judgment unit, configured to perform natural language emotion expression analysis on the current voice communication data. If the emotion feature does not conform to the emotion expression inertia, it indicates that the current user speech data contains reverse expression meaning.

[0048] Compared with the prior art, the beneficial effects of the present invention are: in the understanding of the user's conversation content by the big data model, through the dialogue understanding based on natural language, the content of the conversation and the viewpoints and stances of each speaker in the conversation are evaluated and obtained. Furthermore, the speech content of the speaker can be matched and evaluated according to the viewpoints and stances, so as to judge whether the speech is a conventional expression that conforms to the stance or a reverse expression with emotional filling. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of an intelligent language emotion recognition method.

[0050] Figure 2 It is a flow chart of the stance inheritance step in an intelligent language emotion recognition method.

[0051] Figure 3 It is a block diagram of the composition of an intelligent language emotion recognition system. Specific implementation manners

[0052] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] The following describes in detail the specific implementation manners of the present invention with reference to specific embodiments.

[0054] As Figure 1 described, an intelligent language emotion recognition method provided by an embodiment of the present invention includes the following steps:

[0055] S10. Split and record the voice communication data according to the speaking object, and perform natural language analysis on the voice communication data to obtain its basic semantic content. The voice communication data includes multiple voice segments, and each voice segment independently corresponds to the basic semantic content;

[0056] S20. Perform natural semantic understanding on the basic semantic content to generate several stance keywords, and perform stance statistics on the speaking object to obtain a user stance preference grid, which is used to represent multiple stance keywords of the object and the corresponding statistical fitting rate;

[0057] S30. Obtain the current user's speech data, identify the corresponding speaking object based on voice feature recognition, and perform stance extraction on the user's speech data to obtain a stance expression to be evaluated;

[0058] S40. Compare the stance expression to be evaluated based on the user stance preference grid of the speaking object. If it does not match the user stance preference grid, the current user's speech data contains a reverse expression meaning, and the reverse expression meaning is used to represent an emotional stance expression manner in which the speech content is mutually exclusive with the user's actual expression.

[0059] In this embodiment, an intelligent language emotion recognition method is provided for the understanding of user conversation content by a big data model. Through natural language-based conversation understanding, the content of the conversation and the viewpoints and positions of each speaker in the conversation are evaluated. Furthermore, the speech content of the speaker can be matched and evaluated according to the viewpoint and position, so as to determine whether the speech is a conventional expression that conforms to the position or a reverse expression with emotional filling; in actual communication or debate, people express their own viewpoints through language and evaluate the viewpoints put forward by others. The current natural language recognition method trained with a large amount of data can effectively record, recognize and understand the meaning represented in the conversation content of people in these scenarios. However, in the process of people's language expression in a debate, there may be emotional expression methods. For example, object A does not agree with object B, so an emotional expression method is used. This expression method appears as an agreement with object B in the text content, but in fact it is a negation of object B, that is, an expression method similar to irony. The existing natural language recognition method cannot effectively and accurately judge such speech methods, which may lead to deviations and errors in the understanding of the actual meaning of the conversation; the solution proposed in this embodiment is: on the basis of natural semantic recognition, determine the positions of multiple speakers participating in the conversation, obtain multiple position keywords that can express the object's position, and then when the user expresses again, the position information expressed in his current sentence (i.e., the position expression to be evaluated) can be compared with the previously determined position (position preference grid), so as to judge the content that may have a reverse meaning expressed by the user and determine whether it is a reverse emotional expression method to assist the natural language recognition of the existing technology to achieve a more accurate conversation understanding effect.

[0060] As another preferred embodiment of the present invention, the position keywords include self-position keywords and other-party position keywords. The step of statistically analyzing the positions of the speakers to obtain the user position preference grid specifically includes:

[0061] Based on the natural language understanding model, perform semantic association analysis on a continuous plurality of voice communication data, and then judge the evaluation object corresponding to the position keyword and establish an association relationship;

[0062] Match the evaluation object with the speaker of the current voice communication data. If they match, it is determined as a self-position keyword; if they do not match, it is determined as an other-party position keyword, and the evaluation object is set as the position evaluation object;

[0063] Assign a reference coefficient value for the statistics of the position keywords according to their respective natures, and perform statistics on the position keywords after the reference coefficient assignment based on the object to which they belong. The self-position keywords are used to represent the self-position evaluation, the other-party position keywords are used to represent the position evaluation of non-self, and the reference coefficient value of the self-position keywords is higher than that of the other-party position keywords.

[0064] In this embodiment, during the actual conversation process, the position information contained in the user's language may not be the user's own position, but a retelling or confirmation of the other party's position. Therefore, the position keywords can also be divided into self-position keywords and other-party position keywords. It is possible to judge who the expression object corresponding to the position keyword is by understanding the relevant conversation content, so as to determine whether the current position keyword is a self-position keyword or an other-party position keyword. At the same time, because the other-party position keywords are not obtained from one's own expression, there may be deviations in understanding the positions of others. Therefore, in the statistics of the position keywords of the object, a smaller statistical reference ratio should be used to avoid the deviation of the statistical results of the object's position caused by the understanding deviation of others.

[0065] As Figure 2 shown, as another preferred embodiment of the present invention, it further includes a position inheritance step:

[0066] S51, record the historical user preference grid of the speaking object, and establish a position information library of the speaking object. Each user preference grid in the position information library is associated with the user preference grid of the conversation object in the corresponding communication;

[0067] S52, match the position information library based on the position keywords and the corresponding evaluation objects of the current voice communication data, and retrieve the matching degree of the corresponding position keywords. Select a specified number of user preference grids corresponding to the position keywords in the order of decreasing matching degree as the positions to be screened;

[0068] S53, obtain the position keywords of the second person through the voice communication data of the current conversation, and compare and retrieve the relevant associated user preference grids of the positions to be screened through the position keywords, so as to screen out the current similar position association records based on the matching degree;

[0069] S54, inherit the user preference grid of the similar position association record to be used to narrow down the range of the position preference of the speaking object in the current communication.

[0070] In this embodiment, the step of stance inheritance is supplemented. In some scenarios, there is a large amount of stance information at the beginning of a conversation. However, since the conversation has just started, the content available for stance understanding is limited, and it is impossible to effectively and quickly determine the stance of each object. In order to quickly achieve an accurate stance judgment of the user's communication content, the historical stance of the user can be recorded and matched, and an evaluation can be made based on the historical stance record at the beginning of the communication. It should be noted that the stance matching here includes two parts of content. Since stances are corresponding, the same stance keyword may represent different stances in different scenarios. Therefore, first, the stance keywords expressed by the user currently are screened and matched to obtain multiple user preference networks containing stance keywords, and then further comparison and screening are performed based on the stance keywords of other users in multiple scenarios associated with the current stance keyword (according to the stance of the other party in the current conversation), so as to determine the stance preference grid of the current user that should be inherited specifically.

[0071] As another preferred embodiment of the present invention, it further includes a determination step based on the event effect:

[0072] Perform semantic association analysis on a series of consecutive voice communication data to judge the main event corresponding to the clear stance and the independent event, and the independent event is used to represent the subsidiary sub-event based on the main event;

[0073] Obtain the expected result of the main event, and evaluate the independent event based on the expected result. If the independent event contradicts the expected result, it indicates that the current independent event is a negative event relative to the main event, and a negative mark is made on the independent event;

[0074] If the evaluation expression of the current stance keyword contradicts the meaning of the negative mark, the current user's speech data contains a reverse expression meaning.

[0075] In this embodiment, a supplementary scheme for the event effect determination step is supplemented. This scheme is applicable to the event evaluation in non-debate scenarios. For example, the user's evaluation of the result of an event. For this event, its occurrence and the executor's demand for the result can both be obtained and determined. Therefore, when the event has occurred and the result is fixed, the current result can be judged as positive or negative according to the executor's required result, so as to evaluate the user's speech and determine whether it is an expression with an emotional reverse expression.

[0076] As another preferred embodiment of the present invention, it further includes a judgment step of emotional expression inertia:

[0077] Perform natural language emotional expression analysis based on the user's historical voice communication data to obtain the emotional expression inertia of the user in language use, where the emotional expression inertia is used to characterize the user's language emotion use preference in normal speech expressions;

[0078] Perform natural language emotional expression analysis on the current voice communication data. If the emotional characteristics do not match the emotional expression inertia, the current user's speech data contains reverse expression meaning.

[0079] In this embodiment, the judgment step of emotional inertia is supplemented to record the language use characteristics of the speaker, and then statistically judge the emotional characteristics of the speaker's language expression in language use. For example, when user A conducts a conceptual debate expression related to stance, it usually uses a relatively stable language emotion and neutral vocabulary to express. However, after communicating with user B for a period of time, user A expresses approval of user B's stance, but the language emotion shows a fluctuating state, which indicates that the emotional expression inertia of user A has changed at this moment, and it may be an embodiment of the user's reverse expression. Because in a conversation, when a person persists in debating for their own stance, their emotional expression is coherent and will not change suddenly.

[0080] As Figure 3 shown, the present invention also provides an intelligent language emotion recognition system, which includes:

[0081] A speech understanding module 100, which is used to split and record voice communication data according to the speaker, and perform natural language analysis on the voice communication data to obtain its basic semantic content. The voice communication data includes multiple groups of voice segments, and each voice segment independently corresponds to the basic semantic content;

[0082] A stance splitting module 200, which is used to perform natural semantic understanding on the basic semantic content, generate several stance keywords, and perform stance statistics on the speaker to obtain a user stance preference grid, where the user stance preference grid is used to characterize multiple stance keywords of the object and the corresponding statistical fitting rate;

[0083] A stance positioning module 300, which is used to obtain the current user's speech data, identify and judge the corresponding speaker based on voice characteristics, and perform stance extraction on the user's speech data to obtain the stance expression to be evaluated;

[0084] An emotion evaluation module 400, which is used to compare the stance expression to be evaluated based on the user stance preference grid of the speaker. If it does not match the user stance preference grid, the current user's speech data contains reverse expression meaning, and the reverse expression meaning is used to characterize the emotional stance expression method in which the speech content is mutually exclusive with the user's actual expression.

[0085] As another preferred embodiment of the present invention, the stance keywords include self-stance keywords and other-party stance keywords, and the sentiment evaluation module 400 includes:

[0086] An association evaluation unit, configured to perform semantic association analysis on a plurality of consecutive voice communication data based on a natural language understanding model, and then judge an evaluation object corresponding to the stance keyword, and establish an association relationship;

[0087] An attribution judgment unit, configured to match the evaluation object with the speaker of the current voice communication data. If they match, it is determined as a self-stance keyword; if they do not match, it is determined as an other-party stance keyword, and the evaluation object is set as the stance evaluation object;

[0088] A differential assignment unit, configured to assign a reference coefficient in statistics according to the nature of the stance keyword to which it belongs, and perform statistics on the stance keyword after the reference coefficient assignment based on the belonging object. The self-stance keyword is used to represent self-stance evaluation, and the other-party stance keyword is used to represent the stance evaluation of non-self. The reference coefficient assignment of the self-stance keyword is higher than that of the other-party stance keyword.

[0089] As another preferred embodiment of the present invention, it further includes a stance inheritance module:

[0090] A feature recording unit, configured to record the historical user preference grid of the speaker, establish a stance information library of the speaker, and each user preference grid in the stance information library is associated with the user preference grid of the conversation object in the corresponding communication;

[0091] A retrieval matching unit, configured to perform matching on the stance information library based on the stance keyword and the corresponding evaluation object of the current voice communication data, and retrieve the matching degree of the corresponding stance keyword, and select a rated number of user preference grids corresponding to the stance keyword as the stance to be screened according to the decreasing order of the matching degree

[0092] A cross-retrieval unit, configured to obtain the stance keyword of the second person through the voice communication data of the current conversation, and perform comparison and retrieval on the associated user preference grid of the stance to be screened through this stance keyword, so as to screen out the current similar stance association record based on the matching degree;

[0093] A cross-validation unit, configured to perform user preference grid inheritance on the similar stance association record, so as to narrow down the range of the stance preference situation of the speaker in the current communication.

[0094] As another preferred embodiment of the present invention, it further includes an effect judgment module:

[0095] An event location unit, configured to perform semantic association analysis based on a plurality of consecutive voice communication data, and determine a main event corresponding to a clear stance and independent events, where the independent events are used to represent subordinate sub-events based on the main event;

[0096] An expectation judgment unit, configured to obtain an expected result of the main event, and evaluate the independent event based on the expected result. If the independent event contradicts the expected result, it indicates that the current independent event is a negative event relative to the main event, and a negative mark is made on the independent event;

[0097] An emotion evaluation unit, configured to determine that the current user speech data contains a reverse expression meaning if the evaluation expression of the current stance keyword contradicts the meaning of the negative mark.

[0098] As another preferred embodiment of the present invention, it further includes an expression inertia evaluation module:

[0099] An expression feature analysis unit, configured to perform natural language emotion expression analysis based on the user's historical voice communication data, and obtain the emotion expression inertia of the user in language use, where the emotion expression inertia is used to represent the user's language emotion use preference in normal speech expressions;

[0100] An expression emotion judgment unit, configured to perform natural language emotion expression analysis on the current voice communication data. If the emotion feature does not conform to the emotion expression inertia, it indicates that the current user speech data contains a reverse expression meaning.

[0101] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0102] After considering the specification and the disclosure of the embodiments, those skilled in the art will readily conceive of other embodiments of the present disclosure. The present application is intended to cover any variations, uses, or adaptations of the present disclosure, which follow the general principles of the present disclosure and include well-known common general knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims.

[0103] It should be understood that the present disclosure is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. An intelligent language emotion recognition method, characterized in that: Include: Splitting and recording voice communication data according to the speaking object, performing natural language analysis on the voice communication data to obtain its basic semantic content, the voice communication data including multiple groups of voice segments, and each of the voice segments independently corresponds to the basic semantic content; Performing natural semantic understanding on the basic semantic content, generating a number of stance keywords, and performing stance statistics on the speaking object to obtain a user stance preference grid, wherein the user stance preference grid is used to characterize a number of stance keywords of the object and the corresponding statistical fit rate; Acquire the current user's speech data, determine the corresponding speech object based on voice feature recognition, and extract the stance of the user's speech data to obtain the stance expression to be evaluated; The stance expression to be evaluated is compared based on the user stance preference grid of the speech object. If it does not match the user stance preference grid, the current user speech data contains reverse expression meaning, which is used to characterize the emotional stance expression mode that is inconsistent with the speech content and the actual expression of the user.

2. The intelligent language emotion recognition method according to claim 1, characterized in that: The stance keywords include self-stance keywords and other-stance keywords. The step of performing stance statistics on the speaking object to obtain the user stance preference grid specifically includes: Based on the natural language understanding model, a semantic correlation analysis is performed on a plurality of continuous voice communication data, thereby determining the evaluation object corresponding to the stance keyword and establishing a correlation relationship; Matching the evaluation object with the speaker of the current voice communication data, and if they match, determining them as self-position keywords; if they do not match, determining them as other-party position keywords, and setting the evaluation object as the position evaluation object; A reference coefficient is assigned to the stance keyword in statistics according to its property, and the stance keyword after the reference coefficient is assigned is counted based on the object to which it belongs. The self-stance keyword is used to characterize the self-stance evaluation, and the other-stance keyword is used to characterize the stance evaluation of someone other than the self. The reference coefficient assigned to the self-stance keyword is higher than that of the other-stance keyword.

3. The intelligent language emotion recognition method according to claim 2, characterized in that: It also includes the position inheritance step: Recording the user preference grids of the speaking object's history, and establishing a stance information database of the speaking object, wherein each user preference grid in the stance information database is associated with a user preference grid of the corresponding dialogue object in the communication; Match the stance information database based on the stance keywords and the corresponding evaluation objects of the current voice communication data, retrieve the matching degree of the corresponding stance keywords, and select the user preference grids corresponding to the rated number of stance keywords as the stances to be screened by reducing the sequence order of the matching degree; Obtain the second person's stance keyword through the voice communication data of the current conversation, and compare and retrieve the associated user preference grid of the stance to be screened through the stance keyword, so as to screen out the current similar stance related records based on the matching degree; The user preference grid inheritance is performed on the similar stance association records to narrow the stance preference of the speaker in the current communication.

4. The intelligent language emotion recognition method according to claim 3, characterized in that: It also includes the following decision steps based on event effects: Perform semantic association analysis based on multiple continuous voice communication data to determine the main event and independent event corresponding to the clear position, wherein the independent event is used to represent the subsidiary sub-event based on the main event; Obtaining the expected result of the main event, and evaluating the independent event based on the expected result; if the independent event contradicts the expected result, the current independent event is characterized as a negative event relative to the main event, and the independent event is negatively marked; If the evaluation expression of the current stance keyword contradicts the negative mark meaning, the current user speech data contains the reverse expression meaning.

5. The intelligent language emotion recognition method according to claim 4, characterized in that: It also includes the judgment steps of emotional expression inertia: Performing natural language emotion expression analysis based on the user's historical voice communication data to obtain the user's emotion expression inertia in language use, wherein the emotion expression inertia is used to characterize the user's language emotion use preference in normal speech expression; The current voice communication data is subjected to natural language emotion expression analysis. If the emotion feature does not match the emotion expression inertia, the current user speech data contains reverse expression meaning.

6. An intelligent language emotion recognition system, characterized in that: Include: A speech understanding module, used to split and record speech communication data according to the speaking object, and perform natural language analysis on the speech communication data to obtain its basic semantic content, wherein the speech communication data includes multiple groups of speech segments, and each of the speech segments independently corresponds to the basic semantic content; A stance splitting module is used to perform natural semantic understanding on the basic semantic content, generate a number of stance keywords, and perform stance statistics on the speaking object to obtain a user stance preference grid, wherein the user stance preference grid is used to characterize a plurality of stance keywords of the object and the corresponding statistical fit rate; A stance positioning module is used to obtain the current user's speech data, identify the corresponding speech object based on the sound feature recognition, and extract the stance of the user's speech data to obtain the stance expression to be evaluated; The emotional evaluation module is used to compare the stance expression to be evaluated based on the user stance preference grid of the speech object. If it does not match the user stance preference grid, the current user speech data contains reverse expression meaning, which is used to characterize the emotional stance expression mode that is inconsistent with the speech content and the actual expression of the user.

7. The intelligent language emotion recognition system according to claim 6, characterized in that: The stance keywords include self-stance keywords and other-stance keywords, and the emotion evaluation module includes: A correlation evaluation unit, used to perform semantic correlation analysis on a plurality of continuous voice communication data based on a natural language understanding model, thereby determining the evaluation object corresponding to the stance keyword and establishing a correlation relationship; a judgment unit, configured to match the evaluation object with the speaker of the current voice communication data, and if they match, determine them as self-position keywords; if they do not match, determine them as other-position keywords, and set the evaluation object as the position evaluation object; The differential assignment unit is used to assign a reference coefficient in statistics of the stance keywords according to their properties, and to perform statistics on the stance keywords after the reference coefficient assignment based on the objects to which they belong, wherein the self-stance keywords are used to characterize the self-stance evaluation, and the other-stance keywords are used to characterize the stance evaluation of a person other than the self, and the reference coefficient assignment of the self-stance keywords is higher than that of the other-stance keywords.

8. The intelligent language emotion recognition system according to claim 7, characterized in that: Also includes stance inheritance module: A feature recording unit, used to record the user preference grid of the speech object's history and establish a stance information library of the speech object, wherein each user preference grid in the stance information library is associated with a user preference grid of the corresponding dialogue object in the communication; A search and matching unit is used to match the stance information library based on the stance keywords of the current voice communication data and the corresponding evaluation objects, and to retrieve the matching degree of the corresponding stance keywords, and select the user preference grids corresponding to the rated number of stance keywords as the stances to be screened by reducing the sequence order according to the matching degree; A cross-retrieval unit, used to obtain the second person's stance keyword through the voice communication data of the current conversation, and compare and retrieve the associated user preference grid of the stance to be screened through the stance keyword, so as to screen out the current similar stance related records based on the matching degree; The cross-validation unit is used to perform user preference grid inheritance on the similar stance association records, so as to narrow the stance preference of the speaker in the current communication.

9. The intelligent language emotion recognition system according to claim 8, characterized in that: It also includes the effect judgment module: An event location unit, used to perform semantic association analysis based on a plurality of continuous voice communication data, and determine a main event and an independent event corresponding to a clear position, wherein the independent event is used to represent an auxiliary sub-event based on the main event; An expectation judgment unit, used to obtain the expected result of the main event, and evaluate the independent event based on the expected result. If the independent event contradicts the expected result, the current independent event is characterized as a negative event relative to the main event, and the independent event is negatively marked; The sentiment evaluation unit is used to determine that if the evaluation expression of the current stance keyword contradicts the negative tag meaning, the current user speech data contains the reverse expression meaning.

10. The intelligent language emotion recognition system according to claim 9, characterized in that: Also includes expression inertia evaluation module: An expression feature analysis unit, used to analyze the emotion expression of natural language based on the historical voice communication data of the user, and obtain the emotion expression inertia of the user in language use, wherein the emotion expression inertia is used to characterize the language emotion use preference of the user in the usual speech expression; The emotion expression judgment unit is used to perform natural language emotion expression analysis on the current voice communication data. If the emotion characteristics do not match the emotion expression inertia, the current user speech data contains reverse expression meaning.

Images