Semantic meaning recognition method and device, equipment, medium and program product

Abstract

The invention provides a semantic recognition method and device, equipment, a medium and a program product. Relates to the financial science and technology field. The method comprises the steps of obtaining a to-be-processed question of a user, and determining a first candidate semantic meaning corresponding to the to-be-processed question; when the semantic similarity probability of the first candidate semanteme meets a preset matching condition, determining the first candidate semanteme as a target semanteme of the to-be-processed problem; when the semantic similarity probability of the first candidate semanteme does not meet a preset matching condition, determining a second candidate semanteme corresponding to the to-be-processed problem from the candidate service semanteme with the highest semantic similarity probability in the candidate service semanteme; determining a first integration probability corresponding to the first candidate semantic meaning and a second integration probability corresponding to the second candidate semantic meaning; and based on the first integration probability and the second integration probability, determining a target semantic meaning corresponding to the to-be-processed problem. According to the method, the technical problem of low semantic recognition accuracy in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of text information extraction, and in particular to a language identification method, device, equipment, medium and program product. BACKGROUND

[0002] In the current service field, intelligent customer service systems have become an important tool for improving service efficiency and quality, and are widely used in e-commerce, finance, telecommunications and other industries, and undertake important functions such as answering user questions and processing business. As the core module of the intelligent customer service system, the core function of language identification is to accurately understand the true meaning of the user's input question, so as to match the corresponding answer from the massive question and answer knowledge base, and its performance is directly related to the accuracy and effectiveness of the system and user interaction.

[0003] The existing language identification scheme mainly relies on a semantic similarity algorithm to match the semantic meaning of the user's question with the pre-set question and answer knowledge base, and is supplemented by a rule template and a static threshold to complete language identification.

[0004] However, the existing language identification scheme has the problem of low accuracy of semantic meaning recognition; therefore, how to improve the recognition accuracy of language identification is a technical problem that needs to be solved at present. SUMMARY

[0005] The present application provides a language identification method, device, equipment, medium and program product to solve the technical problem of low accuracy of language identification in the prior art.

[0006] In a first aspect, the present application provides a language identification method, comprising:

[0007] obtaining a user's to-be-processed question, determining a first candidate semantic meaning corresponding to the to-be-processed question;

[0008] when the semantic similarity probability of the first candidate semantic meaning meets a pre-set matching condition, determining the first candidate semantic meaning as the target semantic meaning of the to-be-processed question;

[0009] when the semantic similarity probability of the first candidate semantic meaning does not meet the pre-set matching condition, determining the candidate semantic meaning with the highest semantic similarity probability among the candidate semantic meanings as a second candidate semantic meaning corresponding to the to-be-processed question;

[0010] determining a first integration probability corresponding to the first candidate semantic meaning and a second integration probability corresponding to the second candidate semantic meaning;

[0011] based on the first integration probability and the second integration probability, determining a target semantic meaning corresponding to the to-be-processed question;

[0012] generating response information corresponding to the to-be-processed question according to the target semantic meaning.

[0013] In an optional implementation, determining the first candidate semantic corresponding to the to-be-processed question comprises:

[0014] Determining at least one candidate business semantic corresponding to the to-be-processed question.

[0015] Determining the first candidate semantic corresponding to the to-be-processed question as the candidate business semantic with the highest business category probability in the candidate business semantics.

[0016] In an optional implementation, determining the at least one candidate business semantic corresponding to the to-be-processed question comprises:

[0017] Performing keyword extraction on the to-be-processed question to obtain at least one keyword.

[0018] Determining at least one business category corresponding to the to-be-processed question based on the keywords.

[0019] Determining the candidate business semantic corresponding to the to-be-processed question under each business category based on a business semantic database of each business category.

[0020] In an optional implementation, determining the candidate business semantic corresponding to the to-be-processed question under each business category based on a business semantic database of each business category comprises:

[0021] For any business category, obtaining a business semantic database corresponding to the business category; the business semantic database stores at least one business semantic corresponding to the business category.

[0022] For any business semantic, determining a semantic similarity probability corresponding to the business semantic based on the to-be-processed question.

[0023] Determining the business semantic with the highest semantic similarity probability as the candidate business semantic corresponding to the to-be-processed question under the business category.

[0024] In an optional implementation, after determining the first candidate semantic corresponding to the to-be-processed question, the method further comprises:

[0025] Obtaining a semantic similarity probability of other candidate business semantics.

[0026] If the semantic similarity probability of the first candidate semantic is greater than the semantic similarity probability of other candidate business semantics, determining that the semantic similarity probability of the first candidate semantic satisfies a preset matching condition.

[0027] If the semantic similarity probability of the first candidate semantic meaning is less than or equal to the semantic similarity probability of other candidate business semantic meanings, it is determined that the semantic similarity probability of the first candidate semantic meaning does not satisfy the preset matching condition.

[0028] In an optional implementation, the first candidate semantic meaning or the second candidate semantic meaning is determined as a target candidate semantic meaning, and the first integrated probability or the second integrated probability is a target integrated probability.

[0029] The first integrated probability corresponding to the first candidate semantic meaning and the second integrated probability corresponding to the second candidate semantic meaning are determined.

[0030] The business category probability and the semantic similarity probability corresponding to the target candidate semantic meaning are obtained.

[0031] The business category is weighted based on a business category compensation coefficient corresponding to the business category probability to obtain a business category weighted probability.

[0032] The target integrated probability is determined based on the business category weighted probability and the semantic similarity probability.

[0033] In a second aspect, the present application provides a semantic category device, comprising:

[0034] A first candidate semantic meaning determination module is configured to obtain a to-be-processed question of a user, and determine a first candidate semantic meaning corresponding to the to-be-processed question.

[0035] A first target semantic meaning determination module is configured to determine the first candidate semantic meaning as a target semantic meaning of the to-be-processed question when the semantic similarity probability of the first candidate semantic meaning satisfies a preset matching condition.

[0036] A second candidate semantic meaning determination module is configured to determine a second candidate semantic meaning corresponding to the to-be-processed question when the semantic similarity probability of the first candidate semantic meaning does not satisfy the preset matching condition, the second candidate semantic meaning being a candidate business semantic meaning with the highest semantic similarity probability among all candidate business semantic meanings.

[0037] A semantic probability determination module is configured to determine a first integrated probability corresponding to the first candidate semantic meaning and a second integrated probability corresponding to the second candidate semantic meaning.

[0038] A second target semantic meaning determination module is configured to determine a target semantic meaning corresponding to the to-be-processed question based on the first integrated probability and the second integrated probability.

[0039] In a third aspect, the present application provides an electronic device, comprising a processor and a memory connected in communication with the processor.

[0040] The memory stores computer execution instructions.

[0041] The processor executes the computer-executed instructions stored in the memory to implement the method according to the first aspect.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executed instructions, and the computer-executed instructions are executed by a processor to implement the method according to the first aspect.

[0043] In a fifth aspect, the present application provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the method according to the first aspect.

[0044] The semantic recognition technology provided by the present application determines a first candidate semantic, and determines whether it is the target semantic according to the semantic similarity probability. When the condition is not met, a second candidate semantic is introduced, and the final target semantic is determined by combining the semantic probabilities of the two. In this process, through the setting of double candidate semantics and the comprehensive judgment of probabilities, the user's real semantic can be more accurately captured, and the mis-matching of user problems can be significantly reduced. Since the over-reliance on manual maintenance of rule templates is reduced, not only the labor cost is reduced, but also the conflicts caused by the increase in the number of rule templates are avoided. At the same time, the way of dynamically determining the target semantic based on probability is more flexible than the static threshold judgment, and can be adjusted adaptively according to the actual situation, thereby effectively improving the accuracy and reliability of semantic recognition. BRIEF DESCRIPTION OF DRAWINGS

[0045] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0046] Figure 1 A structure diagram of an intelligent customer service system for supporting semantic recognition provided by the prior art is shown.

[0047] Figure 2 An application scenario diagram of the semantic recognition method provided by the present application is shown.

[0048] Figure 3 A flowchart of the semantic recognition method provided by the embodiment of the present application is shown.

[0049] Figure 4 A structure diagram of an intelligent customer service system for supporting semantic recognition provided by the present application is shown.

[0050] Figure 5 A structure diagram of a semantic recognition device provided by the embodiment of the present application is shown.

[0051] Figure 6 A block diagram of an electronic device provided by the present application is shown.

[0052] The specific embodiments of the application have been shown and described in the above drawings and text. These drawings and text are not meant to limit the scope of the inventive concept in any way but are merely meant to illustrate the inventive concept to one of ordinary skill in the art by reference to a particular embodiment. DETAILED DESCRIPTION

[0053] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The same reference numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments are not meant to represent all embodiments consistent with the application. Rather, they are merely examples of apparatus and methods consistent with some aspects of the application as detailed in the appended claims.

[0054] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards of relevant countries and regions, necessary security measures are taken, public order and good customs are not violated, and appropriate operation portals are provided for users to choose authorization or refusal.

[0055] And the present application involves big data analysis of user information (including but not limited to personal biological characteristics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and uses artificial intelligence technology for automatic decision-making, and makes technical solutions based on automatic decision-making results that have a significant impact on personal rights and interests, provides corresponding operation portals for users to choose to agree or refuse automatic decision-making results; if the user chooses to refuse, the expert decision-making process is entered.

[0056] It should be noted that the language recognition method, device, equipment, storage medium and product provided by the present application can be used in the field of financial technology, and can also be used in any field other than financial technology. The application field of the language recognition method, device, equipment, storage medium and product in the present application is not limited.

[0057] Intent Recognition is a core technology in the field of natural language processing, mainly used to accurately analyze the user's true purpose or demand from the user's natural language text input (such as questions, instructions, statements, etc.). Its core goal is to map the user's natural language expression to a predefined "semantic category". For example, in the intelligent customer service scenario, the user inputs "Why hasn't my order been shipped yet?", and the Intent Recognition technology will classify it as "Querying the order logistics status"; in the intelligent sound box scenario, the user says "Play a song by XXX", which will be identified as "Music playback request".

[0058] As in the background art, existing Intent Recognition solutions mainly rely on semantic similarity algorithms to match user questions with pre-defined semantic questions in the question and answer knowledge base, supplemented by rule templates and static thresholds to complete Intent Recognition.

[0059] However, the above-mentioned Intent Recognition solution has the problem of high mis-matching rate, i.e. low recognition accuracy.

[0060] For example, see Figure 1 In the intelligent customer service system supporting the existing Intent Recognition method, it specifically includes an input receiving module, an Intent Recognition module, a question and answer knowledge base, a system output module, and a user feedback module; wherein the Intent Recognition module further includes a rule template matching sub-module and a similarity model calculation sub-module.

[0061] It should be noted that the rule template, similarity module, and question and answer knowledge base are pre-constructed, trained modules.

[0062] Among them, the rule template: used for rule template matching of the Intent Recognition module. Generally supports regular expression matching, and the rule template is maintained by artificial.

[0063] Similarity model: used for similarity model calculation of the Intent Recognition module. Generally uses algorithms such as bert, fasttext to train semantic similarity calculation models, which can calculate the similarity between two texts, generating a calculation result in the range [0, 1], and the higher the score, the more similar the meaning expressed by the two texts.

[0064] Question and answer knowledge base: the operation personnel maintain the question and answer knowledge used to reply to user questions, including business consultation, business operation, and other types of knowledge.

[0065] On this basis, in the process of Intent Recognition:

[0066] The input receiving module is used to provide an interface for the user to interact with the intelligent customer service system, receive user input, and pass the input data to the Intent Recognition module.

[0067] Language recognition module: used to identify the semantic of user's question, find the most matched knowledge (standard question) in the Q&A knowledge base. This module usually uses rule template matching, semantic similarity model calculation method to realize language recognition. Each language recognition method sets a threshold, when the user question and the standard question in the Q&A knowledge base calculation result exceeds the threshold, the answer corresponding to the standard question is returned.

[0068] System output module: the answer information corresponding to the question in the Q&A knowledge base calculated by the language recognition module is returned to the user through the dialogue interface.

[0069] User feedback: after receiving the system reply, the user can evaluate and feedback satisfaction or dissatisfaction. The operator optimizes the Q&A knowledge base according to the user feedback.

[0070] Exemplarily, taking the user question "why does the transfer prompt account lock" as an example, the implementation process of the prior art solution is described.

[0071] Pre-maintained module part:

[0072] Rule template: empty.

[0073] Similarity model: a trained bert similarity calculation model.

[0074] Q&A knowledge base: the operator maintains the Q&A knowledge of account exception reason, transfer remittance, bank card unlocking, etc.

[0075] In the specific recognition process:

[0076] 1. User input receiving: the user inputs the request through the dialogue interface, and the system receives the user input content "why does the transfer prompt account lock".

[0077] 2. Language recognition module: assuming that only the similarity model calculation method is used for language recognition, the threshold is 0.7, and the rule template matching method is empty. The system performs similarity model calculation, and the calculation result of the user question "why does the transfer prompt account lock" and the standard question in the Q&A knowledge base is [{“question”:"account exception reason","score":"0.65"},{“question”:"transfer remittance","score":"0.75"},{“question”:"bank card unlocking","score":"0.74"}]. Take the standard question "transfer remittance" and "bank card unlocking" with score greater than or equal to the threshold 0.7, and finally the system returns the result with the highest score "transfer remittance" (score is 0.75).

[0078] 3、System output module: the system presents the corresponding answer "please go to the transfer and remittance page, input the recipient's name, account number..." to the user through the dialogue interface.

[0079] 4、User feedback: the user thinks that the system response does not solve his problem, and feedback "unsatisfied". The operation personnel analyzes this record, supplements the rule template "*why* account lock*" to the standard question "account abnormal reason" in the question and answer knowledge base (when "why" and "account lock" appear in the user's question, it can be matched). When the user inputs the question "why does the transfer prompt account lock" again, the system can get the semantic recognition result as "account abnormal reason" through rule template matching, and return the corresponding answer, so as to improve the reply accuracy of the intelligent customer service system.

[0080] In the intelligent customer service system, the semantic recognition module based on the semantic similarity model calculation has the following defects:

[0081] 1、High user question mis-matching rate: using the same semantic similarity model for the question and answer knowledge base, there are cases where the standard question with the highest similarity calculation score does not match the user question (such as mis-matching "why does the transfer prompt account lock" to "transfer and remittance"). At present, the mis-matching rate is reduced by supplementing rule templates, but it requires high manual maintenance cost, and when the number of question knowledge bases and rule templates increases, the rule templates may conflict.

[0082] 2、Poor flexibility of semantic recognition module judgment threshold: static threshold is usually used in use, which cannot be dynamically and automatically optimized according to user feedback.

[0083] In summary, the existing semantic recognition scheme has the problem of high semantic problem mis-matching rate, which not only requires high-cost manual maintenance of rule templates, but also rule templates are prone to conflict; at the same time, its judgment threshold flexibility is poor, and the static threshold cannot be dynamically optimized according to user feedback, further reducing the accuracy of semantic problem recognition.

[0084] Therefore, how to improve the recognition accuracy of semantic recognition is a technical problem to be solved at present.

[0085] The semantic recognition method provided in this application aims to solve the aforementioned technical problems of existing technologies. Specifically, it determines a first candidate semantic meaning and judges whether it is the target semantic meaning based on its semantic similarity probability; when the condition is not met, a second candidate semantic meaning is introduced, and the semantic probabilities of the two are combined to determine the final target semantic meaning. In this process, by setting dual candidate semantic meanings and comprehensively judging probabilities, it can more accurately capture the user's true meaning and significantly reduce the mismatch of user questions; since it reduces the excessive reliance on manually maintained rule templates, it not only reduces labor costs but also avoids conflicts that may be caused by an increase in the number of rule templates; at the same time, the probability-based dynamic determination of the target semantic meaning is more flexible than static threshold judgment and can be adaptively adjusted according to the actual situation, ultimately effectively improving the accuracy and reliability of semantic recognition.

[0086] The semantic recognition method provided in this application can be widely applied to intelligent customer service systems in the financial sector. It can assist the system in accurately handling semantic recognition needs in service scenarios such as self-service business consultation and business processing for customers. When facing internal company scenarios, it can also help the system efficiently complete semantic parsing work in functional scenarios such as employee information inquiry and business operation, thereby improving the service quality and processing efficiency of intelligent customer service systems in the financial sector.

[0087] Furthermore, this solution can be extended to other fields involving human-computer natural language interaction, such as intelligent customer service on e-commerce platforms (handling semantics such as product inquiries and order queries), self-service systems in the telecommunications industry (responding to semantic recognition of package applications and fault reporting), and intelligent Q&A platforms for government services (analyzing semantics such as policy inquiries and business applications from the public), providing accurate and efficient semantic recognition support for intelligent interactive systems in various scenarios.

[0088] For ease of understanding, the following is based on Figure 2 The application scenarios applicable to the embodiments of this application are described below. Figure 2 An application scenario diagram of the semantic recognition method provided in this application. See also... Figure 2 The semantic recognition technology provided in this application involves a data acquisition device 21 and a semantic recognition device 22. Specifically, the data acquisition device 21 can receive text content input by the user through a user interface, treat it as a problem to be processed, and forward it to the semantic recognition device 22 for semantic recognition processing. In some cases, the user input is speech content. In this case, the data acquisition device 21 can convert the speech content to text through a pre-set text conversion module, treat the converted text content as a problem to be processed, and transmit it to the semantic recognition device 22 for semantic recognition processing.

[0089] However, it should be understood that in the above case, the data acquisition device 21 can also transmit the voice content input by the user as the to-be-processed question to the language recognition device 22 for text conversion and language recognition processing. The application does not limit the text conversion processing.

[0090] In the language recognition device 22, a series of recognition processes are performed on the obtained to-be-recognized data to obtain at least one candidate semantic meaning, and the target semantic meaning corresponding to the candidate semantic meaning is obtained through analysis and processing of the candidate semantic meaning.

[0091] Subsequently, in the language recognition device 22, the response information corresponding to the target semantic meaning can also be searched from the preset question and answer knowledge base, and the response information is fed back to the data acquisition device 11, so that the user knows the response information corresponding to the input content.

[0092] Subsequently, in the language recognition device 22, the response information corresponding to the target semantic meaning can also be searched from the preset question and answer knowledge base, and the response information is fed back to the data acquisition device 11, so that the user knows the response information corresponding to the input content.

[0093] The technical solutions of the application and how the technical solutions of the application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the application will be described below with reference to the drawings.

[0094] Figure 3 A flowchart of a language recognition method provided in an embodiment of the application is shown. The method can be executed by a language recognition device, which can be a server or an electronic device. The electronic device is taken as an example for description below, and the method in the embodiment can be realized by software, hardware or a combination of software and hardware, as shown in the figure. The method comprises the following steps: Figure 3

[0095] S301, obtaining a to-be-processed question of a user, and determining a first candidate semantic meaning corresponding to the to-be-processed question.

[0096] In the application, the to-be-processed question can be understood as original information input by a user to an intelligent customer service system through voice, text, image and other interactive forms. These information contains the user's doubts, needs or instructions. For example, the user inputs "how to query the credit card bill" in a financial scenario.

[0097] ​The candidate semantic meaning can be understood as a semantic meaning option that can match the real demand of the user, which is obtained by the intelligent customer service system through a preliminary analysis based on a preset algorithm and model, etc. The semantic meaning identified above is called the first candidate semantic meaning in order to distinguish it from other candidate semantic meanings.

[0098] Specifically, the intelligent customer service system captures the user inputted problem to be processed in real time through a user interaction interface, and converts it into text content that can be analyzed and processed.

[0099] Subsequently, a pre-trained semantic understanding model is called to perform keyword extraction and semantic analysis, etc., and compared with the preset semantic meaning library in the pre-constructed question and answer knowledge base, and the semantic meaning with the highest matching degree is selected as the first candidate semantic meaning.

[0100] In an optional embodiment, the application can analyze the business category to which the problem to be processed belongs when identifying the first candidate semantic meaning, and then determine the first candidate semantic meaning with the highest matching degree based on the business category. In this way, the problem to be processed can be first limited within a specific business category, reducing the interference of cross-business fields, thereby improving the initial matching accuracy of the first candidate semantic meaning and the real semantic meaning of the user, and laying a more reliable foundation for the subsequent semantic identification process.

[0101] Based on this, the process of determining the first candidate semantic meaning corresponding to the problem to be processed can include: determining at least one candidate business semantic meaning corresponding to the problem to be processed; determining the candidate business semantic meaning with the highest business category probability in each candidate business semantic meaning as the first candidate semantic meaning corresponding to the problem to be processed.

[0102] It can be explained that the candidate business semantic meaning refers to a specific semantic type that is closely related to a specific business category and meets the business logic and scene demand. When determining the candidate semantic meaning, at least one business category to which the content to be analyzed belongs can be first identified, and the corresponding candidate business semantic meaning for each business category can be determined.

[0103] When the problem to be processed is classified by business category, the business category probability corresponding to each of the at least one matched business category can be obtained; based on this, the candidate business semantic meaning corresponding to the business category with the highest probability can be determined as the first candidate semantic meaning corresponding to the problem to be processed, in the case that the candidate business semantic meaning corresponding to each business category has been determined. In this way, the initial semantic meaning that is more suitable for the problem to be processed can be selected in combination with the probability weight of the business category, the interference of irrelevant business field semantic meanings is reduced, and the preliminary matching accuracy of the first candidate semantic meaning is improved.

[0104] Based on the above optional implementation, the process of determining the candidate business semantics can include: keyword extraction on the to-be-processed question to obtain at least one keyword; determining at least one business category corresponding to the to-be-processed question based on each keyword; and determining the candidate business semantics corresponding to the to-be-processed question under each business category based on the business semantics database of each business category.

[0105] In the implementation process, the to-be-processed question input by the user can be preprocessed by natural language processing technology, such as word segmentation and stop word removal, to extract keywords that can reflect the core meaning of the content. For example, in a financial scenario, if the user inputs “why is the account locked after the transfer prompt”, the keywords “transfer”, “account”, and “lock” can be extracted.

[0106] The keywords are compared and matched with a preset business category keyword library, and at least one business category to which the to-be-processed question can belong and the corresponding matching probability (i.e., business category probability) are determined according to the matching degree. Alternatively, a business category determination model trained using an algorithm such as FastText is called, and the extracted keywords are input into the business category determination model to obtain the business category and probability distribution (i.e., business category probability) output by the model. For example, after keyword extraction and semantic classification, the probability distribution is obtained as: account security category: 0.78, and transfer operation category: 0.15.

[0107] Subsequently, the information stored in the business semantics database corresponding to each business category can be analyzed to determine the candidate business semantics corresponding to the to-be-processed question under each business category, thereby providing accurate basis for subsequent determination of candidate semantics.

[0108] In an exemplary manner, the implementation of determining the candidate business semantics corresponding to each business category based on the business semantics database can include: for any business category, obtaining the business semantics database corresponding to the business category; the business semantics database stores at least one business semantics corresponding to the business category; for any business semantics, determining a semantic similarity probability corresponding to the business semantics based on the to-be-processed question; and determining the business semantics with the highest semantic similarity probability as the candidate business semantics corresponding to the to-be-processed question under the business category.

[0109] Specifically, any business category is taken as an example, the business semantics database corresponding to the business category is called; the database stores the business semantics under the business category. By analyzing and matching all business semantics in the database with the to-be-processed question, the business semantics with the highest matching degree in the business semantics database corresponding to the to-be-processed question is determined according to the matching result, and the business semantics is determined as the candidate business semantics.

[0110] In the above embodiments, by associating the service category with the exclusive service semantic database, the determination of the candidate service semantics is more logically clear and reliable, thereby improving the matching degree of the candidate service semantics and the user demand, and providing solid support for the subsequent semantic identification accuracy.

[0111] S302, when the semantic similarity probability of the first candidate semantic meets the preset matching condition, determining the first candidate semantic as the target semantic of the problem to be processed.

[0112] In this application, the semantic similarity probability can be explained as the similarity between the text of the candidate semantic and the text of the problem to be processed.

[0113] On this basis, the process of determining whether the semantic similarity probability of the first candidate semantic meets the preset matching condition can include: obtaining the semantic similarity probability of other candidate service semantics; if the semantic similarity probability of the first candidate semantic is greater than the semantic similarity probability of other candidate service semantics, it is determined that the semantic similarity probability of the first candidate semantic meets the preset matching condition; if the semantic similarity probability of the first candidate semantic is less than or equal to the semantic similarity probability of other candidate service semantics, it is determined that the semantic similarity probability of the first candidate semantic does not meet the preset matching condition.

[0114] Specifically, after obtaining the first candidate semantic and other candidate service semantics, the similarity (i.e. semantic similarity probability) between the text of the first candidate semantic and the text of the problem to be processed, and the similarity between the text of each candidate service semantic and the text of the problem to be processed are calculated by a text similarity algorithm (such as cosine similarity, edit distance, etc.).

[0115] Subsequently, the semantic similarity probability of the first candidate semantic is compared with the semantic similarity probability of all other candidate service semantics: if the semantic similarity probability of the first candidate semantic is the largest in all comparison objects, i.e. greater than the semantic similarity probability of any other candidate service semantic, it is determined that it meets the preset matching condition, and then the first candidate semantic can be determined as the target semantic corresponding to the problem to be processed; otherwise, if there is at least one other candidate service semantic whose semantic similarity probability is greater than or equal to that of the first candidate semantic, it is determined that it does not meet the preset matching condition. Based on this, it is indicated that the first candidate semantic may not be the target semantic corresponding to the problem to be processed, and further comprehensive judgment needs to be made in combination with other candidate service semantics to obtain the most accurate recognition result.

[0116] Through the above embodiments, the matching priority of the first candidate semantic intention in all candidate business semantics can be determined based on the transverse comparison of the semantic similarity probability, and it is ensured that the first candidate semantic intention is directly determined as the target semantic intention only when the text similarity of the first candidate semantic intention is significantly higher than other semantic intentions, thereby reducing the misjudgment caused by close or reversed similarity probability, and further improving the accuracy of semantic intention identification.

[0117] S303, when the semantic similarity probability of the first candidate semantic intention does not satisfy the preset matching condition, the candidate business semantic with the highest semantic similarity probability in each candidate business semantic is determined as the second candidate semantic intention corresponding to the problem to be processed.

[0118] Optionally, when it is determined based on the above embodiments that the first candidate semantic intention may not be the target semantic intention corresponding to the problem to be processed, and further comprehensive judgment needs to be made in combination with other candidate business semantics, the candidate business semantic with the highest semantic similarity probability can be selected from all candidate business semantics, and the second candidate semantic intention is determined, so as to expand the reference dimension of semantic intention identification by introducing a secondary candidate, and provide a more comprehensive basis for determining the final target semantic intention subsequently.

[0119] S304, determining a first integrated probability corresponding to the first candidate semantic intention and a second integrated probability corresponding to the second candidate semantic intention.

[0120] In the present application, the semantic probability can be understood as a probability value determined after the comprehensive business category probability and the semantic similarity probability. Based on the above embodiments, to further improve the accuracy of semantic intention identification, the probability data of the first candidate semantic intention and the second candidate semantic intention can be integrated and analyzed to obtain the final semantic intention identification result.

[0121] Optionally, since the integrated probabilities of the first candidate semantic intention and the second candidate semantic intention are calculated based on the same method, it can be determined that the first candidate semantic intention or the second candidate semantic intention is the target candidate semantic intention, the first integrated probability or the second integrated probability is the target integrated probability, and based on this, one optional implementation of determining the first integrated probability corresponding to the first candidate semantic intention and the second integrated probability corresponding to the second candidate semantic intention can include: obtaining the business category probability and the semantic similarity probability corresponding to the target candidate semantic intention; performing weighted processing on the business category based on the business category compensation coefficient corresponding to the business category probability to obtain a business category weighted probability; and determining the target integrated probability based on the business category weighted probability and the semantic similarity probability.

[0122] Specifically, after determining the target candidate semantic (i.e., the first candidate semantic or the second candidate semantic), the business category probability and the semantic similarity probability corresponding to the semantic are extracted. During this period, a preset business category compensation coefficient is also introduced; wherein, the coefficient is used to weight and correct the business category probability according to the actual scene adaptability of different business categories, and the calculation formula can be expressed as "business category weighted probability = business category probability x business category compensation coefficient".

[0123] Subsequently, the business category weighted probability and the semantic similarity probability are integrated by a preset fusion algorithm (such as weighted summation, product normalization, etc.) to obtain the integrated probability (i.e., the first integrated probability or the second integrated probability) corresponding to the target candidate semantic, so as to realize the comprehensive quantification of the business attribute and the text matching degree, and improve the reliability of the semantic probability.

[0124] S305, based on the first integrated probability and the second integrated probability, determining the target semantic corresponding to the problem to be processed.

[0125] In this application, in the case of calculating the first integrated probability corresponding to the first candidate semantic and the second integrated probability corresponding to the second candidate semantic based on the above-mentioned embodiments, the first integrated probability and the second integrated probability are analyzed to obtain the final target semantic.

[0126] Specifically, the first integrated probability of the first candidate semantic and the second integrated probability of the second candidate semantic are compared in value: if the first integrated probability is greater than the second integrated probability, the first candidate semantic is determined as the target semantic corresponding to the problem to be processed; if the second integrated probability is greater than the first integrated probability, the second candidate semantic is determined as the target semantic; if the two integrated probabilities are equal, a secondary determination can be further made in combination with the business scene priority (such as the candidate semantic under high-frequency business scene priority) or the preset rule (such as the candidate semantic with higher semantic similarity probability is selected by default), and finally a unique target semantic is determined.

[0127] Through this direct comparison based on integrated probability and supplementary rule combination mode, the dual dimensions of business classification and text matching can be comprehensively considered, and the selection of the target semantic is more in line with the real needs of the user.

[0128] S306, generating the response information corresponding to the problem to be processed according to the target semantic.

[0129] In this application, the intelligent customer service system finds the response information corresponding to the target semantic in the business semantic database, and feeds back the response information to the user through the interactive interface, so as to improve the service quality of the intelligent customer service system.

[0130] In the technical solution, the first candidate semantic is determined, and whether it is the target semantic is determined according to the semantic similarity probability; when the condition is not met, the second candidate semantic is introduced, and the final target semantic is determined by combining the semantic probabilities of the two. In this process, through the setting of double candidate semantics and the probability comprehensive judgment, the user's real semantic can be more accurately captured, and the mis-matching of the user's problem can be significantly reduced. Since the excessive dependence on manual maintenance of the rule template is reduced, not only the labor cost is reduced, but also the conflicts caused by the increase in the number of rule templates are avoided. At the same time, the target semantic is determined based on the probability, which is more flexible than the static threshold judgment, and can be adjusted adaptively according to the actual situation, thereby effectively improving the accuracy and reliability of semantic identification.

[0131] Next, an example embodiment is provided to specifically describe the provided technical solution.

[0132] For example, Figure 4 The intelligent customer service system provided by the application includes an offline part and an online part; wherein,

[0133] The offline part includes:

[0134] The keyword extraction model is used to extract the domain keywords contained in the user question. Generally, the keyword extraction model is trained using a preset word extraction algorithm. For example, the domain keywords extracted from the question “Why is the account locked when transferring money” are “transfer”, “account”, and “lock”.

[0135] The business classification model is used to identify the business category to which the user question belongs. For example, the user input question can be divided into multiple possible matching business categories, such as the account security category and the transfer operation category, by extracting the keywords.

[0136] Specifically, a training data set is constructed in advance, user questions and knowledge standard questions are collected, and a preset algorithm is used to train the business classification model. The input of the model is the extracted domain keywords, and the output is the business category to which the question belongs and the probability distribution. For example, after keyword extraction and semantic classification, the probability distribution of the question “Why is the account locked when transferring money” is: account security category: 0.78, transfer operation category: 0.15.

[0137] The similarity model is used for semantic similarity calculation of the semantic identification module.

[0138] Optionally, the neural network algorithm can be used to train the semantic similarity calculation model. The similarity between two texts can be calculated, and the calculation result in the range of [0, 1] is generated. The higher the score, the more similar the meaning expressed by the two texts.

[0139] Semantic database: operators maintain the Q&A knowledge for replying to user questions. Additional maintenance is needed for the business category to which each piece of knowledge belongs. The business classification model can be used to automatically identify the business category.

[0140] The online part includes:

[0141] 1. Input receiving module: provides an interface for user interaction with the intelligent customer service system, receives user input, and passes the input data to the semantic analysis module.

[0142] 2. Semantic analysis module: used to identify user semantics. Specifically, two layers of calculation are performed on user questions:

[0143] (1) At least one business category to which the user question belongs and its corresponding probability distribution are obtained through the keyword extraction module and the business classification model.

[0144] (2) Through the similarity calculation model, a standard question corresponding to each business category is obtained from the semantic database as a candidate set.

[0145] 3. Dynamic decision module: based on the calculation results of the previous step, the candidate set is screened and identified to obtain the final recognition result.

[0146] (1) Dynamic threshold matching:

[0147] (1.1) Determine the standard question with the highest business category probability value in the candidate set as the first candidate question; if the semantic similarity of the candidate question is also the highest, the standard question is the final matching result.

[0148] (1.2) If the similarity of the candidate question is not the highest, confidence compensation calculation is performed: adjusted score = original similarity + a x business probability value.

[0149] Where a is the business probability compensation coefficient, which is a parameter for linear adjustment of the original text similarity score based on the confidence of the business classification model output. The semantic probability provides semantic-level confidence compensation to alleviate the local deviation of the text similarity. It can be obtained by training historical misjudgment data.

[0150] Through calculation, the final matching standard question is obtained.

[0151] 4. System output module: returns the answer corresponding to the standard question to the user through the dialogue interaction interface.

[0152] 5. User feedback: after receiving the system reply, the user can evaluate and feedback satisfaction or dissatisfaction.

[0153] Figure 5 The structure of a semantic recognition device provided in an embodiment of the present application is shown in the following figure.Figure 5 The semantic awareness device 50 comprises a first candidate semantic determination module 501, a first target semantic determination module 502, a second candidate semantic determination module 503, a semantic probability determination module 504, a second target semantic determination module 505, and a response information generation module 506. The first candidate semantic determination module 501 is configured to obtain a to-be-processed question of a user, and determine a first candidate semantic corresponding to the to-be-processed question.

[0154] The first candidate semantic determination module 501 is configured to obtain a to-be-processed question of a user, and determine a first candidate semantic corresponding to the to-be-processed question.

[0155] The first target semantic determination module 502 is configured to determine the first candidate semantic as a target semantic of the to-be-processed question when a semantic similarity probability of the first candidate semantic satisfies a preset matching condition.

[0156] The second candidate semantic determination module 503 is configured to determine a second candidate semantic corresponding to the to-be-processed question when the semantic similarity probability of the first candidate semantic does not satisfy the preset matching condition.

[0157] The semantic probability determination module 504 is configured to determine a first integration probability corresponding to the first candidate semantic and a second integration probability corresponding to the second candidate semantic.

[0158] The second target semantic determination module 505 is configured to determine a target semantic corresponding to the to-be-processed question based on the first integration probability and the second integration probability.

[0159] The response information generation module 506 is configured to generate response information corresponding to the to-be-processed question according to the target semantic.

[0160] In an optional implementation, the first candidate semantic determination module 501 comprises:

[0161] A candidate business semantic determination sub-module is configured to determine at least one candidate business semantic corresponding to the to-be-processed question.

[0162] A first candidate semantic determination sub-module is configured to determine a first candidate semantic corresponding to the to-be-processed question from candidate business semantics with the highest business category probability.

[0163] In an optional implementation, the first candidate semantic determination sub-module comprises:

[0164] A keyword extraction unit is configured to perform keyword extraction on the to-be-processed question to obtain at least one keyword.

[0165] A business category determination unit is configured to determine at least one business category corresponding to the to-be-processed question based on the keywords.

[0166] The candidate semantic determining unit is configured to determine, based on the service semantic database of each service category, a candidate service semantic corresponding to the to-be-processed question under each service category.

[0167] In an optional implementation, the candidate semantic determining unit comprises:

[0168] The service semantic database determining subunit is configured to, for any service category, acquire a service semantic database corresponding to the service category, and the service semantic database stores at least one service semantic corresponding to the service category.

[0169] The semantic similarity probability determining subunit is configured to, for any service semantic, determine a semantic similarity probability corresponding to the service semantic based on the to-be-processed question.

[0170] The candidate semantic determining subunit is configured to determine, as a candidate service semantic corresponding to the to-be-processed question under the service category, the service semantic with the highest semantic similarity probability.

[0171] In an optional implementation, the apparatus further comprises:

[0172] The semantic similarity probability determining module is configured to acquire the semantic similarity probability of the other candidate service semantic.

[0173] The first matching module is configured to, if the semantic similarity probability of the first candidate semantic is greater than the semantic similarity probability of the other candidate service semantic, determine that the semantic similarity probability of the first candidate semantic satisfies a preset matching condition.

[0174] The second matching module is configured to, if the semantic similarity probability of the first candidate semantic is less than or equal to the semantic similarity probability of the other candidate service semantic, determine that the semantic similarity probability of the first candidate semantic does not satisfy the preset matching condition.

[0175] In an optional implementation, the first candidate semantic or the second candidate semantic is determined as a target candidate semantic, and the first integration probability or the second integration probability is determined as a target integration probability.

[0176] The semantic probability determining module 504 comprises:

[0177] The probability data acquiring sub-module is configured to acquire the service category probability and the semantic similarity probability corresponding to the target candidate semantic.

[0178] The weighted probability acquiring sub-module is configured to perform weighted processing on the service category based on a service category compensation coefficient corresponding to the service category probability, to obtain a service category weighted probability.

[0179] The target integration probability determining sub-module is configured to determine the target integration probability based on the service category weighted probability and the semantic similarity probability.

[0180] Figure 6 is a block diagram of an electronic device provided by the present application. The device can be a computer, a digital broadcast terminal, or the like. Referring to Figure 6 , the device 800 can include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0181] The processing component 802 usually controls overall operations of the device 800, such as operations associated with displaying, making phone calls, data communications, camera operations, and recording operations. The processing component 802 can include one or more processors 820 to execute instructions to complete all or part of steps of the above methods. In addition, the processing component 802 can include one or more modules to facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0182] The memory 804 is configured to store various types of data to support operations of the device 800. Examples of these data include instructions for any application or methods operating on the device 800, contact data, phonebook data, messages, pictures, videos, and so on. The memory 804 can be realized by any type of volatile or nonvolatile storage devices or a combination thereof, such as static random-access memory (SRAM), electrically-erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic or optical disk.

[0183] The power supply component 806 supplies electrical power for various components of the device 800. The power supply component 806 can include a power supply management system, one or more power sources, and other components associated with generating, managing, and distributing electrical power for the device 800.

[0184] The multimedia component 808 includes a display for the device 800 and a screen for providing an output interface between the device 800 and a user. In some embodiments, the screen can include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors for sensing touch, swiping and gestures on the touch panel. The touch sensor can not only sense a boundary of a touching or swiping action, but also detect duration and pressure related to the touching or swiping action. In some embodiments, the multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When the device 800 is in an operation mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera can receive external multimedia data. Each of the front-facing camera and the rear-facing camera can be a fixed optical lens system or have a focal length and optical zoom capability.

[0185] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC) that is configured to receive an external audio signal when the device 800 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 804 or transmitted via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting an audio signal.

[0186] The input / output interface 812 provides an interface between the processing component 802 and peripheral interface modules, which can be a keyboard, a click wheel, a button, and the like. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0187] The sensor component 814 includes one or more sensors for providing status assessments for various aspects of the device 800. For example, the sensor component 814 can detect an open / closed position of the device 800, relative positioning of components, such as a display and keypad of the device 800, a change in position of the device 800 or a component of the device 800, presence or absence of user contact with the device 800, orientation or acceleration / deceleration of the device 800, and temperature changes of the device 800. The sensor component 814 can include proximity sensor(s) configured to detect presence of a nearby object without any physical contact. The sensor component 814 can further include a light sensor, such as a complementary metal oxide semiconductor (CMOS) or charge coupled device (CCD) sensor, for use in imaging applications. In some embodiments, the sensor component 814 can further include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0188] The communication component 816 is configured to facilitate wired or wireless communication between the device 800 and another device. The device 800 can access a wireless network based on a communication standard, such as WiFi, 4G, or 5G, or a combination thereof. In an example embodiment, the communication component 816 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In an example embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) techniques, Infrared Data Association (IrDA) techniques, Ultra-Wide Band (UWB) techniques, Bluetooth (BT) techniques, and other techniques.

[0189] In an example embodiment, the device 800 can be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electronic elements for executing the above-described methods.

[0190] In an example embodiment, a non-transitory computer-readable storage medium including instructions, such as the memory 804 including instructions, is also provided, which can be executed by the processor 820 of the device 800 to complete the above-described methods. For example, the non-transitory computer-readable storage medium can be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0191] A non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of a server, enables the server to perform the above-described method of language identification.

[0192] Embodiments of the present application also provide a chip running instructions, which is used to execute the technical solutions of the language identification method in the above-described embodiments.

[0193] Embodiments of the present application also provide a computer-readable storage medium, which stores computer execution instructions, when the computer execution instructions are run on a computer, the computer executes the technical solutions of the language identification method in the above-described embodiments.

[0194] Embodiments of the present application also provide a computer program product, which includes a computer program stored in a computer-readable storage medium, at least one processor can read the computer program from the computer-readable storage medium, and the at least one processor executes the computer program to implement the technical solutions of the language identification method in the above-described embodiments.

[0195] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0196] It is to be understood that the application is not limited to the precise details of construction and the exact arrangements of the components described above and illustrated in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the application is limited only by the claims that follow.

[0197] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0198] It is to be understood that the application is not limited to the precise details of construction and the exact arrangements of the components described above and illustrated in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the application is limited only by the claims that follow.

Claims

1. A language identification method, characterized by, The method comprises: acquiring a to-be-processed question of a user, determining a first candidate semantic corresponding to the to-be-processed question; when a semantic similarity probability of the first candidate semantic satisfies a preset matching condition, determining the first candidate semantic as a target semantic of the to-be-processed question; when the semantic similarity probability of the first candidate semantic does not satisfy the preset matching condition, determining a candidate business semantic with a highest semantic similarity probability among each candidate business semantic as a second candidate semantic corresponding to the to-be-processed question; determining a first integration probability corresponding to the first candidate semantic and a second integration probability corresponding to the second candidate semantic; based on the first integration probability and the second integration probability, determining a target semantic corresponding to the to-be-processed question; generating response information corresponding to the to-be-processed question according to the target semantic.

2. The method of claim 1, wherein, The determination of the first candidate semantic corresponding to the to-be-processed question comprises: determining at least one candidate business semantic corresponding to the to-be-processed question; determining a first candidate semantic corresponding to the to-be-processed question as a candidate business semantic with a highest business category probability among each candidate business semantic.

3. The method of claim 2, wherein, The determination of the at least one candidate business semantic corresponding to the to-be-processed question comprises: performing keyword extraction on the to-be-processed question to obtain at least one keyword; based on each keyword, determining at least one business category corresponding to the to-be-processed question; based on a business semantic database of each business category, determining a candidate business semantic corresponding to the to-be-processed question under each business category.

4. The method of claim 3, wherein, The determination of the candidate business semantic corresponding to the to-be-processed question under each business category based on the business semantic database of each business category comprises: for any business category, acquiring a business semantic database corresponding to the business category; the business semantic database stores at least one business semantic corresponding to the business category; for any business semantic, based on the to-be-processed question, determining a semantic similarity probability corresponding to the business semantic; determining a candidate business semantic corresponding to the to-be-processed question under the business category as a business semantic with the highest semantic similarity probability.

5. The method of claim 1, wherein, After determining the first candidate semantic corresponding to the to-be-processed question, the method further comprises: acquiring a semantic similarity probability of other candidate business semantics; if the semantic similarity probability of the first candidate semantic is greater than the semantic similarity probability of other candidate business semantics, it is determined that the semantic similarity probability of the first candidate semantic satisfies the preset matching condition; if the semantic similarity probability of the first candidate semantic is less than or equal to the semantic similarity probability of other candidate business semantics, it is determined that the semantic similarity probability of the first candidate semantic does not satisfy the preset matching condition.

6. The method of claim 1, wherein, determining the first candidate semantic or the second candidate semantic as a target candidate semantic; the first integration probability or the second integration probability is a target integration probability; the determination of the first integration probability corresponding to the first candidate semantic and the second integration probability corresponding to the second candidate semantic comprises: acquiring a business category probability and a semantic similarity probability corresponding to the target candidate semantic; The service category is weighted based on a service category compensation coefficient corresponding to the service category probability, to obtain a service category weighted probability; The target integration probability is determined based on the service category weighted probability and the semantic similarity probability.

7. A speech recognition apparatus characterized by comprising: The apparatus includes: A first candidate semantic determination module configured to obtain a to-be-processed question of a user, and determine a first candidate semantic corresponding to the to-be-processed question; A first target semantic determination module configured to, when a semantic similarity probability of the first candidate semantic satisfies a preset matching condition, determine the first candidate semantic as a target semantic of the to-be-processed question; A second candidate semantic determination module configured to, when the semantic similarity probability of the first candidate semantic does not satisfy the preset matching condition, determine a candidate service semantic with a highest semantic similarity probability among candidate service semantics as a second candidate semantic corresponding to the to-be-processed question; A semantic probability determination module configured to determine a first integration probability corresponding to the first candidate semantic and a second integration probability corresponding to the second candidate semantic; A second target semantic determination module configured to determine a target semantic corresponding to the to-be-processed question based on the first integration probability and the second integration probability; A response information generation module configured to generate response information corresponding to the to-be-processed question according to the target semantic.

8. An electronic device, comprising: It includes: A processor and a memory connected in communication with the processor; The memory stores computer execution instructions; The processor, when executing the computer execution instructions, is configured to implement the semantic classification method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions, when executed by a processor, are configured to implement the semantic classification method according to any one of claims 1 to 6.

10. A computer program product, characterised in that, It includes a computer program, and the computer program, when executed by a processor, implements the semantic classification method according to any one of claims 1 to 6.

Images