Business data processing method, device and equipment based on user session intent recognition

Abstract

This invention discloses a business data processing method, apparatus, and device based on user session intent recognition. First, target user session text data of the target user is acquired. Then, intent recognition is performed on the target user using an intent recognition model or a preset intent tag regular expression to obtain the current target intent recognition result. Next, similarity matching is performed in an intent recognition tag classification library based on the current target intent recognition result to obtain the current tag matching result. Finally, if it is determined that the current tag matching result is not empty, the corresponding current target recommendation information is obtained and sent to the customer service receiving end. This invention enables more accurate intent recognition of user session text data using an intent recognition model or a preset intent tag regular expression to obtain the current target intent recognition result, and can obtain the corresponding current target recommendation information based on the current target intent recognition result to feed back to the customer service receiving end as auxiliary reference information to support user communication.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology in artificial intelligence, and in particular to a business data processing method, apparatus and equipment based on user session intent recognition. Background Technology

[0002] Currently, after using some of the products or services provided by a company, users may contact the company's customer service system or customer service representatives via telephone or online instant messaging if they encounter any questions. During these communications, users may clearly express their needs, such as how to use the product or how to obtain coupons. If the company's customer service system or representatives have not addressed the user's questions in a timely manner during previous communications, they may need to analyze the current communication to determine whether a follow-up call should be made to the user.

[0003] Currently, in analyzing this communication situation, the first approach involves using the legally and compliantly obtained recording data as a basis. This involves designing keyword matching rules based on the experience of human experts, statistically analyzing keyword hits, and determining the user's intent. However, this approach requires experienced human experts to design the rules, and the excessive human intervention results in high labor costs for intent recognition due to its time and effort.

[0004] The second approach involves transcribing legally obtained audio data into text and then using a shallow NLP model (NLP stands for Natural Language Processing) to perform intent recognition on the transcribed data. Before training the shallow NLP model, manually labeled data needs to be accumulated as a training set. If the intent recognition results output by the shallow NLP model are expanded, the model needs to be retrained, making model updates costly. Furthermore, if the transcribed data contains multi-turn long dialogues, shallow NLP models cannot achieve good semantic understanding and classification accuracy, making accurate intent recognition impossible. Summary of the Invention

[0005] This invention provides a business data processing method, apparatus, and device based on user session intent recognition, aiming to solve the problem that when using shallow NLP models to perform intent recognition on user session data in the prior art, it is impossible to achieve accurate intent recognition, and it is also impossible to quickly obtain the recommendation information required by the customer service receiver based on the accurate intent recognition results.

[0006] In a first aspect, embodiments of the present invention provide a business data processing method based on user session intent recognition, applied to a server, comprising: In response to a session intent recognition command, target user session text data corresponding to the target user of the session intent recognition command is obtained; wherein, the target user session text data is obtained by speech recognition of user session voice data obtained through user interaction with the server and with user authorization. The intent recognition result of the current target intent is obtained by performing intent recognition on the target user's conversation text data through a pre-trained intent recognition model or a preset intent label regular expression. Based on the current target intent recognition result, a similarity match is performed in the locally stored intent recognition tag classification library, and a current tag matching result that meets the preset similarity condition with the current target intent recognition result is obtained from the intent recognition tag classification library; If it is determined that the current tag matching result is not empty, then the current target recommendation information corresponding to the current tag matching result is obtained, and the current target recommendation information is sent to the customer service receiving end; wherein, the current target recommendation information is the recommended communication text data or the reply text of the problem to be solved corresponding to the target user.

[0007] Secondly, embodiments of the present invention also provide a business data processing apparatus based on user session intent recognition, configured on a server, comprising: The conversation text data acquisition unit is used to acquire target user conversation text data corresponding to the target user in response to a conversation intent recognition instruction; wherein, the target user conversation text data is obtained by speech recognition of user conversation voice data obtained by the user terminal interacting with the server and with user authorization. The intent recognition unit is used to perform intent recognition on the target user's conversation text data through a pre-trained intent recognition model or a preset intent label regular expression to obtain the current target intent recognition result. The tag matching result acquisition unit is used to perform similarity matching in the locally stored intent recognition tag classification library based on the current target intent recognition result, and to obtain the current tag matching result from the intent recognition tag classification library that meets the preset similarity condition with the current target intent recognition result; The recommendation information acquisition unit is used to acquire the current target recommendation information corresponding to the current tag matching result if it is determined that the current tag matching result is not empty, and send the current target recommendation information to the customer service receiving end; wherein, the current target recommendation information is the recommended communication text data or the reply text of the problem to be solved corresponding to the target user.

[0008] Thirdly, embodiments of the present invention also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect above.

[0009] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the method described in the first aspect above.

[0010] This invention provides a business data processing method, apparatus, and device based on user session intent recognition. The method includes: responding to a session intent recognition instruction, acquiring target user session text data corresponding to the target user of the session intent recognition instruction; wherein, the target user session text data is obtained by speech recognition of user session voice data obtained through user interaction with the server and with user authorization; performing intent recognition on the target user session text data using a pre-trained intent recognition model or a preset intent tag regular expression to obtain a current target intent recognition result; performing similarity matching in a locally stored intent recognition tag classification library based on the current target intent recognition result, and obtaining a current tag matching result from the intent recognition tag classification library whose similarity to the current target intent recognition result meets a preset similarity condition; if it is determined that the current tag matching result is not empty, acquiring current target recommendation information corresponding to the current tag matching result, and sending the current target recommendation information to the customer service receiving end; wherein, the current target recommendation information is recommended communication text data or unresolved problem reply text corresponding to the target user. The embodiments of the present invention can use an intent recognition model or a preset intent tag regular expression to perform more accurate intent recognition on user conversation text data to obtain the current target intent recognition result, and can obtain the corresponding current target recommendation information based on the current target intent recognition result to feed back to the customer service receiving end, as auxiliary reference information to support user communication. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram illustrating an application scenario of the business data processing method based on user session intent recognition provided in an embodiment of the present invention. Figure 2A flowchart illustrating the business data processing method based on user session intent recognition provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the first sub-process of the business data processing method based on user session intent recognition provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the second sub-process of the business data processing method based on user session intent recognition provided in an embodiment of the present invention; Figure 5 A schematic diagram of the third sub-process of the business data processing method based on user session intent recognition provided in an embodiment of the present invention; Figure 6 A schematic block diagram of a service data processing device based on user session intent recognition provided in an embodiment of the present invention; Figure 7 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0015] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0017] Please also refer to Figure 1 and Figure 2 ,in Figure 1 This is a schematic diagram illustrating a scenario of the business data processing method based on user session intent recognition according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the business data processing method based on user session intent recognition provided in an embodiment of the present invention. Figure 1 As shown, the business data processing method based on user session intent recognition provided in this embodiment of the invention is applied to server 10. Both user terminal 20 and customer service receiving terminal 30 are communicatively connected to server 10. User terminal 20 is a smart terminal used by the user (such as a smartphone, tablet, etc.), and customer service receiving terminal 30 is a smart terminal used by customer service personnel (such as a smartphone, tablet, etc.). Figure 2 As shown, the method includes the following steps S110-S140.

[0018] S110. In response to the session intent recognition instruction, acquire the target user session text data corresponding to the target user of the session intent recognition instruction.

[0019] The target user conversation text data is obtained by speech recognition of user conversation voice data obtained through user interaction with the server and with user authorization.

[0020] In this embodiment, the technical solution is described with the server as the execution entity. After each user communicates with customer service via telephone or online instant messaging, the user's authorized voice conversation data is processed through a semantic recognition model to obtain user conversation text data. The following describes the complete technical solution of this application in detail using the specific process of processing the target user's conversation text data obtained after a customer completes a telephone conversation with customer service as an example. After the server obtains the target user's conversation text data (considered the target user), it uses it as data to be processed for subsequent intent recognition.

[0021] S120. Perform intent recognition on the target user's conversation text data using a pre-trained intent recognition model or a preset intent label regular expression to obtain the current target intent recognition result.

[0022] In this embodiment, to more quickly and accurately identify the intent of the target user's conversation text data, a multi-modal identification method can be adopted. For example, a preset intent tag regular expression can be used for short text data, while an intent recognition model can be used for long text data. Through the above multi-modal identification method, the most suitable identification method for the current scenario can be adapted to quickly obtain the current target intent recognition result.

[0023] In one embodiment, such as Figure 3 As shown, step S120 includes: S121. Obtain the conversation role tags corresponding to each conversation statement in the target user conversation text data and add them to the beginning of the corresponding conversation statement. Then, merge the conversation statements with the same conversation role tags and continuous conversation statement relationships and add the corresponding conversation role tags to the beginning of the statement to obtain preprocessed user conversation text data. S122. If it is determined that the conversation role tags included in the preprocessed user conversation text data belong to the first preset role tag set, then the current target intent recognition result corresponding to the preprocessed user conversation text data is obtained through the preset intent tag regular expression. S123. If it is determined that the conversation role tags included in the preprocessed user conversation text data belong to the second preset role tag set, then the intent recognition model is used to perform intent recognition on the target user conversation text data to obtain the current target intent recognition result.

[0024] The second preset character tag set includes a greater number of character tags than the first preset character tag set.

[0025] In this embodiment, to facilitate more accurate intent recognition of the target user's conversation text data, it can be preprocessed. Specifically, the conversation role tags corresponding to each conversation statement in the target user's conversation text data are obtained and added to the beginning of the corresponding conversation statement. Conversation statements with the same conversation role tags and consecutive conversation statement relationships are merged, and the corresponding conversation role tags are added to the beginning of the statement to obtain preprocessed user conversation text data. For example, the target user conversation text data after the addition of conversation role tags is as follows: "User A: Statement 1; User A: Statement 2; Customer Service: Statement 3; User A: Statement 4; Customer Service: Statement 5; User A: Statement 6". At this time, Statement 1 and Statement 2, which both correspond to User A, can be merged, and the corresponding conversation role tags can be added to the beginning of the statement to obtain preprocessed user conversation text data as "User A: Statement 1 + Statement 2; Customer Service: Statement 3; User A: Statement 4; Customer Service: Statement 5; User A: Statement 6".

[0026] In specific implementation, if the first preset role tag set is set to include customer service role tags, and the second preset role tag set is set to include both user role tags and customer service role tags, the preprocessed user conversation text data is analyzed to determine whether the conversation role tags belong to the first preset role tag set. If it is determined that the conversation role tags belong to the first preset role tag set, it indicates that there is only customer service dialogue. In this case, the current target intent recognition result corresponding to the preprocessed user conversation text data can be quickly obtained using a preset intent tag regular expression. If it is determined that the conversation role tags belong to the second preset role tag set, it indicates that there is normal dialogue between the user and customer service. In this case, the intent recognition model is used to perform intent recognition on the target user conversation text data to obtain the current target intent recognition result. Through the above multimodal recognition method, not all scenarios use the intent recognition model, which can effectively reduce the inference resource consumption of the intent recognition model, thereby saving the acquisition cost of intent recognition.

[0027] In one embodiment, the intent recognition model is a large language model with multiple hybrid reasoning modes, and includes at least a non-thinking reasoning mode and a thinking reasoning mode, such as... Figure 4 As shown, step S123 includes: S1231. If it is determined that the current session round number corresponding to the target user's session text data is less than or equal to the first preset round number, then the intent recognition model is used to perform intent recognition in the non-thinking reasoning mode to obtain the current target intent recognition result. S1232. If it is determined that the current session round number corresponding to the target user's session text data is greater than the first preset round number, then the intent recognition model is used to perform intent recognition in the thinking and reasoning mode to obtain the current target intent recognition result.

[0028] In this embodiment, even when using an intent recognition model to identify the intent of the target user's conversational text data, it is necessary to distinguish between the non-thinking reasoning mode and the thinking reasoning mode of this large language model. This is mainly determined by comparing the current conversation round number corresponding to the target user's conversational text data with a first preset round number (e.g., setting the first preset round number to 3 rounds; however, the specific implementation is not limited to the above example value and can be flexibly adjusted to other values ​​according to the user's actual needs). Specifically, when determining the current conversation round number corresponding to the target user's conversational text data, the cumulative occurrence count of each user role tag in the corresponding preprocessed user conversational text data can be counted, and the user role tag with the maximum cumulative occurrence count can be selected and its maximum cumulative occurrence count can be used as the current conversation round number.

[0029] If the current conversation round number corresponding to the target user's conversation text data is less than or equal to a first preset round number, it indicates that the current conversation corresponds to a short text dialogue scenario. The intent recognition model then performs intent recognition in the non-thinking mode (i.e., no-thinking mode) to obtain the current target intent recognition result. If the current conversation round number corresponding to the target user's conversation text data is greater than the first preset round number, it indicates that the current conversation corresponds to a long text dialogue scenario. The intent recognition model then performs intent recognition in the thinking mode (i.e., thinking mode) to obtain the current target intent recognition result. Therefore, using the non-thinking mode of the intent recognition model in short text dialogue scenarios and the thinking mode in long text dialogue scenarios can balance the processing speed and recognition accuracy of the intent recognition model.

[0030] S130. Based on the current target intent recognition result, perform similarity matching in the locally stored intent recognition tag classification library, and obtain the current tag matching result that meets the preset similarity condition with the current target intent recognition result from the intent recognition tag classification library.

[0031] In this embodiment, a similarity match is performed in a locally stored intent recognition tag classification library based on the current target intent recognition result. The first scenario is that a current tag matching result whose similarity to the current target intent recognition result meets a preset similarity condition can be obtained from the intent recognition tag classification library. The second scenario is that a current tag matching result whose similarity to the current target intent recognition result meets the preset similarity condition cannot be obtained from the intent recognition tag classification library. In the first scenario, it means that using the current target intent recognition result as a search condition can obtain a very similar or completely identical current tag matching result from the intent recognition tag classification library, allowing for subsequent recommendation information retrieval processing.

[0032] In the second scenario described above, if the current target intent recognition result cannot yield a very similar or identical current tag matching result in the intent recognition tag classification library, a pre-defined fallback strategy can be employed. This can be achieved through a double fallback approach using a large-scale model verification and transformation process based on the intent recognition model, along with keyword mapping. This ensures that the current tag matching result can ultimately obtain a current tag matching result from the intent recognition tag classification library that meets a pre-defined similarity condition to the current target intent recognition result. Specifically, the large-scale model verification and transformation process involves re-performing intent recognition on the target user's conversation text data using the intent recognition model or a pre-defined intent tag regular expression to obtain the current target intent recognition result. If this current target intent recognition result is not entirely identical to the previous result but can still obtain a current tag matching result from the intent recognition tag classification library that meets a pre-defined similarity condition, then the large-scale model verification and transformation process has been successfully implemented. Alternatively, the current target intent recognition result obtained after large model verification and transformation is based on a preset keyword mapping relationship that includes multiple intent recognition result keywords and convertible intent recognition results. The current target intent recognition result is then transformed into another approximate convertible intent recognition result by looking up the mapping relationship, so as to ensure that it can obtain a current tag matching result from the intent recognition tag classification library that meets the preset similarity condition with the current target intent recognition result.

[0033] The locally stored intent recognition label classification library can be obtained as follows: Acquire a dataset of labeled user conversation text (including multiple user conversation text datasets, each corresponding to a labeled intent recognition label), and aggregate the labeled intent recognition labels corresponding to the dataset to obtain an initial set of labeled intent recognition labels. Then, acquire the word vector corresponding to each initial labeled intent recognition label in the initial set, and perform K-means clustering to obtain multiple clusters. Select the word vector closest to the cluster center in each cluster as candidate word vectors, and form the intent recognition label classification library from the labeled intent recognition labels corresponding to each candidate word vector. This method allows for rapid intelligent clustering analysis based on a large dataset of labeled user conversation text, thus quickly obtaining the intent recognition label classification library.

[0034] S140. If it is determined that the current tag matching result is not empty, then obtain the current target recommendation information corresponding to the current tag matching result, and send the current target recommendation information to the customer service receiving end.

[0035] The current target recommendation information is the recommended communication text data or the reply text of the problem to be solved corresponding to the target user.

[0036] In this embodiment, if it is determined that the current tag matching result is not empty, it means that the current tag matching result corresponds to a specific intent recognition tag. If recommendation information is set in advance for each intent recognition tag in the intent recognition tag classification library in the local knowledge base of the server, after determining the intent recognition tag of the current tag matching result, the recommendation information with the mapping relationship of the intent recognition tag (such as recommended communication text data or reply text of unresolved issues) can be obtained from the knowledge base as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end for the customer service personnel corresponding to the customer service receiving end to use as reference information, and to refer to use in the follow-up communication with the target user.

[0037] In one embodiment, such as Figure 5 As shown, step S140 includes: S141. If it is determined that the current tag matching result belongs to the first preset intent tag set, then the recommended communication text data corresponding to the target user is obtained from the local knowledge base according to the current tag matching result as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end. S142. If it is determined that the current tag matching result belongs to the second preset intent tag set, then obtain the question data to be answered in the target user's conversation text data, and obtain the question reply text to be answered from the local knowledge base as the current target recommendation information, and send the current target recommendation information to the customer service receiving end; S143. If it is determined that the current tag matching result belongs to the third preset intent tag set, then the target user session text data is saved to the local preset storage area.

[0038] There is no overlap between the first preset intent tag set, the second preset intent tag set, and the third preset intent tag set.

[0039] In this embodiment, when specifically setting the first preset intent tag set, the second preset intent tag set, and the third preset intent tag set, the first preset intent tag set is for scenarios where users have a high intention to communicate, and includes intent tags such as consultation details, seeking advice, proactively asking questions, expressing needs, providing feedback, making appointments / handling matters, and proactively engaging in in-depth consultation; the second preset intent tag set is for scenarios where users have a general or moderate intention to communicate, and includes intent tags such as information query, function usage, simple consultation, basic operation, confirming information, routine consultation, and light interaction; the third preset intent tag set is for scenarios where users have a low intention to communicate, and includes intent tags such as single query, extremely simple instructions, passive response, no further needs, ending the conversation, only confirmation, no intention / silence, etc.

[0040] If the current tag matching result is determined to belong to the first preset intent tag set, it indicates that the user has a high communication intention. At this time, recommended communication text data corresponding to the target user can be obtained from the local knowledge base based on the current tag matching result as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end. The customer service personnel corresponding to the receiving end can use the specific dialogue content or recommended dialogue content included in the current target recommendation information as reference data to conduct corresponding follow-up communication with the target user.

[0041] If the current tag matching result is determined to belong to the second preset intent tag set, it indicates that the user has a general or moderate communication intention. Taking the need to promptly answer a customer's unresolved question from the previous conversation as an example, the system first retrieves the unanswered question data from the target user's conversation text data, and then obtains the unanswered question reply text from the local knowledge base as the current target recommendation information. This current target recommendation information is then sent to the customer service receiving end. The customer service personnel at the receiving end can use the unanswered question reply text included in the current target recommendation information as reference data to conduct appropriate follow-up communication with the target user.

[0042] If the current tag matching result is determined to belong to the third preset intent tag set, it indicates that the user has a low communication intention. In this case, the target user's conversation text data is saved to a local preset storage area for use as historical analysis data to optimize communication methods and content when customer service personnel communicate with users. It is evident that by more accurately determining the target preset intent tag set to which the current tag matching result belongs, the system can more accurately filter out the current target recommendation information from the knowledge base that can be used by the customer service receiver in subsequent follow-up communications.

[0043] In one embodiment, the method further includes the following step before step S140: Obtain user behavior trajectory data corresponding to the target user's conversation text data, and user operation behavior features corresponding to the user behavior trajectory data; The user operation behavior features are input into a pre-trained classification model to obtain the current classification result; If it is determined that the similarity between the current classification result and the current label matching result exceeds a preset similarity threshold, then the confidence level of the current label matching result is set to a preset confidence level value. If it is determined that the similarity between the current classification result and the current label matching result does not exceed a preset similarity threshold, then the current label matching result is updated with the current classification result.

[0044] In this embodiment, if a user performs an operation on the user interface of the user interaction system (such as an application) corresponding to the server but fails to obtain the required information or data before communicating online with customer service, in addition to obtaining the target user's conversation text data, the user's user behavior trajectory data for the user interaction system can also be obtained. For example, if the user behavior trajectory data specifically corresponds to page element A being clicked 3 times and viewed for 200 seconds, page element B being clicked once and viewed for 10 seconds, and page element C being clicked once and viewed for 100 seconds, then the user operation behavior features that can be extracted accordingly include three sub-features: [itemA, 3, 200], [itemB, 1, 10], and [itemC, 1, 100]. If the ID corresponding to the page element is used as the index, and the behavior feature columns include click features and browsing features, then after normalizing the click feature values ​​and browsing feature values ​​in the above three sub-features, a 3-row, 2-column user operation behavior feature can be obtained. By inputting user operation behavior features into classification models such as gradient boosting trees and random forests, the current classification result and its corresponding confidence level corresponding to the target user's conversation text data can be obtained.

[0045] If the similarity between the current classification result and the current tag matching result exceeds a preset similarity threshold (e.g., setting the preset similarity threshold to 90%, though this is not limited to the example value and can be adjusted to other values ​​according to user needs), it indicates a high degree of consistency between the intent recognition result obtained from user session data and the intent classification result obtained from user behavior trajectory data. In this case, the confidence level of the current tag matching result can be set to a preset confidence value (preferably 100%). If the similarity between the current classification result and the current tag matching result does not exceed the preset similarity threshold, it indicates a lack of consistency and contradiction between the intent recognition result obtained from user session data and the intent classification result obtained from user behavior trajectory data. In this case, the current tag matching result should be updated with the current classification result. Therefore, by comprehensively analyzing the consistency between the intent recognition results obtained from user session data and user behavior trajectory data, the user's accurate communication intent can be accurately obtained.

[0046] In one embodiment, the method further includes the following after step S140: Obtain the historical cumulative count of each intent recognition tag in the intent recognition tag classification library; Based on the current tag matching result, increment the historical cumulative count of the corresponding target intent recognition tag in the intent recognition tag classification library by one to update the historical cumulative count of the target intent recognition tag.

[0047] In this embodiment, the server can also perform historical cumulative statistics on the number of times each intent recognition tag in the intent recognition tag classification library has been recorded within a set statistical time period (such as the past week, past half month, past month, past quarter, past half year, past year, etc.). For example, when a user's user session text data is processed by intent recognition to determine the current target intent recognition result, and this current target intent recognition result successfully matches one of the intent recognition tags in the intent recognition tag classification library, it is considered that the intent recognition tag has been successfully hit once, and its historical cumulative statistics need to be incremented by one to update the historical cumulative statistics of the target intent recognition tag. After the above processing, the corresponding operation and maintenance personnel of the server can directly visualize the historical cumulative statistics of each intent recognition tag in the current intent recognition tag classification library in the form of bar charts, pie charts, etc., when requesting to view the historical cumulative statistics of each intent recognition tag in the intent recognition tag classification library. It can also display the percentage of the historical cumulative statistics of each intent recognition tag in the intent recognition tag classification library relative to the total historical cumulative statistics of all intent recognition tags, thereby realizing the data dashboard function.

[0048] As can be seen, the implementation of this method can use an intent recognition model or a preset intent tag regular expression to more accurately identify the intent of the user's conversation text data and obtain the current target intent recognition result. It can also obtain the corresponding current target recommendation information based on the current target intent recognition result and feed it back to the customer service receiver as auxiliary reference information to support user communication.

[0049] Figure 6 This is a schematic block diagram of a service data processing device based on user session intent recognition provided in an embodiment of the present invention. Figure 6 As shown, corresponding to the above-described service data processing method based on user session intent recognition, the present invention also provides a service data processing apparatus 100 based on user session intent recognition. This service data processing apparatus 100 includes a unit for executing the above-described service data processing method based on user session intent recognition. Please refer to... Figure 6 The business data processing device 100 based on user session intent recognition includes: a session text data acquisition unit 110, an intent recognition unit 120, a tag matching result acquisition unit 130, and a recommendation information acquisition unit 140.

[0050] The conversation text data acquisition unit 110 is used to acquire the target user conversation text data of the target user corresponding to the conversation intent recognition instruction in response to the conversation intent recognition instruction.

[0051] The target user conversation text data is obtained by speech recognition of user conversation voice data obtained through user interaction with the server and with user authorization.

[0052] In this embodiment, the technical solution is described with the server as the execution entity. After each user communicates with customer service via telephone or online instant messaging, the user's authorized voice conversation data is processed through a semantic recognition model to obtain user conversation text data. The following describes the complete technical solution of this application in detail using the specific process of processing the target user's conversation text data obtained after a customer completes a telephone conversation with customer service as an example. After the server obtains the target user's conversation text data (considered the target user), it uses it as data to be processed for subsequent intent recognition.

[0053] The intent recognition unit 120 is used to perform intent recognition on the target user's conversation text data through a pre-trained intent recognition model or a preset intent label regular expression to obtain the current target intent recognition result.

[0054] In this embodiment, to more quickly and accurately identify the intent of the target user's conversation text data, a multi-modal identification method can be adopted. For example, a preset intent tag regular expression can be used for short text data, while an intent recognition model can be used for long text data. Through the above multi-modal identification method, the most suitable identification method for the current scenario can be adapted to quickly obtain the current target intent recognition result.

[0055] In one embodiment, the intent recognition unit 120 is used to: The conversation role tags corresponding to each conversation statement in the target user conversation text data are obtained and added to the beginning of the corresponding conversation statement. Conversation statements with the same conversation role tags and consecutive conversation statement relationships are merged and the corresponding conversation role tags are added to the beginning of the statement to obtain preprocessed user conversation text data. If it is determined that the conversation role tags included in the preprocessed user conversation text data belong to the first preset role tag set, then the current target intent recognition result corresponding to the preprocessed user conversation text data is obtained through the preset intent tag regular expression; If it is determined that the conversation role tags included in the preprocessed user conversation text data belong to the second preset role tag set, then the intent recognition model is used to perform intent recognition on the target user conversation text data to obtain the current target intent recognition result.

[0056] The second preset character tag set includes a greater number of character tags than the first preset character tag set.

[0057] In this embodiment, to facilitate more accurate intent recognition of the target user's conversation text data, it can be preprocessed. Specifically, the conversation role tags corresponding to each conversation statement in the target user's conversation text data are obtained and added to the beginning of the corresponding conversation statement. Conversation statements with the same conversation role tags and consecutive conversation statement relationships are merged, and the corresponding conversation role tags are added to the beginning of the statement to obtain preprocessed user conversation text data. For example, the target user conversation text data after the addition of conversation role tags is as follows: "User A: Statement 1; User A: Statement 2; Customer Service: Statement 3; User A: Statement 4; Customer Service: Statement 5; User A: Statement 6". At this time, Statement 1 and Statement 2, which both correspond to User A, can be merged, and the corresponding conversation role tags can be added to the beginning of the statement to obtain preprocessed user conversation text data as "User A: Statement 1 + Statement 2; Customer Service: Statement 3; User A: Statement 4; Customer Service: Statement 5; User A: Statement 6".

[0058] In specific implementation, if the first preset role tag set is set to include customer service role tags, and the second preset role tag set is set to include both user role tags and customer service role tags, the preprocessed user conversation text data is analyzed to determine whether the conversation role tags belong to the first preset role tag set. If it is determined that the conversation role tags belong to the first preset role tag set, it indicates that there is only customer service dialogue. In this case, the current target intent recognition result corresponding to the preprocessed user conversation text data can be quickly obtained using a preset intent tag regular expression. If it is determined that the conversation role tags belong to the second preset role tag set, it indicates that there is normal dialogue between the user and customer service. In this case, the intent recognition model is used to perform intent recognition on the target user conversation text data to obtain the current target intent recognition result. Through the above multimodal recognition method, not all scenarios use the intent recognition model, which can effectively reduce the inference resource consumption of the intent recognition model, thereby saving the acquisition cost of intent recognition.

[0059] In one embodiment, the intent recognition model is a large language model with multiple hybrid reasoning modes, and includes at least a non-thinking reasoning mode and a thinking reasoning mode; the step of performing intent recognition on the target user's conversational text data through the intent recognition model to obtain the current target intent recognition result includes: If it is determined that the current session round number corresponding to the target user's session text data is less than or equal to the first preset round number, then the intent recognition model is used to perform intent recognition in the non-thinking reasoning mode to obtain the current target intent recognition result; If it is determined that the current session round number corresponding to the target user's session text data is greater than the first preset round number, then the intent recognition model is used to perform intent recognition in the thinking and reasoning mode to obtain the current target intent recognition result.

[0060] In this embodiment, even when using an intent recognition model to identify the intent of the target user's conversational text data, it is necessary to distinguish between the non-thinking reasoning mode and the thinking reasoning mode of this large language model. This is mainly determined by comparing the current conversation round number corresponding to the target user's conversational text data with a first preset round number (e.g., setting the first preset round number to 3 rounds; however, the specific implementation is not limited to the above example value and can be flexibly adjusted to other values ​​according to the user's actual needs). Specifically, when determining the current conversation round number corresponding to the target user's conversational text data, the cumulative occurrence count of each user role tag in the corresponding preprocessed user conversational text data can be counted, and the user role tag with the maximum cumulative occurrence count can be selected and its maximum cumulative occurrence count can be used as the current conversation round number.

[0061] If the current conversation round number corresponding to the target user's conversation text data is less than or equal to a first preset round number, it indicates that the current conversation corresponds to a short text dialogue scenario. The intent recognition model then performs intent recognition in the non-thinking mode (i.e., no-thinking mode) to obtain the current target intent recognition result. If the current conversation round number corresponding to the target user's conversation text data is greater than the first preset round number, it indicates that the current conversation corresponds to a long text dialogue scenario. The intent recognition model then performs intent recognition in the thinking mode (i.e., thinking mode) to obtain the current target intent recognition result. Therefore, using the non-thinking mode of the intent recognition model in short text dialogue scenarios and the thinking mode in long text dialogue scenarios can balance the processing speed and recognition accuracy of the intent recognition model.

[0062] The tag matching result acquisition unit 130 is used to perform similarity matching in the locally stored intent recognition tag classification library according to the current target intent recognition result, and to obtain the current tag matching result from the intent recognition tag classification library that meets the preset similarity condition with the current target intent recognition result.

[0063] In this embodiment, a similarity match is performed in a locally stored intent recognition tag classification library based on the current target intent recognition result. The first scenario is that a current tag matching result whose similarity to the current target intent recognition result meets a preset similarity condition can be obtained from the intent recognition tag classification library. The second scenario is that a current tag matching result whose similarity to the current target intent recognition result meets the preset similarity condition cannot be obtained from the intent recognition tag classification library. In the first scenario, it means that using the current target intent recognition result as a search condition can obtain a very similar or completely identical current tag matching result from the intent recognition tag classification library, allowing for subsequent recommendation information retrieval processing.

[0064] In the second scenario described above, if the current target intent recognition result cannot yield a very similar or identical current tag matching result in the intent recognition tag classification library, a pre-defined fallback strategy can be employed. This can be achieved through a double fallback approach using a large-scale model verification and transformation process based on the intent recognition model, along with keyword mapping. This ensures that the current tag matching result can ultimately obtain a current tag matching result from the intent recognition tag classification library that meets a pre-defined similarity condition to the current target intent recognition result. Specifically, the large-scale model verification and transformation process involves re-performing intent recognition on the target user's conversation text data using the intent recognition model or a pre-defined intent tag regular expression to obtain the current target intent recognition result. If this current target intent recognition result is not entirely identical to the previous result but can still obtain a current tag matching result from the intent recognition tag classification library that meets a pre-defined similarity condition, then the large-scale model verification and transformation process has been successfully implemented. Alternatively, the current target intent recognition result obtained after large model verification and transformation is based on a preset keyword mapping relationship that includes multiple intent recognition result keywords and convertible intent recognition results. The current target intent recognition result is then transformed into another approximate convertible intent recognition result by looking up the mapping relationship, so as to ensure that it can obtain a current tag matching result from the intent recognition tag classification library that meets the preset similarity condition with the current target intent recognition result.

[0065] The locally stored intent recognition label classification library can be obtained as follows: Acquire a dataset of labeled user conversation text (including multiple user conversation text datasets, each corresponding to a labeled intent recognition label), and aggregate the labeled intent recognition labels corresponding to the dataset to obtain an initial set of labeled intent recognition labels. Then, acquire the word vector corresponding to each initial labeled intent recognition label in the initial set, and perform K-means clustering to obtain multiple clusters. Select the word vector closest to the cluster center in each cluster as candidate word vectors, and form the intent recognition label classification library from the labeled intent recognition labels corresponding to each candidate word vector. This method allows for rapid intelligent clustering analysis based on a large dataset of labeled user conversation text, thus quickly obtaining the intent recognition label classification library.

[0066] The recommendation information acquisition unit 140 is used to acquire the current target recommendation information corresponding to the current tag matching result if it is determined that the current tag matching result is not empty, and send the current target recommendation information to the customer service receiving end.

[0067] The current target recommendation information is the recommended communication text data or the reply text of the problem to be solved corresponding to the target user.

[0068] In this embodiment, if it is determined that the current tag matching result is not empty, it means that the current tag matching result corresponds to a specific intent recognition tag. If recommendation information is set in advance for each intent recognition tag in the intent recognition tag classification library in the local knowledge base of the server, after determining the intent recognition tag of the current tag matching result, the recommendation information with the mapping relationship of the intent recognition tag (such as recommended communication text data or reply text of unresolved issues) can be obtained from the knowledge base as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end for the customer service personnel corresponding to the customer service receiving end to use as reference information, and to refer to use in the follow-up communication with the target user.

[0069] In one embodiment, the recommendation information acquisition unit 140 is used for: If it is determined that the current tag matching result belongs to the first preset intent tag set, then the recommended communication text data corresponding to the target user is obtained from the local knowledge base according to the current tag matching result as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end; If it is determined that the current tag matching result belongs to the second preset intent tag set, then the question data to be answered in the target user's conversation text data is obtained, and the question reply text to be answered and the question data to be answered is obtained from the local knowledge base as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end; If it is determined that the current tag matching result belongs to the third preset intent tag set, then the target user session text data is saved to the local preset storage area.

[0070] There is no overlap between the first preset intent tag set, the second preset intent tag set, and the third preset intent tag set.

[0071] In this embodiment, when specifically setting the first preset intent tag set, the second preset intent tag set, and the third preset intent tag set, the first preset intent tag set is for scenarios where users have a high intention to communicate, and includes intent tags such as consultation details, seeking advice, proactively asking questions, expressing needs, providing feedback, making appointments / handling matters, and proactively engaging in in-depth consultation; the second preset intent tag set is for scenarios where users have a general or moderate intention to communicate, and includes intent tags such as information query, function usage, simple consultation, basic operation, confirming information, routine consultation, and light interaction; the third preset intent tag set is for scenarios where users have a low intention to communicate, and includes intent tags such as single query, extremely simple instructions, passive response, no further needs, ending the conversation, only confirmation, no intention / silence, etc.

[0072] If the current tag matching result is determined to belong to the first preset intent tag set, it indicates that the user has a high communication intention. At this time, recommended communication text data corresponding to the target user can be obtained from the local knowledge base based on the current tag matching result as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end. The customer service personnel corresponding to the receiving end can use the specific dialogue content or recommended dialogue content included in the current target recommendation information as reference data to conduct corresponding follow-up communication with the target user.

[0073] If the current tag matching result is determined to belong to the second preset intent tag set, it indicates that the user has a general or moderate communication intention. Taking the need to promptly answer a customer's unresolved question from the previous conversation as an example, the system first retrieves the unanswered question data from the target user's conversation text data, and then obtains the unanswered question reply text from the local knowledge base as the current target recommendation information. This current target recommendation information is then sent to the customer service receiving end. The customer service personnel at the receiving end can use the unanswered question reply text included in the current target recommendation information as reference data to conduct appropriate follow-up communication with the target user.

[0074] If the current tag matching result is determined to belong to the third preset intent tag set, it indicates that the user has a low communication intention. In this case, the target user's conversation text data is saved to a local preset storage area for use as historical analysis data to optimize communication methods and content when customer service personnel communicate with users. It is evident that by more accurately determining the target preset intent tag set to which the current tag matching result belongs, the system can more accurately filter out the current target recommendation information from the knowledge base that can be used by the customer service receiver in subsequent follow-up communications.

[0075] In one embodiment, the service data processing device 100 based on user session intent recognition further includes: The user operation behavior feature acquisition unit is used to acquire user behavior trajectory data corresponding to the target user's conversation text data, and user operation behavior features corresponding to the user behavior trajectory data. The current classification result acquisition unit is used to input the user operation behavior features into the pre-trained classification model to obtain the current classification result; The first processing unit is configured to set the confidence level of the current label matching result to a preset confidence level if it is determined that the similarity between the current classification result and the current label matching result exceeds a preset similarity threshold. The second processing unit is used to update the current label matching result with the current classification result if it is determined that the similarity between the current classification result and the current label matching result does not exceed a preset similarity threshold.

[0076] In this embodiment, if a user performs an operation on the user interface of the user interaction system (such as an application) corresponding to the server but fails to obtain the required information or data before communicating online with customer service, in addition to obtaining the target user's conversation text data, the user's user behavior trajectory data for the user interaction system can also be obtained. For example, if the user behavior trajectory data specifically corresponds to page element A being clicked 3 times and viewed for 200 seconds, page element B being clicked once and viewed for 10 seconds, and page element C being clicked once and viewed for 100 seconds, then the user operation behavior features that can be extracted accordingly include three sub-features: [itemA, 3, 200], [itemB, 1, 10], and [itemC, 1, 100]. If the ID corresponding to the page element is used as the index, and the behavior feature columns include click features and browsing features, then after normalizing the click feature values ​​and browsing feature values ​​in the above three sub-features, a 3-row, 2-column user operation behavior feature can be obtained. By inputting user operation behavior features into classification models such as gradient boosting trees and random forests, the current classification result and its corresponding confidence level corresponding to the target user's conversation text data can be obtained.

[0077] If the similarity between the current classification result and the current tag matching result exceeds a preset similarity threshold (e.g., setting the preset similarity threshold to 90%, though this is not limited to the example value and can be adjusted to other values ​​according to user needs), it indicates a high degree of consistency between the intent recognition result obtained from user session data and the intent classification result obtained from user behavior trajectory data. In this case, the confidence level of the current tag matching result can be set to a preset confidence value (preferably 100%). If the similarity between the current classification result and the current tag matching result does not exceed the preset similarity threshold, it indicates a lack of consistency and contradiction between the intent recognition result obtained from user session data and the intent classification result obtained from user behavior trajectory data. In this case, the current tag matching result should be updated with the current classification result. Therefore, by comprehensively analyzing the consistency between the intent recognition results obtained from user session data and user behavior trajectory data, the user's accurate communication intent can be accurately obtained.

[0078] In one embodiment, the service data processing device 100 based on user session intent recognition further includes: The historical cumulative count acquisition unit is used to acquire the historical cumulative count of each intent recognition tag in the intent recognition tag classification library; The historical cumulative statistics update unit is used to increment the historical cumulative statistics of the corresponding target intent recognition tag in the intent recognition tag classification library by one according to the current tag matching result, so as to update the historical cumulative statistics of the target intent recognition tag.

[0079] In this embodiment, the server can also perform historical cumulative statistics on the number of times each intent recognition tag in the intent recognition tag classification library has been recorded within a set statistical time period (such as the past week, past half month, past month, past quarter, past half year, past year, etc.). For example, when a user's user session text data is processed by intent recognition to determine the current target intent recognition result, and this current target intent recognition result successfully matches one of the intent recognition tags in the intent recognition tag classification library, it is considered that the intent recognition tag has been successfully hit once, and its historical cumulative statistics need to be incremented by one to update the historical cumulative statistics of the target intent recognition tag. After the above processing, the corresponding operation and maintenance personnel of the server can directly visualize the historical cumulative statistics of each intent recognition tag in the current intent recognition tag classification library in the form of bar charts, pie charts, etc., when requesting to view the historical cumulative statistics of each intent recognition tag in the intent recognition tag classification library. It can also display the percentage of the historical cumulative statistics of each intent recognition tag in the intent recognition tag classification library relative to the total historical cumulative statistics of all intent recognition tags, thereby realizing the data dashboard function.

[0080] As can be seen, embodiments implementing this device can use an intent recognition model or a preset intent tag regular expression to more accurately identify the intent of user conversation text data and obtain the current target intent recognition result. Furthermore, based on the current target intent recognition result, the corresponding current target recommendation information can be obtained and fed back to the customer service receiver as auxiliary reference information to support user communication.

[0081] The aforementioned service data processing device based on user session intent recognition can be implemented as a computer program, which can, for example... Figure 7 It runs on the computer device shown.

[0082] Please see Figure 7 , Figure 7 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. This computer device integrates any of the service data processing devices based on user session intent recognition provided in the embodiments of the present invention.

[0083] See Figure 7 The computer device 400 includes a processor 402, a memory, and a network interface 405 connected via a system bus 401. The memory may include a storage medium 403 and internal memory 404.

[0084] The storage medium 403 may store an operating system 4031 and a computer program 4032. The computer program 4032 includes program instructions that, when executed, cause the processor 402 to perform a business data processing method based on user session intent recognition.

[0085] The processor 402 provides computing and control capabilities to support the operation of the entire computer device.

[0086] The internal memory 404 provides an environment for the computer program 4032 in the storage medium 403 to run. When the computer program 4032 is executed by the processor 402, the processor 402 can execute the above-mentioned business data processing method based on user session intent recognition.

[0087] This network interface 405 is used for network communication with other devices. Those skilled in the art will understand that... Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0088] The processor 402 is used to run a computer program 4032 stored in the memory to implement the business data processing method based on user session intent recognition as described above.

[0089] It should be understood that, in this embodiment of the invention, the processor 402 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0090] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0091] Therefore, the present invention also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the business data processing method based on user session intent recognition as described above.

[0092] The storage medium can be any computer-readable storage medium that can store program code, such as a USB flash drive, external hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0093] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0094] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0095] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0096] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0097] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A business data processing method based on user session intent recognition, applied to a server, characterized in that, include: In response to a session intent recognition command, target user session text data corresponding to the target user of the session intent recognition command is obtained; wherein, the target user session text data is obtained by speech recognition of user session voice data obtained through user interaction with the server and with user authorization. The intent recognition result of the current target intent is obtained by performing intent recognition on the target user's conversation text data through a pre-trained intent recognition model or a preset intent label regular expression. Based on the current target intent recognition result, a similarity match is performed in the locally stored intent recognition tag classification library, and a current tag matching result that meets the preset similarity condition with the current target intent recognition result is obtained from the intent recognition tag classification library; If it is determined that the current tag matching result is not empty, then the current target recommendation information corresponding to the current tag matching result is obtained, and the current target recommendation information is sent to the customer service receiving end; wherein, the current target recommendation information is the recommended communication text data or the reply text of the problem to be solved corresponding to the target user.

2. The method according to claim 1, characterized in that, The step of performing intent recognition on the target user's conversation text data using a pre-trained intent recognition model or a preset intent label regular expression to obtain the current target intent recognition result includes: The conversation role tags corresponding to each conversation statement in the target user conversation text data are obtained and added to the beginning of the corresponding conversation statement. Conversation statements with the same conversation role tags and consecutive conversation statement relationships are merged and the corresponding conversation role tags are added to the beginning of the statement to obtain preprocessed user conversation text data. If it is determined that the conversation role tags included in the preprocessed user conversation text data belong to the first preset role tag set, then the current target intent recognition result corresponding to the preprocessed user conversation text data is obtained through the preset intent tag regular expression; If it is determined that the conversation role tags included in the preprocessed user conversation text data belong to the second preset role tag set, then the intent recognition model is used to perform intent recognition on the target user conversation text data to obtain the current target intent recognition result; wherein, the number of role tags included in the second preset role tag set is greater than the number of role tags included in the first preset role tag set.

3. The method according to claim 2, characterized in that, The intent recognition model is a large language model with multiple hybrid reasoning modes, including at least a non-thinking reasoning mode and a thinking reasoning mode; the process of performing intent recognition on the target user's conversational text data through the intent recognition model to obtain the current target intent recognition result includes: If it is determined that the current session round number corresponding to the target user's session text data is less than or equal to the first preset round number, then the intent recognition model is used to perform intent recognition in the non-thinking reasoning mode to obtain the current target intent recognition result; If it is determined that the current session round number corresponding to the target user's session text data is greater than the first preset round number, then the intent recognition model is used to perform intent recognition in the thinking and reasoning mode to obtain the current target intent recognition result.

4. The method according to claim 1, characterized in that, The step of performing similarity matching in a locally stored intent recognition tag classification library based on the current target intent recognition result, and obtaining a current tag matching result from the intent recognition tag classification library that satisfies a preset similarity condition with the current target intent recognition result, includes: Obtain the current word vector of the current target intent recognition result, and obtain the tag word vector corresponding to each intent recognition tag in the intent recognition tag classification library; The word vector similarity between the current word vector and the tag word vectors of each intent recognition tag in the intent recognition tag classification library is obtained respectively, and candidate tag word vectors with word vector similarity to the current word vector that is greater than the similarity threshold in the preset similarity condition and is the maximum similarity threshold are selected. Obtain the intent recognition label corresponding to the candidate label word vector, and use it as the current label matching result.

5. The method according to claim 1, characterized in that, The step of obtaining the current target recommendation information corresponding to the current tag matching result and sending the current target recommendation information to the customer service receiving end includes: If it is determined that the current tag matching result belongs to the first preset intent tag set, then the recommended communication text data corresponding to the target user is obtained from the local knowledge base according to the current tag matching result as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end; If it is determined that the current tag matching result belongs to the second preset intent tag set, then the question data to be answered in the target user's conversation text data is obtained, and the question reply text to be answered and the question data to be answered is obtained from the local knowledge base as the current target recommendation information, and the current target recommendation information is sent to the customer service receiving end; If it is determined that the current tag matching result belongs to the third preset intent tag set, the target user session text data is saved to the local preset storage area; wherein, the first preset intent tag set, the second preset intent tag set and the third preset intent tag set do not overlap with each other.

6. The method according to claim 1, characterized in that, Before the step of obtaining the current target recommendation information corresponding to the current tag matching result and sending the current target recommendation information to the customer service receiving end, the method further includes: Obtain user behavior trajectory data corresponding to the target user's conversation text data, and user operation behavior characteristics corresponding to the user behavior trajectory data; The user operation behavior features are input into a pre-trained classification model to obtain the current classification result; If it is determined that the similarity between the current classification result and the current label matching result exceeds a preset similarity threshold, then the confidence level of the current label matching result is set to a preset confidence level value; If it is determined that the similarity between the current classification result and the current label matching result does not exceed a preset similarity threshold, then the current label matching result is updated with the current classification result.

7. The method according to claim 1, characterized in that, After the step of obtaining the current target recommendation information corresponding to the current tag matching result and sending the current target recommendation information to the customer service receiving end if it is determined that the current tag matching result is not empty, the method further includes: Obtain the historical cumulative count of each intent recognition tag in the intent recognition tag classification library; Based on the current tag matching result, increment the historical cumulative count of the corresponding target intent recognition tag in the intent recognition tag classification library by one to update the historical cumulative count of the target intent recognition tag.

8. A business data processing device based on user session intent recognition, configured on a server, characterized in that, include: The conversation text data acquisition unit is used to acquire target user conversation text data corresponding to the target user in response to a conversation intent recognition instruction; wherein, the target user conversation text data is obtained by speech recognition of user conversation voice data obtained by the user terminal interacting with the server and with user authorization. The intent recognition unit is used to perform intent recognition on the target user's conversation text data through a pre-trained intent recognition model or a preset intent label regular expression to obtain the current target intent recognition result. The tag matching result acquisition unit is used to perform similarity matching in the locally stored intent recognition tag classification library based on the current target intent recognition result, and to obtain the current tag matching result from the intent recognition tag classification library that meets the preset similarity condition with the current target intent recognition result; The recommendation information acquisition unit is used to acquire the current target recommendation information corresponding to the current tag matching result if it is determined that the current tag matching result is not empty, and send the current target recommendation information to the customer service receiving end; wherein, the current target recommendation information is the recommended communication text data or the reply text of the problem to be solved corresponding to the target user.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the business data processing method based on user session intent recognition as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the business data processing method based on user session intent recognition as described in any one of claims 1-7.

Images