A service intention recognition method, device, equipment and storage medium thereof

Abstract

The embodiment of the application belongs to the technical field of financial science and technology, is applied to a customer business intention recognition scene, and relates to a business intention recognition method, device and equipment and a storage medium thereof, and comprises the following steps: obtaining historical conversation audios between a seat terminal and a customer terminal, performing text conversion on the historical conversation audios to obtain text data; performing intention recognition on the text data by using a preset first conversation intention recognition model; constructing training samples according to an intention recognition result, the text data and a business dictionary; inputting the training samples into a pre-constructed second conversation intention recognition model based on ERNIE to perform training, obtaining a trained second conversation intention recognition model; obtaining to-be-tested conversation data, inputting the to-be-tested conversation data into the trained second conversation intention recognition model, obtaining a second recognition result, predicting a business intention of a target customer according to the second recognition result, and improving the prediction accuracy of the business intention of the customer.

Description

Technical Field

[0001] This application relates to the field of financial technology and is applied to the scenario of customer business intent recognition, and in particular to a business intent recognition method, device, equipment and its storage medium. Background Technology

[0002] With the rapid development of the financial industry, financial companies are involved in an increasingly wide range of businesses. In the financial industry, communication between call center staff and customers is often involved. Many customers will communicate with call center staff to purchase new products or obtain new services.

[0003] Especially in insurance telesales scenarios, agents need to communicate with customers by phone during ITS operations to encourage them to purchase insurance. However, due to the lack of a well-developed dialogue guidance process in the ITS system, the dialogue between junior and intermediate agents and customers is not ideal. Currently, the main approach is to identify the intent of customers' single sentences. For customers whose meaning is clear and unambiguous, their intent can be accurately identified. However, customers often have divergent thinking and express a variety of meanings. Relying solely on the current sentence often leads to ambiguity, resulting in low accuracy in business intent identification. Summary of the Invention

[0004] The purpose of this application is to provide a business intent recognition method, apparatus, device and storage medium to solve the problem that the accuracy of customer business intent recognition results is not high in the prior art.

[0005] To address the aforementioned technical problems, this application provides a method for identifying business intent, employing the following technical solution:

[0006] A method for identifying business intent includes the following steps:

[0007] The historical dialogue recordings between the agent terminal and the client terminal are acquired, and the historical dialogue recordings are converted into text using target speech recognition technology to obtain text data.

[0008] The text data is subjected to intent recognition using a preset dialogue intent first recognition model to obtain a first recognition result;

[0009] Invoke a pre-built business dictionary, and construct training samples based on the first recognition result, the text data, and the business dictionary;

[0010] The training samples are input into a pre-built second dialogue intent recognition model for training to obtain a trained second dialogue intent recognition model, wherein the second dialogue intent recognition model is a text intent recognition model based on ERNIE.

[0011] Acquire the dialogue data to be tested, input the dialogue data to be tested into the trained second dialogue intent recognition model, obtain the second recognition result, and predict the business intent of the target customer based on the second recognition result.

[0012] Furthermore, the step of converting the historical dialogue recordings into text using target speech recognition technology to obtain text data specifically includes:

[0013] The PP-ASR speech recognition component is used to acquire and process the speech signal of the historical dialogue recordings to obtain the speech signal processing results.

[0014] Based on the voice signal processing results, the historical dialogue recordings are classified to obtain the voice sequence of the agent and the voice sequence of the customer.

[0015] The voice sequence of the agent and the voice sequence of the customer are respectively input into a preset acoustic model for encoding processing to obtain the voice encoding result. The preset acoustic model is an acoustic model based on LSTM-CTC structure.

[0016] The speech encoding result is decoded using a preset decoder and language model to obtain the agent's text sequence and the customer's text sequence.

[0017] According to the built-in punctuation processing function of the PP-ASR speech recognition component, the agent's text sequence and the customer's text sequence are respectively punctuated to obtain the agent's text sequence and the customer's text sequence with punctuation.

[0018] According to the dialogue sequence, the text sequence of the agent with punctuation marks and the text sequence of the customer with punctuation marks are serialized and processed to obtain the serialized and processed result as the text data.

[0019] Furthermore, the step of using a preset dialogue intent recognition model to perform intent recognition on the text data and obtain a first recognition result specifically includes:

[0020] The text data is input into the first dialogue intent recognition model;

[0021] Using the first dialogue intent recognition model, the text data is processed sequentially according to the dialogue sequence, including word segmentation, part-of-speech analysis, and stop word removal, to obtain the text processing result.

[0022] The word frequency statistics method is used to count the top N high-frequency words in the text processing results, which are used as intent keywords. The word frequency statistics method includes the TF-IDF word frequency algorithm, and N is a positive integer greater than or equal to 1.

[0023] The intent keyword is used as the first identification result.

[0024] Furthermore, the business dictionary includes financial industry-specific terminology. The step of constructing training samples based on the first recognition result, the text data, and the business dictionary specifically includes:

[0025] Obtain the time information of historical dialogue recordings, and set the timestamp corresponding to the text data of each dialogue based on the time information;

[0026] Based on the timestamp, identify and obtain the first identification result corresponding to each of the text data, and construct the intent keyword set corresponding to each of the text data.

[0027] Using the business dictionary and the timestamp, identify and obtain the financial industry business-specific terms contained in all text data, and construct the industry-specific term set corresponding to all text data.

[0028] Based on the intent keyword set and industry-specific term set corresponding to all text data, construct the association between the intent keyword set and the industry-specific term set;

[0029] Based on the relationship between the intent keyword set and the industry-specific term set, and the intent keyword set and industry-specific term set corresponding to all text data respectively, a relationship comparison map is constructed among the intent keyword set, industry-specific term set, and text data.

[0030] By using a pre-defined thesaurus, the intent keyword set and industry-specific term set corresponding to all text data are expanded to obtain the expanded intent keyword set and industry-specific term set corresponding to all text data.

[0031] The intent keyword set, industry-specific term set, and all text data corresponding to the expanded text data are used as the training samples.

[0032] Furthermore, the step of inputting the training samples into the pre-constructed dialogue intent second recognition model for training to obtain the trained dialogue intent second recognition model specifically includes:

[0033] Obtain training samples that are input into the second dialogue intent recognition model;

[0034] Based on the relationship mapping between the intent keyword set, the industry-specific term set, and the text data, the training samples are divided into different sub-training samples. Each sub-training sample contains the text data of the current dialogue, the expanded intent keyword set corresponding to the text data, and the expanded industry-specific term set corresponding to the text data.

[0035] The text intent recognition model based on ERNIE is used to perform knowledge reinforcement learning on the text data, intent keyword set, and industry-specific term set in all sub-training samples to obtain the text intent recognition model after knowledge reinforcement learning, which serves as the second dialogue intent recognition model after training.

[0036] Furthermore, the step of using the ERNIE-based text intent recognition model to perform knowledge reinforcement learning on the text data, intent keyword set, and industry-specific term set in all sub-training samples to obtain the text intent recognition model after knowledge reinforcement learning specifically includes:

[0037] The ERNIE-based text intent recognition model is used to perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the text data in all sub-training samples.

[0038] Based on the semantic analysis results, causal relationship judgment results, and sentence structure analysis results of all sub-training samples, intent keywords are re-extracted to obtain re-extraction results.

[0039] Based on the re-extraction results and the preset thesaurus, reconstruct the intent keyword set corresponding to all text data, and obtain the reconstructed intent keyword set corresponding to each text data.

[0040] The reconstructed intent keyword set is used to replace the corresponding intent keyword set when the model is input. A relationship mapping graph is constructed among the reconstructed intent keyword set, the text data, and the expanded industry-specific terminology set corresponding to the text data to obtain a text intent recognition model completed by knowledge reinforcement learning.

[0041] Furthermore, the steps of acquiring the dialogue data to be tested, inputting the dialogue data to be tested into the trained dialogue intent second recognition model, obtaining the second recognition result, and predicting the dialogue intent of the target customer based on the second recognition result specifically include:

[0042] The trained second dialogue intent recognition model is used to perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the dialogue data to be tested.

[0043] Obtain the sentence semantic analysis results, sentence causal relationship judgment results, and sentence structure analysis results corresponding to the dialogue data to be tested;

[0044] Based on the sentence semantic analysis results, sentence causal relationship judgment results, and sentence structure analysis results corresponding to the dialogue data to be tested, intent keywords are extracted to obtain the intent keyword set corresponding to the dialogue data to be tested.

[0045] The similarity between the intent keyword set and all reconstructed intent keyword sets is calculated, and the reconstructed intent keyword set with the highest similarity is selected as the selection result.

[0046] Based on the filtering results, the text data corresponding to the reconstructed intent keyword set and the historical dialogue recordings corresponding to the text data are identified during the model training phase.

[0047] The business intent ultimately reached by the customer in the historical dialogue recordings is obtained, and the business intent is fed back to the target customer as a model prediction result.

[0048] To address the aforementioned technical problems, this application also provides a business intent recognition device, which employs the following technical solution:

[0049] A business intent recognition device, comprising:

[0050] The text data acquisition module is used to acquire historical dialogue recordings between the agent terminal and the client terminal, and to convert the historical dialogue recordings into text using target speech recognition technology to obtain text data.

[0051] The first recognition module is used to perform intent recognition on the text data using a preset dialogue intent first recognition model and obtain a first recognition result;

[0052] The training sample construction module is used to call a pre-built business dictionary and construct training samples based on the first recognition result, the text data, and the business dictionary;

[0053] The model training module is used to input the training samples into a pre-built dialogue intent second recognition model for training, and obtain a trained dialogue intent second recognition model, wherein the dialogue intent second recognition model is a text intent recognition model based on ERNIE.

[0054] The second recognition module is used to acquire the dialogue data to be tested, input the dialogue data to be tested into the trained dialogue intent second recognition model, obtain the second recognition result, and predict the business intent of the target customer based on the second recognition result.

[0055] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0056] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the business intent recognition method described above.

[0057] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0058] A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the business intent recognition method described above.

[0059] Compared with the prior art, the embodiments of this application have the following main advantages:

[0060] The business intent recognition method described in this application involves acquiring historical dialogue recordings between an agent terminal and a customer terminal, and converting these recordings into text using target speech recognition technology to obtain text data. A pre-defined first dialogue intent recognition model is then used to recognize the intent in the text data. Training samples are constructed based on the intent recognition results, the text data, and a business dictionary. These training samples are then input into a pre-built second dialogue intent recognition model based on ERNIE for training, resulting in a trained second dialogue intent recognition model. Test dialogue data is acquired and input into the trained second dialogue intent recognition model to obtain a second recognition result. Based on this second recognition result, the business intent of the target customer is predicted. Firstly, the processing results of the original dialogue intent recognition model are obtained, facilitating subsequent business intent recognition processing based on these results. Using an ERNIE-based text intent recognition model to perform further business intent recognition based on the original processing results allows for prediction not only of customer business intent from keywords but also of customer business intent based on the entire communication scenario of the dialogue recording, thus improving the accuracy of customer business intent prediction. Attached Figure Description

[0061] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0063] Figure 2 This is a flowchart of an embodiment of the business intent identification method according to this application;

[0064] Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 201 shown;

[0065] Figure 4yes Figure 2 A flowchart of a specific embodiment of step 202 shown;

[0066] Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 203 shown;

[0067] Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 204 shown;

[0068] Figure 7 yes Figure 6 A flowchart of a specific embodiment of step 603 shown;

[0069] Figure 8 yes Figure 2 A flowchart of a specific embodiment of step 205 shown;

[0070] Figure 9 This is a schematic diagram of a structure of an embodiment of the business intent recognition device according to this application;

[0071] Figure 10 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0072] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0073] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0074] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0075] like Figure 1As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0076] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0077] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0078] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0079] It should be noted that the business intent recognition method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the business intent recognition device is generally set in the server / terminal device.

[0080] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0081] Continue to refer to Figure 2 A flowchart of an embodiment of the business intent recognition method according to this application is shown. The business intent recognition method includes the following steps:

[0082] Step 201: Obtain historical dialogue recordings between the agent terminal and the client terminal, and convert the historical dialogue recordings into text using target speech recognition technology to obtain text data.

[0083] In this embodiment, the agent terminal can be an insurance agent terminal in the financial industry or a sales agent terminal in an e-commerce platform, and the historical dialogue recordings include dialogue recordings when customers communicate business intentions with agents.

[0084] Continue to refer to Figure 3 , Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 201 shown includes:

[0085] Step 301: Obtain the historical dialogue recordings between the agent terminal and the client terminal;

[0086] Step 302: Using the PP-ASR speech recognition component, the historical dialogue recording is collected and processed to obtain the speech signal processing result;

[0087] In this embodiment, the PP-ASR speech recognition component is a tool that provides ASR functionality. It offers various Chinese and English models, supports model training, and allows for model inference via command line. It also supports streaming ASR service deployment. Supporting simultaneous speech recognition and punctuation processing, its application in recognizing historical dialogue recordings between agent terminals and client terminals not only obtains speech recognition results but also allows for setting punctuation marks in the dialogue content, facilitating subsequent business intent recognition.

[0088] Step 303: Based on the voice signal processing results, classify the historical dialogue recordings to obtain the voice sequence of the agent and the voice sequence of the customer.

[0089] Specifically, since historical dialogue recordings contain both agent voices and customer voices, the different speakers often result in different timbres, which in turn cause differences in the voice signals. Based on the voice signal processing results, historical dialogue recordings can be classified to obtain agent voice sequences and customer voice sequences.

[0090] Step 304: Input the voice sequence of the agent and the voice sequence of the customer into a preset acoustic model for encoding processing to obtain the voice encoding result. The preset acoustic model is an acoustic model based on LSTM-CTC structure.

[0091] Specifically, an acoustic model based on LSTM-CTC structure is used to encode the speech sequences of the agent and the customer to obtain speech coding results. The CTC (Connectionist Temporal Classification) criterion is introduced to solve the problem of unequal length between the speech annotation sequence and the speech feature sequence, while the introduction of LSTM ensures the sequentiality of the coding results when encoding the speech sequences of the agent and the customer.

[0092] Step 305: The speech encoding result is decoded using a preset decoder and language model to obtain the agent's text sequence and the customer's text sequence.

[0093] In this embodiment, the language model is consistent with the language category of the historical dialogue recording. For example, if the historical dialogue recording is in Chinese, then the language model is a Chinese language model. Similarly, if the historical dialogue recording is in English, then the language model is an English language model.

[0094] Step 306: According to the punctuation processing function built into the PP-ASR speech recognition component, punctuation processing is performed on the agent's text sequence and the customer's text sequence respectively to obtain the punctuated agent's text sequence and the punctuated customer's text sequence.

[0095] Step 307: According to the dialogue sequence, the text sequence of the agent with punctuation marks and the text sequence of the customer with punctuation marks are serialized and processed to obtain the serialization and processing result as the text data.

[0096] Step 202: Use a preset dialogue intent first recognition model to perform intent recognition on the text data and obtain the first recognition result.

[0097] Specifically, the preset dialogue intent first recognition model refers to the company's original dialogue intent recognition model.

[0098] Continue to refer to Figure 4 , Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 202 shown includes:

[0099] Step 401: Input the text data into the first dialogue intent recognition model;

[0100] Step 402: Using the first dialogue intent recognition model, the text data is processed sequentially according to the dialogue sequence, including word segmentation, part-of-speech analysis, and stop word removal, to obtain the text processing result.

[0101] Step 403: Using the word frequency statistics method, the top N high-frequency words in the text processing results are counted as intent keywords. The word frequency statistics method includes the TF-IDF word frequency algorithm, and N is a positive integer greater than or equal to 1.

[0102] Step 404: Use the intent keyword as the first identification result.

[0103] Step 203: Invoke the pre-built business dictionary and construct training samples based on the first recognition result, the text data, and the business dictionary.

[0104] A business dictionary is constructed. In this embodiment, the business dictionary includes financial industry-specific terms. These financial industry-specific terms include, for example, health insurance, accident and health insurance, insurance application, and policy maintenance.

[0105] Continue to refer to Figure 5 , Figure 5 yes Figure 2 A flowchart of a specific embodiment of step 203 shown includes:

[0106] Step 501: Obtain the time information of historical dialogue recordings, and set the timestamp corresponding to the text data of each dialogue according to the time information;

[0107] Specifically, based on the different time information of the historical dialogue recordings, multiple historical dialogue recordings can be obtained, and timestamps can be set for the corresponding dialogue text data according to the time information, so as to distinguish the text data of each dialogue.

[0108] Step 502: Based on the timestamp, identify and obtain the first identification result corresponding to each of the text data, and construct the intent keyword set corresponding to each of the text data;

[0109] Step 503: Using the business dictionary and the timestamp, identify and obtain the financial industry business-specific terms contained in all text data respectively, and construct the industry-specific term set corresponding to all text data respectively;

[0110] Step 504: Based on the intent keyword set and industry-specific term set corresponding to all text data, construct the association between the intent keyword set and the industry-specific term set;

[0111] Step 505: Based on the relationship between the intent keyword set and the industry-specific term set, and the intent keyword set and industry-specific term set corresponding to all text data respectively, construct a relationship comparison map among the intent keyword set, industry-specific term set, and text data.

[0112] By obtaining the intent keyword set and industry-specific term set corresponding to all text data, the processing results of the original dialogue intent recognition model can be obtained, which facilitates subsequent business intent recognition processing based on the original processing results.

[0113] Step 506: Expand the intent keyword set and industry-specific term set corresponding to all text data using a preset thesaurus to obtain the expanded intent keyword set and industry-specific term set corresponding to all text data.

[0114] By using a thesaurus to expand the intent keyword set and industry-specific term set corresponding to all text data, the input training samples are more sufficient when performing business intent recognition processing again, which facilitates the improvement of the accuracy of business intent recognition.

[0115] Step 507: Use the intent keyword set, industry-specific term set, and all text data corresponding to the expanded text data as the training samples.

[0116] Step 204: Input the training samples into the pre-constructed second dialogue intent recognition model for training to obtain the trained second dialogue intent recognition model, wherein the second dialogue intent recognition model is a text intent recognition model based on ERNIE.

[0117] Continue to refer to Figure 6 , Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 204 shown includes:

[0118] Step 601: Obtain training samples input into the second dialogue intent recognition model;

[0119] Step 602: Based on the relationship mapping between the intent keyword set, the industry-specific term set, and the text data, the training samples are divided into different sub-training samples. Each sub-training sample contains the text data of the current dialogue, the expanded intent keyword set corresponding to the text data, and the expanded industry-specific term set corresponding to the text data.

[0120] Step 603: The text intent recognition model based on ERNIE is used to perform knowledge reinforcement learning on the text data, intent keyword set, and industry-specific term set in all sub-training samples to obtain the text intent recognition model after knowledge reinforcement learning, which serves as the second dialogue intent recognition model after training.

[0121] Specifically, ERNIE is a knowledge-enhanced continuous learning semantic understanding framework proposed by Baidu. It combines big data pre-training with rich knowledge from multiple sources. Through continuous learning technology, it continuously absorbs knowledge in terms of vocabulary, structure, and semantics from massive text data, enabling the model to continuously evolve. It supports semantic tasks such as entity prediction, sentence causal relationship judgment, and article sentence structure reconstruction. It can obtain natural language information from multiple dimensions such as lexical, syntactic, and semantic data from training samples, thereby improving the accuracy of business intent recognition.

[0122] Continue to refer to Figure 7 , Figure 7 yes Figure 6 A flowchart of a specific embodiment of step 603 shown includes:

[0123] Step 701: The text intent recognition model based on ERNIE is used to perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the text data in all sub-training samples respectively;

[0124] By employing an ERNIE-based text intent recognition model to perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the text data in all sub-training samples, the model not only predicts customer business intent from keywords but also trains the correlation between text data and customer business intent based on the entire communication scenario of the recorded dialogue, thereby improving the accuracy of customer business intent prediction.

[0125] Step 702: Based on the semantic analysis results, causal relationship judgment results, and sentence structure analysis results of all sub-training samples, perform intent keyword re-extraction to obtain the re-extraction results;

[0126] Step 703: Based on the re-extraction results and the preset thesaurus, reconstruct the intent keyword set corresponding to all text data, and obtain the reconstructed intent keyword set corresponding to each text data.

[0127] Step 704: Replace the original intent keyword set with the reconstructed intent keyword set when the model was input, and construct a relationship mapping graph among the reconstructed intent keyword set, the text data, and the expanded industry-specific terminology set corresponding to the text data to obtain the text intent recognition model completed by knowledge reinforcement learning.

[0128] By re-extracting intent keywords and reconstructing the intent keyword set corresponding to all text data, the recognition results of the original dialogue intent recognition model are corrected and updated. Combined with the entire dialogue voice context, it is more conducive to the accuracy of predicting customer business intent.

[0129] Step 205: Obtain the dialogue data to be tested, input the dialogue data to be tested into the trained dialogue intent second recognition model, obtain the second recognition result, and predict the business intent of the target customer based on the second recognition result.

[0130] Continue to refer to Figure 8 , Figure 8 yes Figure 2 A flowchart of a specific embodiment of step 205 shown includes:

[0131] Step 801: Using the trained second dialogue intent recognition model, perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the dialogue data to be tested;

[0132] Step 802: Obtain the sentence semantic analysis results, sentence causal relationship judgment results, and sentence structure analysis results corresponding to the dialogue data to be tested;

[0133] Step 803: Based on the sentence semantic analysis results, sentence causal relationship judgment results, and sentence structure analysis results corresponding to the dialogue data to be tested, extract intent keywords to obtain the intent keyword set corresponding to the dialogue data to be tested;

[0134] Step 804: Calculate the similarity between the intent keyword set and all reconstructed intent keyword sets, and select the reconstructed intent keyword set with the highest similarity as the selection result;

[0135] In this embodiment, the similarity calculation between the intent keyword set and all reconstructed intent keyword sets can be performed using either the cosine similarity algorithm or the edit distance method. The purpose is to select the target intent keyword set from all reconstructed intent keyword sets.

[0136] Step 805: Based on the filtering results, identify the text data corresponding to the reconstructed intent keyword set during the model training phase, and the historical dialogue recordings corresponding to the text data.

[0137] Step 806: Obtain the business intent that the customer ultimately reached in the historical dialogue recording, and feed the business intent back to the target customer as the model prediction result.

[0138] Since the business intent of the dialogue data to be tested is unknown, while the business intent of customers in historical dialogue recordings is known, by filtering the target intent keyword set from all reconstructed intent keyword sets, the corresponding customers in the historical dialogue recordings can be identified. Based on the business intent of the customers, the business intent of the target customers can be predicted, ensuring a more accurate prediction of the target customers' business intent.

[0139] This application obtains historical dialogue recordings between agent terminals and customer terminals, and performs text conversion on these recordings using target speech recognition technology to obtain text data. It then uses a pre-defined first dialogue intent recognition model to identify the intent in the text data. Based on the intent recognition results, the text data, and a business dictionary, training samples are constructed. These training samples are input into a pre-constructed second dialogue intent recognition model based on ERNIE for training, resulting in a trained second dialogue intent recognition model. Test dialogue data is then obtained and input into the trained second dialogue intent recognition model to obtain a second recognition result. Based on this second recognition result, the business intent of the target customer is predicted. By first obtaining the processing results of the original dialogue intent recognition model, subsequent business intent recognition processing can be performed again based on the original processing results. Using an ERNIE-based text intent recognition model to perform business intent recognition again based on the original processing results allows for prediction of customer business intent not only from keywords but also from the overall communication scenario of the dialogue recording, thus improving the accuracy of customer business intent prediction.

[0140] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0141] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0142] In this embodiment, historical dialogue recordings between the agent terminal and the client terminal are acquired, and the historical dialogue recordings are converted into text to obtain text data. A preset dialogue intent recognition model is used to recognize the intent in the text data. Training samples are constructed based on the intent recognition results, the text data, and a business dictionary. The training samples are input into a pre-constructed ERNIE-based dialogue intent recognition model for training, resulting in a trained dialogue intent recognition model. Test dialogue data is acquired and input into the trained dialogue intent recognition model to obtain a second recognition result. Based on the second recognition result, the business intent of the target customer is predicted. By using an ERNIE-based text intent recognition model to perform business intent recognition again based on the original processing results, the prediction of customer business intent is not only based on keywords but also on the overall communication scenario of the dialogue recording, thus improving the accuracy of customer business intent prediction.

[0143] Further reference Figure 9 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a business intent recognition device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0144] like Figure 9 As shown, the business intent recognition device 900 described in this embodiment includes: a text data acquisition module 901, a first recognition module 902, a training sample construction module 903, a model training module 904, and a second recognition module 905. Wherein:

[0145] The text data acquisition module 901 is used to acquire historical dialogue recordings between the agent terminal and the client terminal, and to convert the historical dialogue recordings into text using target speech recognition technology to obtain text data.

[0146] The first recognition module 902 is used to perform intent recognition on the text data using a preset dialogue intent first recognition model and obtain a first recognition result;

[0147] The training sample construction module 903 is used to call a pre-built business dictionary and construct training samples based on the first recognition result, the text data and the business dictionary;

[0148] The model training module 904 is used to input the training samples into the pre-constructed dialogue intent second recognition model for training, and obtain the trained dialogue intent second recognition model, wherein the dialogue intent second recognition model is a text intent recognition model based on ERNIE.

[0149] The second recognition module 905 is used to acquire the dialogue data to be tested, input the dialogue data to be tested into the trained dialogue intent second recognition model, obtain the second recognition result, and predict the business intent of the target customer based on the second recognition result.

[0150] This application obtains historical dialogue recordings between agent terminals and customer terminals, and performs text conversion on these recordings using target speech recognition technology to obtain text data. It then uses a pre-defined first dialogue intent recognition model to identify the intent in the text data. Based on the intent recognition results, the text data, and a business dictionary, training samples are constructed. These training samples are input into a pre-constructed second dialogue intent recognition model based on ERNIE for training, resulting in a trained second dialogue intent recognition model. Test dialogue data is then obtained and input into the trained second dialogue intent recognition model to obtain a second recognition result. Based on this second recognition result, the business intent of the target customer is predicted. By first obtaining the processing results of the original dialogue intent recognition model, subsequent business intent recognition processing can be performed again based on the original processing results. Using an ERNIE-based text intent recognition model to perform business intent recognition again based on the original processing results allows for prediction of customer business intent not only from keywords but also from the overall communication scenario of the dialogue recording, thus improving the accuracy of customer business intent prediction.

[0151] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0152] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0153] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 10 , Figure 10 This is a basic structural block diagram of the computer device in this embodiment.

[0154] The computer device 10 includes a memory 10a, a processor 10b, and a network interface 10c that are interconnected via a system bus. It should be noted that only the computer device 10 with components 10a-10c is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0155] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0156] The memory 10a includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 10a may be an internal storage unit of the computer device 10, such as the hard disk or memory of the computer device 10. In other embodiments, the memory 10a may also be an external storage device of the computer device 10, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Of course, the memory 10a may include both internal storage units and external storage devices of the computer device 10. In this embodiment, the memory 10a is typically used to store the operating system and various application software installed on the computer device 10, such as computer-readable instructions for a business intent recognition method. In addition, the memory 10a can also be used to temporarily store various types of data that have been output or will be output.

[0157] In some embodiments, the processor 10b may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 10b is typically used to control the overall operation of the computer device 10. In this embodiment, the processor 10b is used to execute computer-readable instructions stored in the memory 10a or to process data, such as executing computer-readable instructions for the business intent recognition method.

[0158] The network interface 10c may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 10 and other electronic devices.

[0159] The computer device proposed in this embodiment belongs to the field of financial technology and is applied in the scenario of customer business intent recognition. This application obtains historical dialogue recordings between agent terminals and customer terminals, and performs text conversion on the historical dialogue recordings using target speech recognition technology to obtain text data; it then uses a preset dialogue intent first recognition model to perform intent recognition on the text data; based on the intent recognition result, text data, and business dictionary, it constructs training samples; it inputs the training samples into a pre-constructed dialogue intent second recognition model based on ERNIE for training, obtaining a trained dialogue intent second recognition model; it obtains test dialogue data, inputs the test dialogue data into the trained dialogue intent second recognition model, obtains the second recognition result, and predicts the target customer's business intent based on the second recognition result. Firstly, the processing result of the original dialogue intent recognition model is obtained, facilitating subsequent business intent recognition processing based on the original processing result. Using an ERNIE-based text intent recognition model to perform business intent recognition again based on the original processing result allows for prediction of customer business intent not only from keywords but also from the overall communication scenario of the dialogue recording, improving the accuracy of customer business intent prediction.

[0160] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by a processor to cause the processor to perform the steps of the business intent recognition method described above.

[0161] The computer-readable storage medium proposed in this embodiment belongs to the field of financial technology and is applied in the scenario of customer business intent recognition. This application obtains historical dialogue recordings between agent terminals and customer terminals, and performs text conversion on the historical dialogue recordings using target speech recognition technology to obtain text data; it then uses a preset dialogue intent first recognition model to perform intent recognition on the text data; based on the intent recognition result, text data, and business dictionary, it constructs training samples; it inputs the training samples into a pre-constructed dialogue intent second recognition model based on ERNIE for training, obtaining a trained dialogue intent second recognition model; it obtains test dialogue data, inputs the test dialogue data into the trained dialogue intent second recognition model, obtains the second recognition result, and predicts the target customer's business intent based on the second recognition result. Firstly, it obtains the processing result of the original dialogue intent recognition model, facilitating subsequent business intent recognition processing based on the original processing result. Using an ERNIE-based text intent recognition model to perform business intent recognition again based on the original processing result allows for prediction of customer business intent not only from keywords but also from the overall communication scenario of the dialogue recording, improving the accuracy of customer business intent prediction.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0163] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for identifying business intent, characterized in that, Includes the following steps: The historical dialogue recordings between the agent terminal and the client terminal are acquired, and the historical dialogue recordings are converted into text using target speech recognition technology to obtain text data. The text data is subjected to intent recognition using a preset dialogue intent first recognition model to obtain a first recognition result; Invoke a pre-built business dictionary, and construct training samples based on the first recognition result, the text data, and the business dictionary; The training samples are input into a pre-built second dialogue intent recognition model for training to obtain a trained second dialogue intent recognition model, wherein the second dialogue intent recognition model is a text intent recognition model based on ERNIE. Acquire test dialogue data, input the test dialogue data into the trained dialogue intent second recognition model, obtain the second recognition result, and predict the target customer's business intent based on the second recognition result, specifically including: The trained second dialogue intent recognition model is used to perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the dialogue data to be tested. Obtain the sentence semantic analysis results, sentence causal relationship judgment results, and sentence structure analysis results corresponding to the dialogue data to be tested; Based on the sentence semantic analysis results, sentence causal relationship judgment results, and sentence structure analysis results corresponding to the dialogue data to be tested, intent keywords are extracted to obtain the intent keyword set corresponding to the dialogue data to be tested. The similarity between the intent keyword set and all reconstructed intent keyword sets is calculated, and the reconstructed intent keyword set with the highest similarity is selected as the selection result. Based on the filtering results, the text data corresponding to the reconstructed intent keyword set and the historical dialogue recordings corresponding to the text data are identified during the model training phase. The business intent ultimately reached by the customer in the historical dialogue recordings is obtained, and the business intent is fed back to the target customer as a model prediction result.

2. The business intent recognition method according to claim 1, characterized in that, The step of converting the historical dialogue recordings into text data using target speech recognition technology specifically includes: The PP-ASR speech recognition component is used to acquire and process the speech signal of the historical dialogue recordings to obtain the speech signal processing results. Based on the voice signal processing results, the historical dialogue recordings are classified to obtain the voice sequence of the agent and the voice sequence of the customer. The voice sequence of the agent and the voice sequence of the customer are respectively input into a preset acoustic model for encoding processing to obtain the voice encoding result. The preset acoustic model is an acoustic model based on LSTM-CTC structure. The speech encoding result is decoded using a preset decoder and language model to obtain the agent's text sequence and the customer's text sequence. According to the built-in punctuation processing function of the PP-ASR speech recognition component, the agent's text sequence and the customer's text sequence are respectively punctuated to obtain the agent's text sequence and the customer's text sequence with punctuation. According to the dialogue sequence, the text sequence of the agent with punctuation marks and the text sequence of the customer with punctuation marks are serialized and processed to obtain the serialized and processed result as the text data.

3. The business intent recognition method according to claim 1, characterized in that, The step of using a preset dialogue intent recognition model to perform intent recognition on the text data and obtain a first recognition result specifically includes: The text data is input into the first dialogue intent recognition model; Using the first dialogue intent recognition model, the text data is processed sequentially according to the dialogue sequence, including word segmentation, part-of-speech analysis, and stop word removal, to obtain the text processing result. The word frequency statistics method is used to count the top N high-frequency words in the text processing results, which are used as intent keywords. The word frequency statistics method includes the TF-IDF word frequency algorithm, and N is a positive integer greater than or equal to 1. The intent keyword is used as the first identification result.

4. The business intent identification method according to claim 1 or 3, characterized in that, The business dictionary includes specialized terms for the financial industry. The step of constructing training samples based on the first recognition result, the text data, and the business dictionary specifically includes: Obtain the time information of historical dialogue recordings, and set the timestamp corresponding to the text data of each dialogue based on the time information; Based on the timestamp, identify and obtain the first identification result corresponding to each of the text data, and construct the intent keyword set corresponding to each of the text data. Using the business dictionary and the timestamp, identify and obtain the financial industry business-specific terms contained in all text data, and construct industry-specific term sets corresponding to all text data. Based on the intent keyword set and industry-specific term set corresponding to all text data, construct the association between the intent keyword set and the industry-specific term set; Based on the relationship between the intent keyword set and the industry-specific term set, and the intent keyword set and industry-specific term set corresponding to all text data respectively, a relationship comparison map is constructed among the intent keyword set, industry-specific term set, and text data. By using a pre-defined thesaurus, the intent keyword set and industry-specific term set corresponding to all text data are expanded to obtain the expanded intent keyword set and industry-specific term set corresponding to all text data. The intent keyword set, industry-specific term set, and all text data corresponding to the expanded text data are used as the training samples.

5. The business intent recognition method according to claim 4, characterized in that, The step of inputting the training samples into the pre-constructed second dialogue intent recognition model for training to obtain the trained second dialogue intent recognition model specifically includes: Obtain training samples that are input into the second dialogue intent recognition model; Based on the relationship mapping between the intent keyword set, the industry-specific term set, and the text data, the training samples are divided into different sub-training samples. Each sub-training sample contains the text data of the current dialogue, the expanded intent keyword set corresponding to the text data, and the expanded industry-specific term set corresponding to the text data. The text intent recognition model based on ERNIE is used to perform knowledge reinforcement learning on the text data, intent keyword set, and industry-specific term set in all sub-training samples to obtain the text intent recognition model after knowledge reinforcement learning, which serves as the second dialogue intent recognition model after training.

6. The business intent recognition method according to claim 5, characterized in that, The step of using the ERNIE-based text intent recognition model to perform knowledge reinforcement learning on the text data, intent keyword set, and industry-specific term set in all sub-training samples to obtain the text intent recognition model after knowledge reinforcement learning specifically includes: The ERNIE-based text intent recognition model is used to perform sentence semantic analysis, sentence causal relationship judgment, and sentence structure analysis on the text data in all sub-training samples. Based on the semantic analysis results, causal relationship judgment results, and sentence structure analysis results of all sub-training samples, intent keywords are re-extracted to obtain re-extraction results. Based on the re-extraction results and the preset thesaurus, reconstruct the intent keyword set corresponding to all text data, and obtain the reconstructed intent keyword set corresponding to each text data. The reconstructed intent keyword set is used to replace the corresponding intent keyword set when the model is input. A relationship mapping graph is constructed among the reconstructed intent keyword set, the text data, and the expanded industry-specific terminology set corresponding to the text data to obtain a text intent recognition model completed by knowledge reinforcement learning.

7. A business intent recognition device, characterized in that, The service intent recognition device implements the steps of the service intent recognition method as described in any one of claims 1 to 6, and the service intent recognition device comprises: The text data acquisition module is used to acquire historical dialogue recordings between the agent terminal and the client terminal, and to convert the historical dialogue recordings into text using target speech recognition technology to obtain text data. The first recognition module is used to perform intent recognition on the text data using a preset dialogue intent first recognition model and obtain a first recognition result; The training sample construction module is used to call a pre-built business dictionary and construct training samples based on the first recognition result, the text data, and the business dictionary; The model training module is used to input the training samples into a pre-built dialogue intent second recognition model for training, and obtain a trained dialogue intent second recognition model, wherein the dialogue intent second recognition model is a text intent recognition model based on ERNIE. The second recognition module is used to acquire the dialogue data to be tested, input the dialogue data to be tested into the trained dialogue intent second recognition model, obtain the second recognition result, and predict the business intent of the target customer based on the second recognition result.

8. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the business intent identification method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the business intent recognition method as described in any one of claims 1 to 6.

Images