Intention distribution method and device, equipment, storage medium and program product

Abstract

The invention discloses an intention distribution method and device, equipment, a storage medium and a program product, and relates to the technical field of natural language processing, and the intention distribution method comprises the steps: extracting a financial question input by a user in a financial question-answering system, and obtaining a financial entity; performing cache comparison processing on the financial problem based on the financial cache library, and generating a cache comparison failing result when a cache problem matched with the financial problem does not exist; when it is detected that the cache comparison does not pass a result, performing intention recall processing on the financial entity according to the financial intention vector library to obtain a to-be-selected financial intention; and performing intention recognition on the to-be-selected financial intention through the financial intention recognition model to obtain a user intention corresponding to the financial question. The intention recall processing is performed based on the financial intention vector library, so that the financial intention recognition model only needs to be based on the recalled to-be-selected financial intention, and the processing time delay of the financial intention recognition model is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to an intent distribution method, apparatus, device, storage medium and program product. Background Art

[0002] With the rapid development of the field of natural language, understanding the meaning expressed by people has become the most urgent need in the field of natural language. More and more application fields use artificial intelligence to interact with people in natural language, and the complete intelligent autonomous collaboration redefines the way people interact with various software and machines. The development of artificial intelligence has evolved from early supervised learning models that rely on labeled data (such as traditional machine learning, CNN, RNN, LSTM, etc.), to pre-trained models that combine unlabeled data pre-training and labeled data fine-tuning (such as BERT, Transformer, etc.), to the current large models based on large-scale unlabeled data pre-training, instruction fine-tuning and human alignment (such as GPT, LLaMa series, etc.). This process marks the evolution of intent understanding models from processing small-scale data to big data, from small models to large models, and from dedicated models to general models.

[0003] The large language model has almost omnipotent intent understanding and response capabilities, which has attracted people's attention to the seamless interaction between people and the large model. The ability to complete multiple tasks through natural language interaction demonstrates multi-scenario, multi-purpose, and interdisciplinary processing capabilities, bringing new methods and opportunities for intent understanding. In the financial field, how to combine the understanding and reasoning capabilities of the large language model to provide users with the most appropriate and rigorous services is currently a key point. The current intent recognition in the financial field mainly uses multiple sub-intent BERT models to achieve hot-swappable of the overall system, but such an implementation is unacceptable in terms of response delay for intent understanding based on the large language model. Summary of the invention

[0004] The main purpose of this application is to provide an intent distribution method, device, equipment, storage medium and program product, aiming to solve the technical problem of high system latency in the existing financial field intent recognition implemented through multiple self-intention BERT models.

[0005] To achieve the above objectives, the present application proposes an intent distribution method, which includes:

[0006] Extract the financial questions input by users in the financial question-answering system to obtain financial entities;

[0007] Performing cache comparison processing on the financial question based on the financial cache library, and generating a cache comparison failure result when there is no cache question matching the financial question;

[0008] When a cache comparison failure result is detected, the financial entity is subjected to intent recall processing according to the financial intent vector library to obtain the financial intent to be selected;

[0009] The financial intention to be selected is identified through a financial intention identification model to obtain the user intention corresponding to the financial question.

[0010] In one embodiment, the financial cache library includes a short-term financial cache library and a long-term financial cache library;

[0011] The step of performing cache comparison processing on the financial problem based on the financial cache library includes:

[0012] Performing cache comparison processing on the financial problem based on the short-term financial cache library;

[0013] When there is a short-term cache problem identical to the financial problem in the short-term financial cache library, taking the short-term cache problem as a cache problem matching the financial problem;

[0014] and / or, performing cache comparison processing on the financial problem based on a long-term financial cache library;

[0015] When there is a long-term cache problem in the long-term financial cache library whose similarity with the financial problem is greater than a preset cache similarity, the long-term cache problem is used as a cache problem matching the financial problem.

[0016] In one embodiment, when there is a long-term cached question in the long-term financial cache library whose similarity to the financial question is greater than a preset cache similarity, the step of using the long-term cached question as a cached question matching the financial question further includes:

[0017] Performing question recall processing on the financial question based on the long-term financial cache library to obtain a long-term cache question whose similarity to the financial question is greater than a preset cache similarity;

[0018] The long-term cached question is used as a cached question matching the financial question, and the user intention corresponding to the long-term question is obtained.

[0019] In one embodiment, the step of performing intent recall processing on the financial entity according to the financial intent vector library to obtain the financial intent to be selected includes:

[0020] Recalling, by the intention recall module, an initial financial intention whose similarity with the financial entity is higher than a preset similarity in the financial intention vector library;

[0021] A preset number of to-be-selected financial intentions are selected based on the similarity between the initial financial intention and the financial entity.

[0022] In one embodiment, the method further comprises:

[0023] Get the seed annotated financial corpus;

[0024] Expanding the seed annotated financial corpus to obtain an expanded financial corpus;

[0025] Determine the negative and positive example labeling problems based on the expanded financial corpus;

[0026] Performing intent recall based on the negative example labeling problem and the positive example labeling problem to obtain training data;

[0027] The initial financial intention recognition model is trained based on the training data to obtain a financial intention recognition model.

[0028] In one embodiment, the method further comprises:

[0029] Performing intent clustering on the training data to obtain an intent clustering result;

[0030] Obtaining intent comparison training data according to the intent clustering result;

[0031] A financial intention comparison model is trained according to the intention comparison training data to obtain a financial intention comparison model.

[0032] In addition, to achieve the above-mentioned purpose, the present application also proposes an intention distribution device, the intention distribution device comprising:

[0033] The financial entity extraction module is used to extract the financial questions input by the user in the financial question-answering system to obtain financial entities;

[0034] A cache module, configured to perform cache comparison processing on the financial problem based on a financial cache library, and generate a cache comparison failure result when there is no cache entity matching the financial entity;

[0035] An intention recall module, used to perform intention recall processing on the financial entity according to the financial intention vector library to obtain the financial intention to be selected when a cache comparison failure result is detected;

[0036] The intention recognition module is used to perform intention recognition on the financial intention to be selected through a financial intention recognition model to obtain the user intention corresponding to the financial question.

[0037] In addition, to achieve the above-mentioned purpose, the present application also proposes an intent distribution device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the intent distribution method described above.

[0038] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the intention distribution method described above are implemented.

[0039] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the intention distribution method described above are implemented.

[0040] One or more technical solutions proposed in this application have at least the following technical effects:

[0041] This application obtains financial entities by extracting financial questions input by users in a financial question-and-answer system; performs cache comparison processing on financial questions based on a financial cache library, and generates a cache comparison failure result when there is no cache question matching the financial question; when a cache comparison failure result is detected, performs intent recall processing on financial entities based on a financial intent vector library to obtain a financial intent to be selected; performs intent recognition on the financial intent to be selected through a financial intent recognition model to obtain the user intent corresponding to the financial question. Since the financial questions input by users are first compared through a cache library, the number of times the financial intent recognition model is used is reduced, and the processing efficiency of high-frequency questions is improved; before the financial recognition model performs intent recall processing based on a financial intent vector library, the financial intent recognition model only needs to be based on the recalled financial intent to be selected, and does not need to be based on the full intent vector library for recognition, which reduces the data that the financial intent recognition model needs to process and reduces the processing delay of the financial intent recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A flowchart of the first embodiment of the method for distributing the intent of this application is provided;

[0045] Figure 2 A flowchart diagram of the second embodiment of the method for distributing the intention of this application is provided;

[0046] Figure 3 A flowchart of the third embodiment of the method for distributing the intent of this application is provided;

[0047] Figure 4 A schematic diagram of the process of clustering intent corpus in one implementation of the intent distribution method of the present application;

[0048] Figure 5 This is a schematic diagram of the module structure of the distribution device according to an embodiment of the present application;

[0049] Figure 6 A schematic diagram of the device structure of the hardware operating environment involved in the distribution method in the embodiment of the present application.

[0050] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0051] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0052] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0053] The main solution of the embodiment of the present application is: extract the financial questions input by the user in the financial question-and-answer system to obtain financial entities; perform cache comparison processing on the financial questions based on the financial cache library, and generate a cache comparison failure result when there is no cache entity matching the financial entity; when the cache comparison failure result is detected, perform intent recall processing on the financial entity according to the financial intent vector library to obtain the financial intent to be selected; perform intent recognition on the financial intent to be selected through the financial intent recognition model to obtain the user intent corresponding to the financial question.

[0054] At present, the intent similarity in financial scenarios is strong and the rejection capability is required to be high. In the same scenario, there are many intents with similar semantics. For example, in the credit card operation scenario, there are many different limit increase instructions such as limit increase, fixed limit increase, and temporary limit increase. Directly using a large language model that has not been fine-tuned has poor accuracy and cannot distinguish similar intent scenarios well. At the same time, the capability scope of a certain intent is limited and can only handle a part of the functions. The current BERT-type intent model often needs to construct a large amount of corpus to achieve sufficient improvement for such situations.

[0055] In addition, the error repair latency requirements in financial scenarios are high, and hot plugging can be achieved without affecting the normal use of other intents. When using the BERT model for intent classification, if a downstream intent process is unavailable, hot repair can often only be achieved by hot patching the code or replacing the model.

[0056] At the same time, financial scenarios have high requirements for overall stability and latency. The reasoning time of large language models often exceeds 1 second, and the overall latency is higher than that of previous intent distribution systems. There are often a large number of similar problems in financial wealth scenarios.

[0057] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a computer, server, etc., or an electronic device, virtual device, etc. that can realize the above functions. The following takes the intention distribution device (referred to as the distribution device) as an example to illustrate this embodiment and the following embodiments.

[0058] Based on this, the embodiment of the present application provides an intention distribution method, referring to Figure 1 , Figure 1 This is a flowchart of the first embodiment of the distribution method of the present application.

[0059] In this embodiment, the intention distribution method includes steps S10 to S40:

[0060] Step S10, extracting the financial questions input by the user in the financial question-answering system to obtain financial entities.

[0061] It should be noted that the financial question-answering system can be a system based on a large language model that can intelligently search, classify and recommend answers to professional questions in the financial industry. The financial question-answering system can integrate natural language processing, knowledge graphs, deep learning and other technologies, receive questions raised by users in natural language, and provide users with accurate and comprehensive financial business information through parsing, matching, reasoning and other processes.

[0062] It is understandable that the above-mentioned financial issues may be issues related to the financial field, which may cover various aspects such as financial markets, financial institutions, financial products, and financial supervision. By analyzing the financial issues input by users, intelligent question and answer can be achieved to provide users with comprehensive and accurate recommended answers.

[0063] It should be noted that in natural language processing, an entity can be an object used in the system to describe or represent a specific or abstract thing. In a financial question-answering system, a financial entity can be related to the financial field, and can be a specific financial product or service, such as stocks, bonds, funds, deposits, etc., or an abstract financial concept, such as interest rate, exchange rate, risk, etc., which is not limited in the embodiments of this application.

[0064] It is understandable that in the financial question-and-answer system, financial entities may have clear attributes and / or relationships, and each financial entity may be identified by these attributes and / or relationships, which helps the financial question-and-answer system to more accurately understand and answer users' financial questions.

[0065] It should be noted that in natural language processing, intent can be the action or purpose that the user wants to complete by asking questions. In the financial question-answering system, intent is the key to understanding the financial questions input by users. The same intent may exist for different user expressions. By accurately identifying the financial entities and user intent in financial questions, accurate answers to financial questions can be achieved.

[0066] It is understandable that the user may input financial questions by voice input, text input or other input methods, which are not limited in the present embodiment. The financial questions may be extracted by slot extraction or other extraction methods, which are not limited in the present embodiment.

[0067] In a specific implementation, the distribution device can obtain the financial questions input by the user into the financial question-answering system, and by extracting the slots of the financial questions, the financial entities corresponding to the financial questions can be obtained. By analyzing the financial entities, the user's user intention can be determined, thereby achieving an accurate answer to the financial questions.

[0068] Step S20, performing cache comparison processing on the financial question based on the financial cache library, and generating a cache comparison failure result when there is no cache question matching the financial question.

[0069] It should be noted that in order to achieve intent distribution, the present application designs a cache mechanism. Specifically, the embodiment of the present application can cache an intent given by a large language model and a user question after slot extraction and replacement through a financial cache library, and the intent can be made a cache intent, the entity corresponding to the intent is a cache entity, and the user question corresponding to the intent is a cache question. For the cache entity, cache intent, and cache question, there may be a certain time limit (that is, the cache storage time). Within the cache storage time range, if there is a financial entity that matches the cache entity and / or a user question that matches the cache question, the large language model will not be used further for intent recognition, but the matching cache intent in the financial cache library will be directly returned.

[0070] Specifically, the embodiment of the present application can compare the financial question and / or financial entity input by the user through the financial cache library. When there is a matching cache entity and / or cache question in the financial cache library, the cache intent corresponding to the cache entity / cache question can be returned. In this way, the use of the financial intent recognition model is reduced, and the efficiency and speed of intent distribution / intent recognition are improved.

[0071] It is understandable that when the cache entities, cache intents, and cache issues stored in the cache library have been stored for a period of time equal to the cache storage period, the content may be cleared to speed up the operation of the system.

[0072] It should be understood that slot extraction may refer to identifying and extracting predefined slot information from a given text or question. Such slot information may correspond to specific entities or attributes in the text or question, such as fund name, fund type, fund code, etc. Through slot extraction, the efficiency of financial entity extraction can be improved.

[0073] It should be noted that when the slot extraction is successful to obtain the user intent corresponding to the financial entity and the user question, the distribution device can obtain the user question template corresponding to the user intent, and replace the slot of the user question template with the financial entity obtained by the slot extraction to obtain the cached question for caching after the slot replacement. The cached question is cached by the cache library, and then the user question input by the user is compared with the cached question in the cache library, which can reduce the system's usage delay on high-frequency questions. For ease of description, the embodiments of the present application and the following embodiments take financial questions as an example to illustrate the solution of the present application.

[0074] In a specific implementation, the embodiment of the present application can cache the user intents identified and processed by the financial intent recognition model and the user questions corresponding to the user after the slot extraction and replacement through the financial cache library. Within the cache storage time range, if there is a matching similar user intent, the user intent corresponding to the cached question can be directly returned. Through the design of the cache mechanism, the delay problem of the large language model on high-frequency and general user questions is greatly reduced, and the use efficiency of the system is improved.

[0075] Step S30, when a cache comparison failure result is detected, the financial entity is subjected to intent recall processing according to the financial intent vector library to obtain a financial intent to be selected;

[0076] Step S40: performing intent recognition on the financial intention to be selected through a financial intention recognition model to obtain the user intention corresponding to the financial question.

[0077] It should be noted that when there are no questions matching the financial questions in the financial cache library, that is, when the cache comparison fails, the current user question or financial entity may correspond to a large number of user intents. Directly giving the full amount to the large language model for intent distribution / intent recognition may result in a decrease in the overall intent recognition effect and recognition speed. The embodiment of the present application designs an intent vector recall model to identify financial entities and financial questions input by the user, returns the correlation between all intents and user questions in the financial intent vector library, and determines the financial intent to be selected from all intents based on the correlation, thereby narrowing the recognition range of the financial intent recognition model and improving the recognition speed of the financial intent recognition model.

[0078] It can be understood that the above-mentioned financial intent vector library is a database or data storage platform that stores all or a large amount of user intents. By comparing financial questions and financial entities with user intents in the financial intent vector library, a number of intents (the number can be selected according to the actual application) whose financial entity similarity and / or financial question similarity are at the forefront or greater than the preset similarity are recalled to obtain the financial intent to be selected. The financial intent to be selected and the financial questions and financial entities input by the user are identified by the financial recognition model, and the user intent that is closest to the financial question and / or financial entity input by the user can be obtained.

[0079] Specifically, the step of performing intent recall processing on the financial entity according to the financial intent vector library to obtain the financial intent to be selected includes: recalling, through the intent recall module, initial financial intentions in the financial intent vector library whose similarity with the financial entity is higher than a preset similarity; and selecting a preset number of financial intentions to be selected based on the similarity between the initial financial intention and the financial entity.

[0080] It is understandable that the specific value of the above-mentioned preset similarity can be selected according to the actual application situation, and the embodiment of the present application is not limited to this.

[0081] It should be noted that the embodiment of the present application implements hot plugging of vector content (user intent) through the intent vector library. In the application, the financial intent vector library can recall the most similar intents based on the user's financial problems and / or financial entities. The hot plugging of user intents is implemented through the financial intent vector library, which realizes the overall control of the effectiveness of a certain intent by the overall intent distribution system, and ensures that when problems occur in the downstream process of the intent, the overall intent can be offline by adding and deleting operations in the financial intent vector library.

[0082] It should be noted that the above-mentioned financial recognition model is a large language model used for question and answering in the financial field. Through the solution of this application, the intention understanding ability and intention understanding scope of the large language model in the financial field can be improved, and the accuracy and efficiency of intention recognition and distribution in the financial field can be improved.

[0083] The embodiment of the present application obtains financial entities by extracting financial questions input by users in a financial question-and-answer system; performs cache comparison processing on financial questions based on a financial cache library, and generates a cache comparison failure result when there is no cache question matching the financial question; when a cache comparison failure result is detected, performs intent recall processing on financial entities based on a financial intent vector library to obtain a financial intent to be selected; performs intent recognition on the financial intent to be selected through a financial intent recognition model to obtain the user intent corresponding to the financial question. Since the financial questions input by users are first compared through a cache library, the number of times the financial intent recognition model is used is reduced, and the processing efficiency of high-frequency questions is improved; before the financial recognition model performs intent recall processing based on a financial intent vector library, the financial intent recognition model only needs to be based on the recalled financial intent to be selected, and does not need to be based on the full intent vector library for recognition, which reduces the data that the financial intent recognition model needs to process and reduces the processing delay of the financial intent recognition model.

[0084] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2In the embodiment of the present application, the financial cache library includes a short-term financial cache library and a long-term financial cache library;

[0085] The step of performing cache comparison processing on the financial problem based on the financial cache library includes:

[0086] Step S21A, performing cache comparison processing on the financial question based on the short-term financial cache library;

[0087] Step S22A, when there is a short-term cache problem identical to the financial problem in the short-term financial cache library, taking the short-term cache problem as a cache problem matching the financial problem;

[0088] Step S21B, and / or, performing cache comparison processing on the financial problem based on the long-term financial cache library;

[0089] Step S22B: when there is a long-term cached question in the long-term financial cache library whose similarity with the financial question is greater than a preset cache similarity, the long-term cached question is used as a cached question matching the financial question.

[0090] It should be noted that the cache mechanism in the embodiment of the present application can mainly include two types of caches: short-term cache and long-term cache. Correspondingly, the financial cache library can include a short-term financial cache library and a long-term financial cache library. Among them, the cache module corresponding to the short-term cache mainly caches problems of a shorter time (such as minutes or seconds), while the cache module corresponding to the long-term cache can cache problems of a longer time (such as permanent level).

[0091] Specifically, for short-term cache, financial questions can be matched through the short-term financial cache library. When there is a short-term cache question that is the same as the financial question in the short-term financial cache library, the short-term cache question can be returned as a cache question that matches the financial question input by the user, and the user intent corresponding to the cache question is determined. For long-term cache, the embodiment of the present application designs a question recall module, through which a simple, general, and annotated user intent corpus can be used as the source of the question vector library (that is, the long-term financial cache library), and the financial question input by the user is matched through the question vector library. When there is a long-term cache question in the question vector library whose similarity with the financial question is greater than the preset cache similarity, the long-term cache question can be returned as a cache question that matches the financial question input by the user, and the user intent corresponding to the cache question is determined. That is, when there is a long-term cache question in the long-term financial cache library whose similarity with the financial question is greater than the preset cache similarity, the step of using the long-term cache question as a cache question matching the financial question also includes: performing question recall processing on the financial question based on the long-term financial cache library to obtain a long-term cache question whose similarity with the financial question is greater than the preset cache similarity; using the long-term cache question as a cache question matching the financial question, and obtaining a user intention corresponding to the long-term question.

[0092] The embodiment of the present application performs cache comparison processing on financial questions based on a short-term financial cache library; when there is a short-term cache question identical to the financial question in the short-term financial cache library, the short-term cache question is used as a cache question matching the financial question; and / or, performs cache comparison processing on financial questions based on a long-term financial cache library; when there is a long-term cache question in the long-term financial cache library whose similarity to the financial question is greater than a preset cache similarity, the long-term cache question is used as a cache question matching the financial question. Because the cache mechanism is designed, user questions that meet the conditions are returned with the user intent of the cached results, which greatly reduces the delay problem of the large language model on high-frequency and general user questions.

[0093] Based on the first embodiment and / or the second embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those of the above-mentioned embodiment 1 and / or the second embodiment can refer to the above introduction, and will not be repeated later. Figure 3 In an embodiment of the present application, the method further includes:

[0094] Step S100, obtaining a seed annotated financial corpus;

[0095] Step S200, expanding the seed annotated financial corpus to obtain an expanded financial corpus;

[0096] Step S300, determining negative example labeling problems and positive example labeling problems based on the expanded financial corpus.

[0097] It should be noted that in order to make full use of the generation capacity of the large language model and reduce the pressure on business annotation personnel, the embodiment of the present application uses the various financial corpus intentions annotated by business personnel and the open domain negative data as the annotated seed data, and completes the expansion of the overall corpus through the common data enhancement methods of natural language processing (such as entity replacement, word replacement, etc.) and large language model polishing generation and rewriting. The overall corpus data includes positive example data, negative example data and open domain problem data of each financial user's intention. The distribution of these three parts of data can be roughly the same, and can also be modified according to the needs of actual applications. The embodiment of the present application does not limit this.

[0098] It should be understood that the expanded financial corpus is obtained by sampling and expanding the seed annotated corpus. The positive example annotation problem and the negative example annotation problem can be obtained by annotating the expanded financial corpus with positive examples and negative examples.

[0099] It is understandable that in machine learning and natural language processing, positive data and negative data are two basic types of data used to train and supervise learning models. Among them, positive data can be samples that meet or satisfy specific rules, conditions or categories, such as financial news, financial reports, financial regulations, etc.; negative data are samples that do not meet or satisfy specific rules, conditions or categories, such as false financial news, wrong financial reports, etc.

[0100] It should be noted that the embodiment of the present application may be provided with a financial intent recall vector module to realize intent recall. The financial intent recall vector module may be obtained through training, and can be well used as a pre-module of the financial intent recognition model in the reasoning process to filter out data that is obviously irrelevant to the user's question. In the training data construction stage, the financial intent recall vector module can provide the ability to construct negative examples of positive data. In the normal reasoning process, several intents that are most similar to the user's question can be returned, and the negative candidate intents of the positive example are several intentions that are relatively dissimilar to the user's intent. By adding such negative candidate intents for the positive example, the understanding of the scope of intent capabilities can be fully enhanced.

[0101] It is understandable that the financial intent recall vector module can perform training searches in the financial intent library based on the negative example labeling problem to obtain negative example data; and perform training searches in the financial intent vector library based on the positive example labeling problem to obtain positive example data. At the same time, the financial intent recall vector module can also return negative example data of several positive examples that are relatively dissimilar to the user's question (specifically, the similarity can be lower than the preset similarity threshold). These positive example data, negative example data, and negative example data of positive examples can all be used as training data for the financial intent recognition model.

[0102] Step S400, performing intent recall based on the negative example labeling problem and the positive example labeling problem to obtain training data;

[0103] Step S500: training the initial financial intention recognition model based on the training data to obtain a financial intention recognition model.

[0104] In some implementations of the embodiments of the present application, the embodiments of the present application can also use Cot data generation, generate outputs of user intent and intent capability analysis that meets user needs through prompt guidance models, and finally provide intent classification results. Prompt guidance ensures the accuracy and effectiveness of the overall generated data and speeds up the overall data generation process.

[0105] In some implementations of the present application, due to the large number of financial intents, complex business logic, and a large number of corpora, it takes too much manpower for business personnel to mark the similarities between corpora under different intents. As shown in Figure 4, the present application can measure the similarity between intents through corpus clustering, and then obtain the similarity between corpora.

[0106] Specifically, the corpus pairs within the first intent can be extracted as training corpus with a similarity of 1, the corpus pairs in the first intent and the second intent can be extracted as training corpus with a similarity of 0.75, and the corpus pairs between the third intent that has no similarity relationship with the first intent can be extracted as training corpus with a similarity of 0, thereby realizing the construction of vector training corpus. The first intent can be any intent, and the second intent can be a similar intent to the first intent.

[0107] That is, the method also includes: performing intent clustering on the training data to obtain intent clustering results; obtaining intent comparison training data based on the intent clustering results; and training a financial intent comparison model based on the intent comparison training data to obtain a financial intent comparison model.

[0108] It is understandable that the obtained financial intent comparison model can be used for intent labeling, cache comparison, question recall and intent recall, and the embodiments of the present application are not limited to this.

[0109] The embodiment of the present application obtains a seed annotated financial corpus; performs corpus expansion on the seed annotated financial corpus to obtain an expanded financial corpus; determines negative and positive example annotation problems based on the expanded financial corpus; performs intent recall based on the negative and positive example annotation problems to obtain training data; and trains an initial financial intent recognition model based on the training data to obtain a financial intent recognition model. Since a dataset for fine-tuning the entity recognition capability of the financial intent recognition model is completed through a systematic corpus construction process, the ability to understand the user's intent on financial issues and the scope of intent capabilities is improved by constructing a special dataset composition and Cot's output method, thereby improving the accuracy of intent recognition.

[0110] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the intended distribution method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0111] This application also provides an intention distribution device, please refer to Figure 5 , the intention distribution device comprises:

[0112] The financial entity extraction module 10 is used to extract the financial questions input by the user in the financial question-answering system to obtain financial entities;

[0113] A cache module 20, configured to perform cache comparison processing on the financial problem based on a financial cache library, and generate a cache comparison failure result when there is no cache entity matching the financial entity;

[0114] The intention recall module 30 is used to perform intention recall processing on the financial entity according to the financial intention vector library to obtain the financial intention to be selected when the cache comparison fails.

[0115] The intention recognition module 40 is used to perform intention recognition on the financial intention to be selected through a financial intention recognition model to obtain the user intention corresponding to the financial question.

[0116] The intent distribution device provided by the present application adopts the intent distribution method in the above-mentioned embodiment, which can solve the technical problem of high system latency in the existing financial field intent recognition implemented through multiple self-intention BERT models. Compared with the prior art, the beneficial effects of the intent distribution device provided by the present application are the same as the beneficial effects of the intent distribution method provided by the above-mentioned embodiment, and the other technical features in the intent distribution device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0117] The present application provides an intent distribution device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the intent distribution method in the above-mentioned embodiment one.

[0118] Reference below Figure 6 , which shows a schematic diagram of the structure of an intent distribution device suitable for implementing the embodiments of the present application. The intent distribution device in the embodiments of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The intended distribution device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0119] like Figure 6 As shown, the intention distribution device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the intention distribution device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the intention distribution device to communicate with other devices wirelessly or wired to exchange data. Although the figure shows the intention distribution device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0120] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0121] The intent distribution device provided by the present application adopts the intent distribution method in the above embodiment, which can solve the technical problem of high system latency in the existing financial field where intent recognition is implemented through multiple self-intention BERT models. Compared with the prior art, the beneficial effects of the intent distribution device provided by the present application are the same as the beneficial effects of the intent distribution method provided by the above embodiment, and the other technical features in the intent distribution device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0122] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0123] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0124] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the intention distribution method in the above-mentioned embodiment.

[0125] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0126] The computer-readable storage medium may be included in the intended distribution device; or may exist independently without being assembled into the intended distribution device.

[0127] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the intended distribution device, the intended distribution device:

[0128] Extract the financial questions input by users in the financial question-answering system to obtain financial entities;

[0129] Performing cache comparison processing on the financial question based on the financial cache library, and generating a cache comparison failure result when there is no cache question matching the financial question;

[0130] When a cache comparison failure result is detected, the financial entity is subjected to intent recall processing according to the financial intent vector library to obtain the financial intent to be selected;

[0131] The financial intention to be selected is identified through a financial intention identification model to obtain the user intention corresponding to the financial question.

[0132] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0133] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0134] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0135] The readable storage medium provided in this application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned intent distribution method, and can solve the technical problem that the existing intent recognition in the financial field is implemented through multiple self-intention BERT models and the system latency is high. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the intent distribution method provided in the above-mentioned embodiment, which will not be repeated here.

[0136] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned intention distribution method when executed by a processor.

[0137] The computer program product provided by this application can solve the technical problem of high system latency in the existing financial field intent recognition implemented through multiple self-intent BERT models. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the intent distribution method provided by the above embodiment, which will not be repeated here.

[0138] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for distributing intentions, characterized in that: The method comprises: Extract the financial questions input by users in the financial question-answering system to obtain financial entities; Performing cache comparison processing on the financial question based on the financial cache library, and generating a cache comparison failure result when there is no cache question matching the financial question; When a cache comparison failure result is detected, the financial entity is subjected to intent recall processing according to the financial intent vector library to obtain the financial intent to be selected; The financial intention to be selected is identified through a financial intention identification model to obtain the user intention corresponding to the financial question.

2. The intention distribution method according to claim 1, characterized in that: The financial cache library includes a short-term financial cache library and a long-term financial cache library; The step of performing cache comparison processing on the financial problem based on the financial cache library includes: Performing cache comparison processing on the financial problem based on the short-term financial cache library; When there is a short-term cache problem identical to the financial problem in the short-term financial cache library, taking the short-term cache problem as a cache problem matching the financial problem; and / or, performing cache comparison processing on the financial problem based on a long-term financial cache library; When there is a long-term cache problem in the long-term financial cache library whose similarity with the financial problem is greater than a preset cache similarity, the long-term cache problem is used as a cache problem matching the financial problem.

3. The intention distribution method according to claim 2, characterized in that: When there is a long-term cache problem in the long-term financial cache library whose similarity to the financial problem is greater than a preset cache similarity, the step of using the long-term cache problem as a cache problem matching the financial problem further includes: Performing question recall processing on the financial question based on the long-term financial cache library to obtain a long-term cache question whose similarity to the financial question is greater than a preset cache similarity; The long-term cached question is used as a cached question matching the financial question, and the user intention corresponding to the long-term question is obtained.

4. The intention distribution method according to claim 1, characterized in that: The step of performing intention recall processing on the financial entity according to the financial intention vector library to obtain the financial intention to be selected includes: Recalling, by the intention recall module, an initial financial intention whose similarity with the financial entity is higher than a preset similarity in the financial intention vector library; A preset number of to-be-selected financial intentions are selected based on the similarity between the initial financial intention and the financial entity.

5. The intention distribution method according to claim 1, characterized in that: The method further comprises: Get the seed annotated financial corpus; Expanding the seed annotated financial corpus to obtain an expanded financial corpus; Determine the negative and positive example labeling problems based on the expanded financial corpus; Performing intent recall based on the negative example labeling problem and the positive example labeling problem to obtain training data; The initial financial intention recognition model is trained based on the training data to obtain a financial intention recognition model.

6. The intention distribution method according to claim 5, characterized in that: The method further comprises: Performing intent clustering on the training data to obtain an intent clustering result; Obtaining intent comparison training data according to the intent clustering result; A financial intention comparison model is trained according to the intention comparison training data to obtain a financial intention comparison model.

7. An intention distribution device, characterized in that: The intention distribution device comprises: The financial entity extraction module is used to extract the financial questions input by the user in the financial question-answering system to obtain financial entities; A cache module, configured to perform cache comparison processing on the financial problem based on a financial cache library, and generate a cache comparison failure result when there is no cache entity matching the financial entity; An intention recall module, used to perform intention recall processing on the financial entity according to the financial intention vector library to obtain the financial intention to be selected when a cache comparison failure result is detected; The intention recognition module is used to perform intention recognition on the financial intention to be selected through a financial intention recognition model to obtain the user intention corresponding to the financial question.

8. An intention distribution device, characterized in that: The device comprises: a memory, a processor, and an intent distribution program stored in the memory and executable on the processor, wherein the intent distribution program is configured to implement the steps of the intent distribution method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores an intent distribution program, which, when executed by a processor, implements the steps of the intent distribution method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the intention distribution method according to any one of claims 1 to 6 are implemented.

Images