Question intention-based financial question answer generation method and device, equipment and storage medium

Abstract

The invention discloses a questioning intention-based financial question answer generation method, apparatus and device, and a storage medium, relates to the technical field of financial questions and answers, and discloses a questioning intention-based financial question answer generation method, which comprises the steps of obtaining a preset financial vocabulary set, a financial question and answer knowledge base and financial questions proposed by a user; performing intention recognition and business element extraction according to the financial question and the preset financial vocabulary set to obtain a question intention and a target business element respectively; determining a target business field according to the target business elements; determining a target skill interface according to the questioning intention, a target business element and the target business field; and determining a target answer in the financial question and answer knowledge base according to the target skill interface. Accurate and professional answering of financial questions of users is achieved, and the question and answer requirements of banks in the multi-service field can be met.

Description

Technical Field

[0001] This application relates to the field of financial question-answering technology, and in particular to a method, apparatus, device, and storage medium for generating answers to financial questions based on question intent. Background Technology

[0002] With the continuous development of the banking market and the increasing variety of financial products, users' needs for information related to banking products are becoming more diversified and complex. Both banking professionals and ordinary users need to quickly and accurately acquire knowledge in different financial fields such as bill of exchange business, cross-border business, and guarantee business.

[0003] Financial knowledge is inherently highly specialized, frequently updated, covers a wide range of business areas, and contains a large number of technical terms. However, traditional financial question-and-answer systems typically rely on simple keyword matching methods. While these methods can find relevant fragments from large amounts of text, they lack semantic understanding, often resulting in irrelevant or inaccurate answers. A drawback of traditional solutions is their inability to accurately identify the financial terminology contained in user questions, leading to insufficient ability to understand user intent and extract business elements.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a method, apparatus, device, and storage medium for generating answers to financial questions based on the intent of the question, aiming to solve the technical problem that the accuracy of answers is difficult to guarantee due to the complexity of professional terminology in financial question-and-answer scenarios.

[0006] To achieve the above objectives, this application proposes a method for generating answers to financial questions based on the intent of the question, the method comprising:

[0007] Acquire a pre-defined financial vocabulary set, a financial Q&A knowledge base, and financial questions submitted by users;

[0008] Based on the financial question and the preset financial vocabulary set, intent recognition and business element extraction are performed to obtain the question intent and target business element, respectively.

[0009] The target business area is determined based on the target business elements;

[0010] The target skill interface is determined based on the questioning intent, the target business elements, and the target business domain.

[0011] The target answer is determined in the financial Q&A knowledge base based on the target skill interface.

[0012] In one embodiment, the step of performing intent recognition and business element extraction based on the financial question and the preset financial vocabulary set to obtain the question intent and target business elements respectively includes:

[0013] Obtain a standard business element library, preset similarity thresholds, and preset edit distance thresholds;

[0014] The financial issue is segmented into words to obtain product slots, and the product slots are matched with words according to the preset financial vocabulary set to obtain subdivided business elements.

[0015] Based on the standard business element library, the preset similarity threshold, the preset edit distance threshold, and the subdivided business elements, business elements are extracted to obtain the target business elements;

[0016] Based on the financial question and the target business element, the intent of the question is identified.

[0017] In one embodiment, the step of extracting target business elements based on the standard business element library, the preset similarity threshold, the preset edit distance threshold, and the subdivided business elements includes:

[0018] Calculate the business similarity between the segmented business element and the elements in the standard business element library;

[0019] When the business similarity is greater than the preset similarity threshold, the standard business element corresponding to the business similarity is taken as the target business element.

[0020] When the business similarity is less than or equal to the preset similarity threshold, the edit distance between the pinyin of the subdivided business element and the pinyin of the element in the standard business element library is calculated;

[0021] If the edit distance is less than the preset edit distance threshold, the standard business element corresponding to the edit distance is taken as the target business element.

[0022] In one embodiment, the step of determining the target business domain based on the target business element includes:

[0023] Obtain a preset business domain classification table and a preset business domain quantity threshold, wherein the preset business domain classification table contains a mapping relationship between business domains, corresponding product types, and business scenario keywords;

[0024] Based on the target business elements, determine product type information, business scenario keywords, and business context information;

[0025] The product type information and the business scenario keywords are matched with the contents of the preset business domain classification table to obtain the business domain matching results, and the number of successfully matched business domains is determined.

[0026] When the number of matched business areas is less than or equal to the preset threshold for the number of business areas, the business areas whose matching results are successful are taken as the target business areas;

[0027] When the number of matched business domains is greater than the preset threshold for the number of business domains, the target business domain is determined from the business domain matching results based on the business context information.

[0028] In one embodiment, the step of determining the target skill interface based on the question intent, the target business element, and the target business domain includes:

[0029] Obtain the current user's channel code and determine the corresponding channel orchestration configuration table based on the channel code;

[0030] Based on the target business area and the question intent, a list of candidate skill interfaces is determined in the channel orchestration configuration table;

[0031] The candidate skill interface list is filtered for relevance based on the target business elements to obtain the relevance filtering results, and the target skill interface is determined based on the relevance filtering results.

[0032] In one embodiment, after the steps of obtaining the current user's channel code and determining the corresponding channel orchestration configuration table based on the channel code, the method further includes:

[0033] When no corresponding channel orchestration configuration table is found based on the channel code, the public database is obtained, and the general skill interface corresponding to the public database is used as a candidate skill interface.

[0034] If no filtering requirement is detected in the candidate skill interface, an alternative skill interface is obtained, wherein the alternative skill interface includes basic question and answer interfaces for each business domain.

[0035] If the filtering requirement is not detected in the backup skill interface, the basic skill interface is obtained and used as the target skill interface.

[0036] When the filtering requirement is detected to be included in the backup skill interface, the backup skill interface is used as the target skill interface.

[0037] In one embodiment, the step of determining the target answer in the financial question-and-answer knowledge base based on the target skill interface includes:

[0038] The search results are obtained by performing a search on the financial Q&A knowledge based on the target skill interface.

[0039] The criteria for determining whether the search results are valid answers are: containing detailed business information, having unambiguous answer information, and correctly matching the question intent.

[0040] When the search result is a valid answer, the valid answer is formatted to obtain the target answer;

[0041] When the search results are empty or not a valid answer, a preset basic word set is obtained, and it is determined whether the financial question matches the keywords in the preset basic word set.

[0042] When the financial question is matched with keywords in the preset basic word set, the target answer is determined based on the keywords.

[0043] In addition, to achieve the above objectives, this application also proposes a financial question answer generation device based on question intent, the financial question answer generation device based on question intent includes: a data acquisition module, used to acquire a preset financial vocabulary set, a financial question and answer knowledge base and financial questions raised by users;

[0044] The intent recognition module is used to perform intent recognition and business element extraction based on the financial question and the preset financial vocabulary set, so as to obtain the question intent and the target business element respectively.

[0045] The domain determination module is used to determine the target business domain based on the target business elements.

[0046] The interface determination module is used to determine the target skill interface based on the question intent, the target business element, and the target business domain.

[0047] The question-answering module is used to determine the target answer in the financial question-answering knowledge base based on the target skill interface.

[0048] Furthermore, to achieve the above objectives, this application also proposes a financial question answer generation device based on question intent, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the financial question answer generation method based on question intent as described above.

[0049] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the financial question answer generation method based on question intent as described above.

[0050] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the financial question answer generation method based on question intent as described above.

[0051] One or more technical solutions proposed in this application have at least the following technical effects:

[0052] By introducing a pre-set financial vocabulary set and combining it with intent recognition and business element extraction, this approach solves the problems of inaccurate identification of financial terminology and inability to correctly understand user question intent in general large models. By constructing a financial question-and-answer knowledge base and determining the target answer from this knowledge base based on the target skill interface, it avoids the high maintenance costs of traditional rule bases and the poor semantic understanding of simple retrieval technologies. Furthermore, by determining the target skill interface according to the target business domain and question intent, it provides a unified processing path for business question-and-answer in different domains, solving the problem of separate creation and maintenance of question-and-answer for different domains, which incurs high manpower costs. Compared with existing technologies, the pre-set financial vocabulary set improves the accuracy of identifying user financial question intent and extracting business elements, thereby ensuring the rationality of subsequently determining the target business domain and target skill interface. Finally, by obtaining the target answer from the financial question-and-answer knowledge base through the target skill interface, it improves the accuracy and professionalism of financial question answers. Attached Figure Description

[0053] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart is provided for Embodiment 1 of the method for generating answers to financial questions based on question intent in this application;

[0056] Figure 2 A partial knowledge base diagram provided for Embodiment 1 of the financial question answer generation method based on question intent in this application;

[0057] Figure 3 This is a schematic diagram of the business element extraction process provided in Embodiment 1 of the financial question answer generation method based on question intent in this application;

[0058] Figure 4This is a schematic diagram of the overall question-and-answer architecture provided in Embodiment 1 of the financial question answer generation method based on question intent in this application;

[0059] Figure 5 A flowchart illustrating Embodiment 2 of the method for generating answers to financial questions based on question intent in this application;

[0060] Figure 6 A simplified flowchart illustrating the financial question answer generation method based on question intent provided in Embodiment 2 of this application;

[0061] Figure 7 This is a schematic diagram of the module structure of the financial question answer generation device based on question intent in an embodiment of this application;

[0062] Figure 8 This is a schematic diagram of the device structure of the hardware operating environment involved in the financial question answer generation method based on question intent in the embodiments of this application.

[0063] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0064] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0065] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0066] The main solution of this application embodiment is as follows: acquiring a preset financial vocabulary set, a financial Q&A knowledge base, and financial questions raised by users; performing intent recognition and business element extraction based on the financial questions and the preset financial vocabulary set to obtain the question intent and target business element respectively; determining the target business domain based on the target business element; determining the target skill interface based on the question intent, target business element, and target business domain; and determining the target answer in the financial Q&A knowledge base based on the target skill interface.

[0067] In this embodiment, for ease of description, the following description will focus on the device that identifies and generates answers to financial questions based on the intent of the question.

[0068] Because existing technologies require traditional banking and financial question-answering systems to be built based on rule bases, which necessitate the manual writing of a large number of rules, and because banking product knowledge is updated rapidly and business is complex, resulting in extremely high maintenance costs, systems built based on simple retrieval technologies cannot understand the semantics of user questions and often return irrelevant or inaccurate results. Directly applied general-purpose models, due to a lack of in-depth optimization for financial knowledge, cannot accurately identify professional terms in user questions, leading to insufficient accuracy and professionalism in the answers. Furthermore, different business areas within the banking industry require question-answering but lack unified management, necessitating extremely high human resource costs to create and maintain question-answering systems for each area.

[0069] This application provides a solution that addresses the problems of inaccurate identification of financial terminology and inability to correctly understand user question intent by introducing a pre-set financial vocabulary set and combining it for intent recognition and business element extraction. It constructs a financial question-and-answer knowledge base and determines the target answer from this knowledge base based on the target skill interface, avoiding the high maintenance costs of traditional rule bases and the poor semantic understanding of simple retrieval technologies. Furthermore, by determining the target skill interface according to the target business domain and question intent, it provides a unified processing path for business question-and-answer in different domains, solving the problem of separate creation and maintenance of question-and-answer for different domains, which incurs high manpower costs. Compared with existing technologies, the pre-set financial vocabulary set improves the accuracy of identifying user financial question intent and extracting business elements, thereby ensuring the rationality of subsequently determining the target business domain and target skill interface. Finally, the target answer is obtained from the financial question-and-answer knowledge base through the target skill interface, improving the accuracy and professionalism of financial question answers.

[0070] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as a financial question answer generation device based on question intent. The following description uses a financial question answer generation device based on question intent as an example to illustrate this embodiment and the subsequent embodiments.

[0071] Based on this, embodiments of this application provide a method for generating answers to financial questions based on question intent, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the financial question answer generation method based on question intent in this application.

[0072] In this embodiment, the method for generating answers to financial questions based on question intent includes steps S10 to S50:

[0073] Step S10: Obtain a preset financial vocabulary set, a financial Q&A knowledge base, and financial questions raised by users;

[0074] The pre-defined financial glossary is a collection of professional terms specifically built for the banking and finance sector. It includes various professional expressions in the banking and public finance field, such as product names, business process keywords, and business rule terms. These terms are derived from internal bank business documents, product descriptions, and industry standards, and have been screened and organized to help accurately identify key information in financial issues. This avoids errors in the segmentation of professional terms caused by general vocabulary segmentation, ensuring the accuracy of subsequent intent recognition and business element extraction.

[0075] The financial Q&A knowledge base is a collection of knowledge built upon the bank's Retrieval-Augmented Generation Center (RAG Center), comprising two parts: a product knowledge base and a corporate knowledge base. The product knowledge base stores knowledge about over 280 products, existing in both document and question-and-answer pairs formats. Documents store complete information for each product with segmentation tags inserted across different modules. Question-and-answer pairs generate combinations of questions and answers for different modules within each product. Financial questions submitted by users refer to inquiries related to banking and financial services that users input when using the bank's Q&A service.

[0076] In one feasible implementation, before acquiring the preset financial vocabulary set, financial Q&A knowledge base, and user-submitted financial questions, the following steps are included: Product knowledge base creation: Based on the bank's RAG center, a corporate banking product document and QA knowledge base (with Chinese explanations) is built, importing knowledge from over 280 corporate banking product knowledge bases. The product document knowledge base stores product knowledge in document form, generating a document for each product. Different modules of the product are directly marked with slice tags. The documents are imported into the RAG center, and the model is initiated to slice and self-learn the documents. The product QA knowledge base stores product knowledge in QA pairs, generating QA pairs for different modules of each product, such as (Q: What are the advantages of the product? A: The advantages of the product, etc.). The QA pairs for each product are imported into the RAG center, and the model is initiated to self-learn. Corporate banking domain knowledge base creation: Based on the bank's RAG center, knowledge bases are built for areas such as salary payment, bills, cross-border transactions, cash pooling, and settlement. Each domain's knowledge base stores knowledge in either document or QA pair form, and the model is initiated to self-learn.

[0077] Reference Figure 2 , Figure 2 This is a partial knowledge base diagram of the first embodiment of the financial question answer generation method based on question intent in this application. Figure 2As shown, each row represents a different knowledge base, from top to bottom: Knowledge Base ID 13408, original knowledge base ID 240750532, knowledge base name: ACS Certification Materials, knowledge base source: document library; Knowledge Base ID 10258, original knowledge base ID 228780238, knowledge base name: Inclusive Finance Q&A Resource Library, knowledge base source: document library; Knowledge Base ID 10249, original knowledge base ID 228540150, knowledge base name: Inclusive Finance Q&A Library, knowledge base source: QA Library; Knowledge Base ID 13344, original knowledge base ID 240766685, knowledge base name: Centralized Income Intelligent Accounting QA Document Library, knowledge base source: document library; Knowledge Base ID 13360, original knowledge base ID 240... 767017, Knowledge Base Name: Income Accounting QA Base, Knowledge Base Source: QA Base; Knowledge Base ID: 12536, Original Knowledge Base ID: 239664400, Knowledge Base Name: Domestic Certificate Knowledge Base - 10098165122051219154, Knowledge Base Source: Document Base; Knowledge Base ID: 12630, Original Knowledge Base ID: 240740592, Knowledge Base Name: Cross-border Business QA Base, Knowledge Base Source: QA Base; Knowledge Base ID: 12631, Original Knowledge Base ID: 240740700, Knowledge Base Name: Cross-border Business Document Base, Knowledge Base Source: Document Base; Knowledge Base ID: 10734, Original Knowledge Base ID: 230761774, Knowledge Base Name: Salary & Benefits QA Base (Internal), Knowledge Base Source: QA Base.

[0078] Step S20: Based on the financial question and the preset financial vocabulary set, perform intent recognition and business element extraction to obtain the question intent and target business elements respectively;

[0079] Intent identification refers to the process of analyzing a user's financial question to determine the type of core need the user intends to achieve through that question. For example, if a user inquires about the product application process, their intent is "business process consultation intent"; if a user inquires about product advantages, their intent is "product information inquiry intent." The core of intent identification is to extract the user's potential needs from the expression of the financial question, providing a basis for determining the subsequent processing path. By combining a pre-set financial vocabulary set and large-scale model capabilities, it is possible to more accurately determine the type of the user's core need.

[0080] Business element extraction refers to the process of filtering out key information related to banking and financial services from user-submitted financial questions. This key information includes product names, business areas, and business scenarios. For example, if a user asks, "What is the discount rate for bills?", the extracted business elements would be "bill business, discount rate". Business element extraction must ensure the accuracy and completeness of the extracted information, as it is crucial for subsequently determining the target business area and target skill interface. Using a pre-set financial vocabulary set can prevent the incorrect breakdown of professional business elements and improve extraction accuracy.

[0081] Question intent is the result of intent recognition, a clear definition of the user's core financial needs, such as "product information inquiry intent," "business process consultation intent," and "business rule confirmation intent." Different question intents correspond to different processing methods. For example, "business process consultation intent" requires retrieving relevant business process knowledge to answer, while "product information inquiry intent" requires retrieving relevant product details knowledge. The accuracy of the question intent directly affects whether the subsequent processing steps are correct.

[0082] The target business element is the final result of business element extraction; it is the verified and standardized key business information extracted from financial questions. During the extraction process, initial business elements are first obtained, and then undergo a series of verification and standardization processes, such as word segmentation, similarity matching, and pinyin edit distance matching, to ensure that the obtained business elements are accurate and conform to the bank's internal standard expressions. For example, potentially non-standard expressions in user questions such as "bill discounting business" are standardized to "bill discounting business." The target business element is the core basis for subsequently determining the target business area.

[0083] Understandably, a large model is used to process the acquired financial questions, extracting initial business elements. These initial business elements may include product names and business keywords directly mentioned in the user's question. Next, Jieba word segmentation technology is used to segment these initial business elements. During segmentation, financial terms from a pre-defined financial vocabulary set are prioritized to avoid splitting complete financial terms into multiple ordinary words, thus obtaining subdivided business elements. Then, the similarity between the subdivided business elements and elements in a standard business element library is calculated. If the similarity exceeds a first pre-defined threshold, the corresponding standard business element is used as the target business element. If the similarity does not exceed the first pre-defined threshold, the edit distance between the pinyin of the subdivided business element and the pinyin of elements in the standard business element library is calculated. If the edit distance is less than a second pre-defined threshold, the corresponding standard business element is also used as the target business element. Finally, the financial question and the determined target business elements are input into the large model, which analyzes the core needs of the user's question, identifies, and outputs the question's intent.

[0084] In one feasible implementation, step S20 may include steps S21 to S24:

[0085] Step S21: Obtain the standard business element library, preset similarity threshold, and preset edit distance threshold;

[0086] The standard business element library is a collection of standardized business element information. This information is pre-organized and determined based on business rules, product characteristics, and industry standards in the financial sector. The library contains various standard elements involved in financial business, such as standard product names, standard business types, and standard processing steps. Each element has a unified expression format to avoid identification errors caused by differences in expression.

[0087] The preset similarity threshold is a pre-set critical value used to judge the degree of similarity between subdivided business elements and elements in the standard business element library. The preset edit distance threshold is a pre-set critical value used to judge the degree of difference between the pinyin of subdivided business elements and the pinyin of standard business elements.

[0088] Step S22: The financial question is segmented to obtain product slots, and the product slots are matched with words according to the preset financial vocabulary set to obtain subdivided business elements.

[0089] Word segmentation is the process of breaking down a user's financial question into multiple independent words or phrases according to certain rules. In this embodiment, Jieba Segmentation (Jieba) is used for word segmentation. This technology combines language habits and lexical features to break down a continuous text sequence into meaningful lexical units.

[0090] Product slots are words or phrases directly related to financial products extracted from financial questions after word segmentation. They are the core objects of subsequent business element analysis. These words or phrases typically include information such as product names and related services.

[0091] Lexical matching involves comparing extracted product slots with words in a pre-defined financial vocabulary set to determine if a product slot contains any of the pre-defined financial terms. During the comparison, each word in a product slot is checked against the pre-defined financial vocabulary set. If a word exists, it is extracted; otherwise, the original word structure of the product slot is preserved. Its purpose is to filter out specialized financial terms from product slots, preventing the breakdown of specialized terms and ensuring that the subsequently obtained segmented business elements conform to the professional expression habits of the financial field.

[0092] The segmented business elements are the result of word matching, that is, the preset financial words selected from the product slots, or the original words in the product slots that are retained when no preset financial words are matched.

[0093] Understandably, the process involves segmenting user-submitted financial questions using Jieba segmentation technology. Following language conventions and the characteristics of financial terminology, the questions are broken down into multiple independent words or phrases. Content directly related to financial products is extracted from these words or phrases, resulting in product slots. Next, a vocabulary matching process is initiated, comparing each extracted product slot with a pre-defined financial vocabulary set. The process checks if the product slot contains any professional terms from the pre-defined financial vocabulary set. If a product slot contains a term that matches the pre-defined financial vocabulary set, that term is used as a sub-business element. If no term matches the pre-defined financial vocabulary set, the product slot itself is used as a sub-business element, thus completing the transformation from financial questions to sub-business elements.

[0094] Step S23: Extract business elements based on the standard business element library, the preset similarity threshold, the preset edit distance threshold, and the subdivided business elements to obtain the target business element;

[0095] Understandably, the process involves comparing each segmented business element with a standard element in the standard business element library, calculating the similarity between the segmented business element and each standard element. This similarity is then compared to a preset similarity threshold. If the similarity between a standard element and a segmented business element is greater than or equal to the preset threshold, they are considered a match, and the standard element is directly designated as the target business element. If the similarity between all standard elements and segmented business elements is less than the preset threshold, the pinyin of the segmented business element and the pinyin of each standard element in the standard business element library are further extracted, and the edit distance between the pinyin of the segmented business element and the pinyin of each standard element is calculated. This edit distance is then compared to a preset edit distance threshold. If the edit distance between the pinyin of a standard element and the pinyin of a segmented business element is less than or equal to the preset edit distance threshold, they are considered a match, and the standard element is designated as the target business element. If the edit distance between the pinyin of all standard elements and the pinyin of segmented business elements is greater than the preset edit distance threshold, the accuracy of the segmented business element needs to be re-verified, or the standard business element library needs to be supplemented and improved, until a target business element that meets the specifications is determined.

[0096] In one feasible implementation, step S23 may include steps S231 to S234:

[0097] Step S231: Calculate the business similarity between the segmented business element and the elements in the standard business element library;

[0098] Business similarity is an indicator that measures the degree of similarity between a specific business element and each element in a standard business element library in terms of business meaning and relevance of expression. Its calculation incorporates business logic in the financial field, comparing not only the superficial overlap of the text but also considering the relevance of business attributes (such as the product's industry and the business processing scenario).

[0099] The elements in the standard business element library are standardized business information units stored within the library that conform to business norms and industry standards in the financial field. The process involves retrieving all standard elements potentially related to the subdivided business elements from the library (usually starting with a preliminary screening based on the business domain of the subdivided business elements to narrow down the comparison scope; for example, if a subdivided business element involves "XinFuTong," standard elements within the XinFuTong business domain are retrieved first). Next, a preset similarity calculation method is used (this method combines text matching and business attribute relevance; text matching compares the text overlap ratio between the subdivided business element and the standard element, while business attribute relevance considers the degree of association between the two in dimensions such as business domain, service recipients, and processing procedures). The subdivided business element is then compared with each of the screened standard elements. Finally, the business similarity values ​​between the subdivided business element and each standard element are recorded to form a similarity list, preparing for subsequent comparisons with a preset similarity threshold.

[0100] Step S232: When the business similarity is greater than the preset similarity threshold, the standard business element corresponding to the business similarity is taken as the target business element.

[0101] The standard business element corresponding to business similarity refers to the specific element from the standard business element library that corresponds to the business similarity value calculated with the subdivided business elements.

[0102] Extract the similarity value of each business and its corresponding standard business element. Then, compare each business similarity value with a preset similarity threshold one by one to determine if the value is greater than the preset similarity threshold. If a business similarity value is greater than the preset similarity threshold, it means that its corresponding standard business element and subdivided business element are highly matched at the business level and meet the requirements of the target business element. At this time, the standard business element is directly determined as the target business element. If there are multiple business similarity values ​​that are greater than the preset similarity threshold, the standard business element with the highest business similarity value is selected as the target business element (if the highest similarity value corresponds to multiple standard elements, further screening is required based on the specific business scenario of the subdivided business element to ensure the uniqueness and accuracy of the target business element). After determining the target business element, record the relevant information of the element (such as its business domain, business definition, etc.) to provide data support for subsequent steps.

[0103] Step S233: When the business similarity is less than or equal to the preset similarity threshold, calculate the edit distance between the pinyin of the subdivided business element and the pinyin of the element in the standard business element library;

[0104] The pinyin representation of subdivided business elements involves converting the corresponding Chinese characters of these elements into a sequence of pinyin characters. This conversion process follows pinyin rules, including tone markings. To simplify calculations, pinyin without tones can be used, but it must be consistent with the conversion rules for standard business element pinyin. The pinyin representation of elements in the standard business element library involves converting the Chinese character representation of each standard element in the library into a sequence of pinyin characters. Edit distance is a metric that measures the degree of difference between two strings; it refers to the minimum number of single-character edit operations required to convert one string into another.

[0105] According to the preset pinyin conversion rules, the Chinese character descriptions of the subdivided business elements are converted into corresponding pinyin strings to obtain the pinyin of the subdivided business elements. All standard elements previously involved in business similarity calculations are retrieved from the standard business element library, and their Chinese character descriptions are converted into pinyin strings according to the same conversion rules as the pinyin of the subdivided business elements to obtain the element pinyin in the standard business element library. Using a preset edit distance calculation algorithm (such as the Levinstein distance algorithm), the pinyin of each subdivided business element is calculated against the pinyin of each standard business element to determine the minimum number of single-character editing operations required to convert the pinyin of the subdivided business element to the pinyin of each standard business element. The edit distance values ​​between the pinyin of the subdivided business element and the pinyin of each standard business element are recorded to form an edit distance list, preparing for subsequent comparison with a preset edit distance threshold.

[0106] Step S234: When the editing distance is less than the preset editing distance threshold, the standard business element corresponding to the editing distance is taken as the target business element.

[0107] The standard business element corresponding to the edit distance refers to the specific element from the standard business element library that corresponds to the edit distance value calculated from the pinyin of the subdivided business elements.

[0108] Extract each edit distance value and its corresponding standard business element; compare each edit distance value with a preset edit distance threshold one by one to determine whether the value is less than the preset edit distance threshold; if an edit distance value is less than the preset edit distance threshold, it means that the difference between the pinyin of its corresponding standard business element and the pinyin of its subdivided business element is very small, and the corresponding business element is highly matched in business meaning, which meets the requirements of the target business element; determine the standard business element as the target business element; if there are multiple edit distance values ​​that are all less than the preset edit distance threshold, select the standard business element with the smallest edit distance value as the target business element.

[0109] Step S24: Based on the financial question and the target business element, perform intent identification to obtain the questioning intent.

[0110] Understandably, the question intent is the final result of intent identification in this step; it is the user's specific expression of needs related to the target business element. Examples include inquiries about bill discounting procedures, salary and fee standards for the "XinFuTong" product, and document preparation for cross-border declarations. These intents not only include the user's need type but also relate to the corresponding target business element. Their function is to bind user needs to specific business elements, ensuring accurate matching of knowledge resources corresponding to that business element and need type, thus avoiding irrelevant answers.

[0111] Reference Figure 3 , Figure 3 This is a schematic diagram of the business element extraction process for the first embodiment of the financial question answer generation method based on question intent in this application. Figure 3 As shown, starting with a user inquiry, the process first extracts product slots using a large model, then segments these slots using financial terminology, calculates the similarity between the segmented words and the product name, and outputs a standard product name based on the similarity threshold. This process consists of two main parts: similarity-based matching of standard products and pinyin edit distance matching of standard products. In the similarity-based matching of standard products part, the product name is first extracted from financial terminology, then the similarity between the segmented words and the product name is calculated. If a match is successful, the standard product name is output. If no match is found, the process moves to the pinyin edit distance matching of standard products part. The pinyin of the product name is extracted from the standard product pool, the edit distance between the segmented pinyin and the product name pinyin is calculated, and the standard product name is output based on the edit distance threshold. After product name standardization, the user question is replaced, then a RAG search is performed, and finally, the large model generates the answer.

[0112] Step S30: Determine the target business area based on the target business elements;

[0113] It's important to note that the target business area refers to a specific business category within the bank that relates to the user's financial questions. Examples include areas such as bills, cross-border transactions, payroll services, cash pooling, and settlement. Each business area corresponds to a set of specialized business knowledge and processing procedures. Once the target business area is identified, the corresponding knowledge base and skill interfaces can be accessed to ensure professional and accurate answers are provided to users. Identifying the target business area is a crucial step in achieving precise question-and-answer sessions across different business areas.

[0114] Understandably, the process involves extracting product type information and business scenario keywords from the identified target business elements. Next, a pre-defined business area classification table is retrieved from within the bank, recording the mapping relationship between each business area and its corresponding product type and business scenario keywords. The extracted product type information and business scenario keywords are then matched against the contents of the pre-defined business area classification table to see if a completely matching business area exists. If a unique matching business area is found, that business area is directly designated as the target business area. If multiple candidate business areas are matched, for example, if the product type information simultaneously matches the mapping relationship of two business areas, further filtering is performed using contextual information from the financial question, such as analyzing other statements in the question to determine the business focus, ultimately determining the unique target business area.

[0115] In one feasible implementation, step S30 may include: obtaining a preset business domain classification table and a preset business domain quantity threshold, wherein the preset business domain classification table contains a mapping relationship between business domains, corresponding product types, and business scenario keywords; determining product type information, business scenario keywords, and business context information based on the target business element; matching the product type information and the business scenario keywords with the content in the preset business domain classification table to obtain business domain matching results, and determining the number of successfully matched business domains based on the business domain matching results; when the number of matched business domains is less than or equal to the preset business domain quantity threshold, taking the successfully matched business domains as target business domains; when the number of matched business domains is greater than the preset business domain quantity threshold, determining the target business domain from the business domain matching results based on the business context information.

[0116] The pre-defined business area classification table is a structured table pre-built based on the bank's financial business attributes, product characteristics, and service scenarios. Its core components include the mapping relationship between business areas, corresponding product types, and business scenario keywords. Business areas refer to the specific financial business categories defined within the bank, such as bill business, cross-border business, payroll services, cash pooling, and settlement business. Corresponding product types are the specific financial product categories covered under each business area; for example, the bill business area includes bank acceptance bills and commercial acceptance bills, while the cross-border business area includes cross-border wire transfers and letters of credit. Business scenario keywords are terms that identify typical service scenarios within that business area; for example, the bill business area includes bill discounting, bill acceptance, and bill payment, while the cross-border business area includes cross-border declaration, foreign exchange settlement, and cross-border fund transfer. The purpose of this table is to provide a clear mapping basis for matching target business elements with business areas, ensuring that business area matching has a unified and standardized reference standard.

[0117] The preset threshold for the number of business domains is a critical value set in advance based on the complexity of internal business and the accuracy requirements of user questions. It is used to determine whether the number of successfully matched business domains needs further filtering. For example, if it is set to 1 or 2, if the number of successfully matched business domains does not exceed this threshold, it means that the matching results are relatively concentrated, and the target business domain can be directly determined; if it exceeds this threshold, it means that there are multiple candidate domains in the matching results, and further filtering is required based on additional information. Its purpose is to provide a quantitative judgment standard for processing business domain matching results, avoiding deviations in the determination of the target business domain due to too many candidate domains.

[0118] Product type information is extracted from target business elements and identifies the category to which a financial product belongs. It is one of the core bases for linking business areas. For example, if the target business element is the bank acceptance bill discounting process, the extracted product type information would be bank acceptance bill; if the target business element is cross-border wire transfer application materials, the extracted product type information would be cross-border wire transfer. This information needs to match the corresponding product type field in the preset business area classification table to provide product-dimensional support for business area matching.

[0119] Business scenario keywords are words extracted from target business elements that reflect the business service scenario and are used to help identify the business domain. For example, the business scenario keyword extracted from the target business element "bill acceptance business processing time" is "bill acceptance"; the business scenario keyword extracted from the target business element "cross-border fund transfer limit query" is "cross-border fund transfer". Their function is to match them with the business scenario keyword field in the preset business domain classification table, supplementing and verifying the matching results of the business domain from the scenario dimension, and improving the matching accuracy.

[0120] Business context information is extracted from the overall semantics of the financial questions raised by users, providing supplementary information about the business scenario. This includes user identity (e.g., corporate user, individual user), business processing stage (e.g., business consultation, document preparation, process handling), and related business needs (e.g., fee inquiries, limit inquiries, anomaly handling). For example, if a user's question is about the process for corporate users to discount bank acceptance bills, the extracted business context information would include: user identity: corporate user; business processing stage: process consultation. Similarly, if a user's question is about the limit for cross-border wire transfers for individual users, the extracted business context information would include: user identity: individual user; related business needs: limit inquiries. Its purpose is to provide additional filtering criteria when there are too many matching business areas, ensuring a high degree of alignment between the target business area and the user's actual business scenario.

[0121] The business domain matching result is obtained by comparing product type information and business scenario keywords with the corresponding product type and business scenario keywords in the preset business domain classification table. The result set includes successfully matched business domains and unmatched business domains. For example, if the product type information "bank acceptance bill" matches the corresponding product type in the "bill business" field of the preset business domain classification table, and the business scenario keyword "bill discounting" matches the business scenario keyword in the "bill business" field, then the "bill business" field in the business domain matching result will be a successful match, while other fields will be unmatched. Similarly, if the product type information "cross-border wire transfer" partially matches the corresponding product types in both the "cross-border business" and "settlement business" fields, and the business scenario keyword "foreign exchange settlement" matches the business scenario keyword in the "cross-border business" field, then the "cross-border business" field in the business domain matching result will be a successful match, while the "settlement business" field will be unmatched.

[0122] The number of matched business domains is the total number of business domains that successfully matched in the business domain matching results. For example, if the business domains that successfully matched in the business domain matching results are the bill business domain and the cross-border business domain, then the number of matched business domains is 2; if only the XinFuTong business domain matched successfully, the number of matched business domains is 1. This number is the core basis for determining whether it is necessary to further filter the target business domains by combining business context information.

[0123] The target business area is the specific business area that is ultimately determined and highly relevant to the user's financial questions. It is a crucial prerequisite for subsequent matching of target skill interfaces and retrieval of the financial Q&A knowledge base. For example, areas such as bill business and cross-border business, determined through matching, may become target business areas, and their accuracy directly affects the precision of subsequent Q&A services.

[0124] Understandably, this involves obtaining a preset business domain classification table and a preset business domain quantity threshold. The preset business domain classification table is retrieved from pre-stored business configuration resources within the bank, ensuring the table contains a complete mapping relationship between business domains, corresponding product types, and business scenario keywords. Simultaneously, the preset business domain quantity threshold is extracted from the configuration file, specifying its exact value. Based on the target business elements, product type information, business scenario keywords, and business context information are determined. Semantic analysis is performed on the target business elements, extracting content identifying product categories as product type information. Simultaneously, considering the overall semantics of the user's financial question, content such as user identity, business processing stage, and business-related needs is extracted as business context information. The product type information and business scenario keywords are then matched against the content in the preset business domain classification table to obtain business domain matching results and determine the number of matched business domains. The extracted product type information is compared one by one with the corresponding product types in each business area of ​​the preset business area classification table. If they are completely consistent or highly related, the area is determined to be successfully matched in the product type dimension. The business scenario keywords are compared one by one with the business scenario keywords of each business area. If there are identical or semantically similar keywords, the area is determined to be successfully matched in the scenario dimension. Combining the matching results of the two dimensions, if a business area is successfully matched in at least one dimension, it is included in the successfully matched category of the business area matching results; otherwise, it is included in the unsuccessfully matched category. The total number of business areas in the successfully matched category is counted to obtain the number of matched business areas.

[0125] The target business domain is determined based on the following scenarios. If the number of matching business domains is less than or equal to the preset threshold, it indicates that there are few candidate business domains and the matching results are concentrated. In this case, all successfully matched business domains in the matching results are directly used as the target business domain. If only one match is successful, it is directly determined as the unique target business domain. If multiple matches are successful and do not exceed the threshold, they are all used as target business domains, and further refined based on other information. If the number of matching business domains is greater than the preset threshold, it indicates that there are too many candidate business domains. It is necessary to filter them based on the previously extracted business context information. For example, if the business context information is user identity: enterprise user, then business domains targeting enterprise users are prioritized as target business domains. If the business context information is business-related requirement: fund settlement, then business domains containing fund settlement scenarios are prioritized as target business domains. Finally, a unique or a set of precisely matched target business domains are determined.

[0126] Step S40: Determine the target skill interface based on the questioning intent, the target business element, and the target business domain;

[0127] It should be noted that the target skill interface is selected from multiple skill interfaces that match the question intent, target business elements, and target business domain. It serves as the crucial link for retrieving the target answer from the financial Q&A knowledge base, and its determination must follow a pre-defined invocation strategy. The accuracy of the target skill interface directly determines the accuracy of subsequent knowledge retrieval and answer generation.

[0128] Understandably, based on the identified target business domain, all candidate skill interfaces corresponding to that business domain are selected. For example, when the target business domain is the bill business domain, the candidate skill interfaces are all the skill interfaces related to bill business configured within the bank. Then, combined with the previously obtained question intent and target business elements, the candidate skill interfaces are further filtered. Specifically, the calling strategy and processing capabilities of each candidate skill interface are analyzed to determine whether it can handle the requirements corresponding to the question intent and whether it can match the key information in the target business elements. If the configuration information of a candidate skill interface can fully meet the processing requirements of the question intent and the target business elements, then the candidate skill interface is determined as the target skill interface.

[0129] Step S50: Determine the target answer in the financial question-and-answer knowledge base according to the target skill interface.

[0130] Understandably, the target answer is content obtained from a financial Q&A knowledge base that accurately answers the user's financial questions. The target answer must be generated based on knowledge resources accessed via the target skill interface, and its content must meet the professional and accuracy requirements of the financial field. Its sources fall into two categories: one is standard answers directly obtained from QA pairs in the financial Q&A knowledge base; the other is answers generated by summarizing and generating information from document content in the knowledge base using large models (such as FinGPT, Financial Generative Pre-trained Transformer). The target answer must accurately match the user's question intent and target business elements, ensuring that the user can clearly understand and obtain the required financial information.

[0131] Furthermore, the identified target skill interface is activated. Authentication is performed based on the interface's configuration information (such as service code, ragToken authentication information, and access address). Upon successful authentication, access to the financial Q&A knowledge base is granted. Next, the target skill interface filters relevant knowledge content from the financial Q&A knowledge base based on the previously obtained question intent and target business elements. If knowledge content directly matches the question intent and target business elements (such as a corresponding QA pair), it is used as the initial answer. If no directly matching content is found, or if the knowledge in the document needs to be summarized to form a more concise answer, a large model (such as FinGPT) is invoked to process the relevant document content in the knowledge base, generating an answer that conforms to the question intent and target business elements. Finally, the initial answer or the answer generated by the large model is validated to ensure that the answer is accurate, professional, unambiguous, and fully answers the user's question. The validated answer is then determined as the target answer.

[0132] Reference Figure 4 , Figure 4 This is a schematic diagram of the overall question-answering architecture of the first embodiment of the financial question answer generation method based on question intent in this application. Figure 4 As shown, the first step is to input user questions. These questions can come from different assistants, such as the Yidian Tong mobile app, CRM smart assistant, Taifutong smart assistant, domestic certificate assistant, bill operation assistant, fee collection accounting assistant, cross-border declaration assistant, and mobile phone payment machine. The second step is the internal KYC agent (Knowledge, Technology, and Accounts) control center, which involves scheduling strategies and user permission handling. Scheduling strategies include single-skill library scheduling based on intent maps, single-skill library scheduling for supported channels, and multi-skill library scheduling for supported channels. User permission handling involves KYC questions containing sensitive information; the control center will authenticate based on the branch to which the employee belongs. Next is the fusion engine (multi-turn conversation), which migrates and integrates with the internal KYC agent (Knowledge, Technology, and Accounts) control center to handle user questions. The third step involves sending user questions to various knowledge bases and tool libraries for processing. These include the RAG Center's corporate product knowledge base, the Help Center knowledge base, the Yuanjingfu corporate knowledge base, the Xinfutong Xiaozhao Cloud knowledge base, the Xinfutong RAG knowledge base, the bill business knowledge base, the revenue accounting business knowledge base, the declaration business document knowledge base, the AI ​​knowledge base, the operations and maintenance knowledge base, and FinGPT's domestic certificate FinGPT. Additionally, it includes the Yidiantong marketing tool library and Yunce, among others. The diagram also shows different processes and information return paths for KYC, OTHER, Xinfutong channels, designated channel Q&A, domestic certificate channels, FUNCTION, and migration. Finally, the TW Agent (intent recognition) and the AI ​​Smart Factory (intent recognition) participate in the identification and processing of user questions.

[0133] In one feasible implementation, step S50 may include: performing a search in the financial Q&A knowledge according to the target skill interface to obtain search results; determining whether the search results are valid answers, wherein the criteria for a valid answer are containing detailed business information, unambiguous answer information, and correctly matching the question intent; when the search results are valid answers, formatting the valid answers to obtain a target answer; when the search results are empty or not valid answers, obtaining a preset basic terminology set and determining whether the financial question matches the keywords in the preset basic terminology set; when the financial question matches the keywords in the preset basic terminology set, determining the target answer based on the keywords.

[0134] Financial Q&A knowledge is a collection of professional content stored in the financial Q&A knowledge base to answer users' financial questions. It covers product information, processing procedures, required materials, business rules, and frequently asked questions for various financial services within the bank. This knowledge exists in the form of documents or QA pairs / question-answer pairs. Examples include documents on the processing procedures for bank acceptance drafts in the corporate banking product knowledge base, QA pairs on required materials for cross-border declarations in the cross-border business QA database, and documentation explaining the conditions for activating payroll services in the payroll service knowledge base. Its purpose is to provide content sources for target skill interface retrieval, ensuring that professional information matching user questions can be obtained.

[0135] The search results are a collection of information obtained by the target skill interface after querying financial Q&A knowledge based on the question intent and target business elements. Search results may include directly matched QA answers, fragments of content from documents, or no relevant information found. For example, if the target skill interface is the bill business interface, and the question intent is to query the bank acceptance bill discounting process, the search results might be QA answers related to the bank acceptance bill discounting process in the bill business knowledge base; if no relevant process information is found, the search results will be empty. Its function is to serve as the basis for subsequently determining valid answers and generating target answers.

[0136] An effective answer is a search result that meets the preset judgment criteria, which include three core dimensions. The first dimension is to contain detailed business information, that is, the answer should cover the key elements of business handling. For example, the handling process should include the order of steps, the departments or systems involved, the handling time limit, and the required materials should list the specific document names and requirements. The second dimension is that the answer information is unambiguous, that is, the expression is clear and there is no ambiguous expression. For example, the handling time is 3 working days instead of the handling time is relatively fast, and a copy of the business license needs to be provided instead of relevant enterprise documents need to be provided. The third dimension is to correctly match the question intention, that is, the content of the answer is consistent with the user's core needs. For example, if the user's intention is to query the cross-border wire transfer fee, the effective answer should focus on the cross-border wire transfer fee standard rather than the cross-border wire transfer handling process. Its role is to screen out high-quality search results that can truly answer the user's questions and avoid the output of invalid information.

[0137] Formatting is a process of standardizing the presentation form of effective answers, aiming to make the answers clearer and easier to read and conform to the user's reading habits. The content to be sorted includes adjusting the text layout, such as listing the process steps, material lists in points, and supplementing necessary business identifiers, such as marking the first step and the second step before the process steps, and unifying the professional term expressions, such as unifying "bill discounting" as "note discounting".

[0138] The preset basic word set is a pre-organized set of keywords for basic business in the financial field. These keywords correspond to high-frequency basic questions or general business concepts in the industry, such as account opening, transfer, loss reporting, balance query, password modification, business handling time, branch address, etc. Each keyword is associated with standardized basic answer content. For example, for account opening, when opening an enterprise account, you need to bring materials such as the original copy of the business license, the original ID card of the legal person, and the articles of association, and go to the corporate business outlet of our bank for handling. The handling time is about 2 working days; the business handling time is associated with the corporate business handling time of our bank from 9:00 to 17:00 on working days. Some branches can handle simple corporate business on weekends. Please refer to the branch notice for details. Its role is to provide a fallback answer basis for basic financial questions when the search results are invalid or empty, ensuring that there is no situation of no answer.

[0139] Keywords are specific words in the preset basic word set. Each keyword corresponds to a type of basic financial question and a standardized answer. For example, loss reporting corresponds to questions related to bank card or bill loss reporting, and balance query corresponds to questions related to account balance query. These keywords can directly relate to the user's basic needs. Their role is to quickly locate the corresponding basic answer and generate the target answer by matching the words in the user's financial questions.

[0140] Understandably, the search results are obtained by retrieving financial Q&A knowledge based on the target skill interface. The target skill interface carries the question intent and target business elements, and uses preset search algorithms, such as keyword matching and semantic similarity matching algorithms, to search for relevant content in the corresponding financial Q&A knowledge sections, such as bill business knowledge and cross-border business knowledge. The search results are then integrated with the QA answers or document fragments found. If no relevant content is found, the search results are empty.

[0141] To determine whether a search result is a valid answer, the criteria are: containing detailed business information, providing unambiguous information, and correctly matching the query intent. The search results are reviewed one by one: First, determine if it contains detailed business information. For example, if the search result is about the bill discounting process, it needs to confirm whether it includes key information such as process steps and the relevant department. Second, determine if the answer is unambiguous, checking for vague statements and ensuring that each piece of information is clear and unambiguous. Finally, determine if it correctly matches the query intent. For example, if the query intent is to query cross-border declaration materials, it needs to confirm that the search results revolve around cross-border declaration materials and not other irrelevant content. If all three criteria are met, the search result is considered a valid answer; if any criterion is not met, the search result is considered an invalid answer.

[0142] The search results are processed according to different scenarios. If the search result is a valid answer, it is formatted to obtain the target answer. Key information in the valid answer is broken down, numbered, or supplemented with business identifiers according to preset format specifications. If the search result is empty or not a valid answer, a preset basic terminology set is retrieved, and it is determined whether the financial question matches the keywords in the preset basic terminology set. The preset basic terminology set is retrieved from pre-configured resources within the bank, and the words in the financial question are compared one by one with the keywords in the basic terminology set. For example, if the financial question is "How to open a corporate account," the comparison is made to see if it contains the keyword "account opening"; if the financial question is "How to handle a lost bank card," the comparison is made to see if it contains the keyword "reporting loss." If there are words in the financial question that are consistent with or semantically similar to the keywords in the basic terminology set, the two are considered a match; otherwise, they are considered a mismatch. When the financial question matches the keywords in the preset basic terminology set, the target answer is determined based on the keywords. After finding a keyword that matches the financial question, the standardized basic answer associated with that keyword is retrieved and directly used as the target answer.

[0143] This embodiment provides a method for generating answers to financial questions based on question intent. By constructing a preset financial vocabulary set, a financial question-and-answer knowledge base, and using large models for intent recognition and business element extraction, it solves the technical problems in financial question-and-answer systems, such as errors in the segmentation of professional terms caused by general vocabulary segmentation and inaccurate matching between user questions and knowledge base content. This achieves the beneficial effects of improving the accuracy and professionalism of financial question answers and enhancing user experience.

[0144] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 Step S40 of the financial question answer generation method based on question intent further includes steps S41 to S43:

[0145] Step S41: Obtain the channel code of the current user and determine the corresponding channel orchestration configuration table based on the channel code;

[0146] It's important to note that the channel code is a unique identifier used to determine the specific channel a user accesses the Q&A service. These channels encompass various user interaction entry points within the industry. Additionally, the channel orchestration configuration table is a pre-built structured table that records the orchestration rules for skill interfaces for different access channels. The table contains a mapping relationship between channel codes and corresponding callable skill interfaces, skill interface call priority, and interface access permissions.

[0147] Step S42: Determine the candidate skill interface list in the channel orchestration configuration table according to the target business area and the question intent;

[0148] It should be noted that the candidate skill interface list is a collection of multiple skill interfaces selected from the channel orchestration configuration table that match both the target business domain and the question intent. The list includes basic information such as the name, service scope, and calling parameters of each skill interface. Its purpose is to provide a range of alternatives for further screening of target skill interfaces, ensuring that multiple suitable interfaces are available for selection and optimization.

[0149] Understandably, the initial screening should begin based on the target business domain. This screening involves comparing the service business domain of each skill interface in the configuration table with the target business domain. For example, if the target business domain is cross-border business, all skill interfaces whose service scope includes cross-border business are selected from the configuration table, excluding interfaces from other domains such as invoice processing and salary payment processing, thus narrowing the interface selection scope. From the business domain matching interfaces obtained in the initial screening, further filtering is performed based on the question intent. The service function of each interface is compared to the question intent. All skill interfaces that simultaneously meet both the target business domain and question intent are integrated to form a candidate skill interface list, preparing for subsequent relevance-based screening.

[0150] Step S43: Perform relevance filtering on the candidate skill interface list based on the target business elements to obtain the relevance filtering results, and determine the target skill interface based on the relevance filtering results.

[0151] Understandably, relevance screening is the process of determining the degree of relevance between each interface in the candidate skill interface list and the user's question based on the target business elements, and eliminating interfaces with low relevance. Screening dimensions include whether the interface supports the product type in the target business elements, whether it can handle the business type in the elements, and whether it covers the detailed requirements of the elements. Its purpose is to further optimize the candidate skill interface list, ensuring that the interfaces in the list highly match the business details of the user's question. The result of relevance screening is a set of skill interfaces that are highly relevant to the target business elements. The target skill interface is the skill interface that best matches the user's financial question, determined from the relevance screening results.

[0152] Understandably, for the obtained candidate skill interface list, the service parameters and service scope of each candidate interface are extracted one by one, such as the product types supported by the interface, the business types it can handle, and the detailed requirements it can respond to. The service parameters and service scope of each candidate interface are compared with the target business elements one by one to determine their relevance: if the interface supports the standard product name in the target business element, can handle the business type in the element, and covers the detailed requirements in the element, then the interface is considered highly relevant to the user's problem; if the interface only partially matches the target business element, such as supporting product types but not covering the detailed requirements, then the relevance is considered moderate; if the interface does not match the core information in the target business element at all, then the relevance is considered low. Based on the relevance determination results, interfaces with high relevance are retained, while those with moderate and low relevance are removed, forming a relevance filtering result. The target skill interface is determined based on the relevance filtering result: if the filtering result contains only one interface, that interface is directly determined as the target skill interface; if the filtering result contains multiple interfaces, the call priority of each interface in the channel orchestration configuration table is referenced, and the interface with the highest priority is selected as the target skill interface; if the priorities are the same, the interface with the faster response time is selected based on historical service response efficiency.

[0153] In one feasible implementation, after step S41, the method may further include: when no corresponding channel orchestration configuration table is found based on the channel code, obtaining a public database and using the general skill interface corresponding to the public database as a candidate skill interface; when no filtering requirement is detected in the candidate skill interface, obtaining a backup skill interface, wherein the backup skill interface includes basic question-and-answer interfaces for each business domain; when no filtering requirement is detected in the backup skill interface, obtaining a basic skill interface and using the basic skill interface as a target skill interface; when the filtering requirement is detected in the backup skill interface, using the backup skill interface as the target skill interface.

[0154] The corporate banking database is a dedicated resource repository within the bank for storing knowledge, data, and interface configuration information related to corporate banking services. This database covers core content across the entire corporate banking business scope, including corporate product information, corporate business processing rules, corporate customer service data, and general skill interface configurations adapted for corporate banking services.

[0155] General skill interfaces are pre-defined skill interfaces in the corporate database that are adaptable to most corporate business scenarios, not limited to a specific channel or niche business scenario. These interfaces have the capability to handle basic business needs in the corporate sector. For example, a general corporate product query interface can query basic information about most corporate products, a general corporate business process consultation interface can answer common corporate business processing procedures, and a general corporate document query interface can provide a list of basic documents required for most corporate businesses. Their purpose is to serve as a basic interface resource to meet users' general corporate business Q&A needs when there are no channel-specific skill interface configurations, avoiding service interruptions due to missing channel configurations.

[0156] The backup skill interfaces are a pre-configured set of skills interfaces covering basic Q&A functions across various business areas. These include basic Q&A interfaces for bill of exchange business, cross-border business, salary and benefits payment, cash pooling, and settlement. Each business area's basic Q&A interface can handle the fundamental business needs within that area. For example, the bill of exchange business basic Q&A interface can answer the basic processes of bill issuance, discounting, and redemption; the cross-border business basic Q&A interface can answer the basic rules of cross-border declaration and foreign exchange settlement. Their purpose is to provide more granular business area basic interface support when general skill interfaces cannot meet filtering needs, further expanding the interface coverage and reducing situations where needs cannot be matched.

[0157] The basic skills interface is a pre-built interface within the industry, possessing general basic question-and-answer capabilities applicable to the financial field, and is not limited to any specific business area. This interface stores the most basic and universal business knowledge in the financial field, such as general processing times for financial transactions, basic service rules within the industry, and explanations of common financial terms. It can handle basic financial questions raised by users. Its role is to serve as a last resort interface, ensuring that basic question-and-answer services can still be provided when the general skills interface and backup skills interface cannot meet the screening requirements, thus avoiding a situation where no interface is available.

[0158] Understandably, when querying the corresponding channel orchestration configuration table based on the channel code, if the query result is empty, a fallback mechanism is activated to retrieve the corporate database from the pre-stored business resource library within the bank. After retrieving the corporate database, the pre-defined general skill interfaces in the database are extracted. Since these general skill interfaces have the ability to adapt to most corporate business scenarios, they are directly used as candidate skill interfaces to provide basic interface resources for subsequent selection.

[0159] The process involves determining whether candidate skill interfaces include filtering requirements. Specifically, the service scope and functionality of general skill interfaces are compared one by one with the filtering requirements determined based on the target business domain, question intent, and target business elements to check if a general skill interface that meets the filtering requirements exists. If no candidate skill interface contains filtering requirements, backup skill interfaces are obtained from the internal interface resource library. These backup skill interfaces cover basic Q&A interfaces for various business domains such as bills, cross-border transactions, and salary payment services, providing more granular basic business support. For the obtained backup skill interfaces, the process is again determined whether they contain filtering requirements. Similarly, by comparing the service functions of the backup skill interfaces with the filtering requirements, it is checked whether a suitable interface exists. If no backup skill interface contains filtering requirements, a basic skill interface within the bank is obtained as a last resort and directly designated as the target skill interface to ensure that basic Q&A services can be provided. If a backup skill interface contains filtering requirements, there is no need to obtain other interfaces; the backup skill interface containing filtering requirements is directly designated as the target skill interface.

[0160] This embodiment provides a method for generating answers to financial questions based on question intent. By introducing channel codes and channel orchestration configuration tables, and combining multi-level filtering techniques based on target business areas, question intent, and target business elements, it solves the technical problems of poor skill interface adaptability and inaccurate matching between user questions and skill interfaces caused by different access channels in financial Q&A services. This achieves the beneficial effects of improving the accuracy and adaptability of financial question answers and optimizing user experience.

[0161] For example, to help understand the implementation process of the financial question answer generation method based on question intent obtained by combining this embodiment with the above embodiment one, please refer to... Figure 6 , Figure 6 A simplified flowchart illustrating a method for generating answers to financial questions based on question intent is provided, specifically:

[0162] The process is mainly divided into five parts: input / output interfaces, interface authentication, session information generation, multi-round rewriting, and large-scale model orchestration. The input / output interfaces include two URLs. The interface authentication part mentions clientld authentication, obtaining the current user information and storing it in memory. The session information generation part involves obtaining Session information, looking up the table based on the Sessionid; if the Session information does not exist, a new session is created; if it exists, it continues to be used; simultaneously, SessionLog information is obtained, a new session log is generated, inserted, and returned. The large-scale model orchestration part is the core of the entire process. It mentions querying the channel orchestration configuration table based on the channel code; if not found, it checks the public database; configuring the skill call interfaces as follows, filled in according to priority. Based on the skill call interface list, the skill interface call configuration is queried; non-existent skill interfaces are skipped. The skill interface list is traversed to obtain the call strategy based on the skill interface call configuration. If no strategy is matched, it is skipped; if a currently supported strategy is matched, rag includes configuring the service code, obtaining the ragToken for authentication based on the service code + openld authentication, and configuring the AI ​​smart factory, involving access address, key, and input / output mapping. The method has unified input and output, returning streaming or non-streaming output based on the input streaming / non-streaming flags. Input parameters include sessionId (session ID), sessionLogId (session log ID), responseMode (streaming or non-streaming), question, systemId (system ID), and necessary user-related information, which is obtained through cached event-by-event information. Output parameters include sessionId (session ID), content (response content), streamItemTypeANSWER (document reference), END (streaming end), aiAnswerTypeEnumSUMMARIZED (valid answer for the main model), CREATED (empty retrieval, fallback answer), LLM_DEFAULT (fallback answer for the main model), streamAnswerStatusEnumAPPEND (text appending mode), REPLACE (text replacement mode), specialParams (personalized response, such as document references with inconsistent formats between main models, placed here for upstream free parsing), and relatedContent (document references in common mode). In exception handling, when accessing downstream fails, the response is encapsulated as a fallback answer. aiAnswerTypeEnumLLM_DEFAULT - Large model fallback answer, no error thrown. An additional frame is appended to the doc frame and placed at the end, and the final response is returned to the upstream for parsing and personalized response. Fallback judgment: Configure a fallback term set; a match is considered a fallback. In cases where the rag might contain invalid frames in the first few frames, the final fallback frame will not be returned by default if it matches an invalid frame with the fallback term.The encapsulation intelligently determines whether it is a fallback answer. Deep Thinking skill scenarios are not currently supported; in scenarios using Deep Thinking, it must be placed as the lowest priority skill. Aspects uniformly print input and output logs. Traversing the skill interface list can also call strategies to access skill interfaces, parsing the first frame's answer, categorizing it as a valid or invalid answer. For valid answers (aiAnswerTypeEnumSUMMARIZED - valid answer for the large model), a response is assembled and output to the upstream, using a card template, and the SessionLog information is updated in the last frame to save the answer. For invalid answers (aiAnswerTypeEnumCREATED - empty search, fallback answer LLM_DEFAULT - fallback answer for the large model), if it is not the last strategy, the next calling strategy is used to continue the question-and-answer process; if it is the last strategy, the business fallback answer configured in the intelligent question-and-answer scenario is returned.

[0163] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the method for generating answers to financial questions based on the intent of the question. Any simple modifications based on this technical concept are within the scope of protection of this application.

[0164] This application also provides a device for generating answers to financial questions based on the intent of the question; please refer to [reference needed]. Figure 7 The financial question answer generation device based on question intent includes:

[0165] Data acquisition module 10 is used to acquire a preset financial vocabulary set, a financial Q&A knowledge base, and financial questions raised by users;

[0166] The intent recognition module 20 is used to perform intent recognition and business element extraction based on the financial question and the preset financial vocabulary set, so as to obtain the question intent and the target business element respectively.

[0167] Domain determination module 30 is used to determine the target business domain based on the target business elements;

[0168] The interface determination module 40 is used to determine the target skill interface based on the questioning intent, the target business element, and the target business domain.

[0169] The question-answering module 50 is used to determine the target answer in the financial question-answering knowledge base based on the target skill interface.

[0170] The financial question answer generation device based on question intent provided in this application, employing the financial question answer generation method based on question intent in the above embodiments, can solve the technical problem of difficulty in ensuring the accuracy of answers in financial question-and-answer scenarios due to the complexity of professional terminology. Compared with the prior art, the beneficial effects of the financial question answer generation device based on question intent provided in this application are the same as those of the financial question answer generation method based on question intent provided in the above embodiments, and other technical features in the financial question answer generation device based on question intent are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0171] In one embodiment, the intent recognition module 20 is further configured to: acquire a standard business element library, a preset similarity threshold, and a preset edit distance threshold; perform word segmentation on the financial question to obtain product slots, and perform word matching on the product slots according to the preset financial vocabulary set to obtain subdivided business elements; extract business elements according to the standard business element library, the preset similarity threshold, the preset edit distance threshold, and the subdivided business elements to obtain target business elements; and perform intent recognition based on the financial question and the target business elements to obtain the question intent.

[0172] In one embodiment, the intent recognition module 20 is further configured to calculate the business similarity between the segmented business element and the elements in the standard business element library; when the business similarity is greater than the preset similarity threshold, the standard business element corresponding to the business similarity is used as the target business element; when the business similarity is less than or equal to the preset similarity threshold, the edit distance between the pinyin of the segmented business element and the pinyin of the elements in the standard business element library is calculated; when the edit distance is less than the preset edit distance threshold, the standard business element corresponding to the edit distance is used as the target business element.

[0173] In one embodiment, the domain determination module 30 is further configured to obtain a preset business domain classification table and a preset business domain quantity threshold, wherein the preset business domain classification table contains a mapping relationship between business domains, corresponding product types, and business scenario keywords; determine product type information, business scenario keywords, and business context information based on the target business element; match the product type information and the business scenario keywords with the content in the preset business domain classification table respectively to obtain a business domain matching result, and determine the number of successfully matched business domains based on the business domain matching result; when the number of matched business domains is less than or equal to the preset business domain quantity threshold, the business domains with successfully matched business domains are taken as target business domains; when the number of matched business domains is greater than the preset business domain quantity threshold, the target business domain is determined from the business domain matching result based on the business context information.

[0174] In one embodiment, the interface determination module 40 is further configured to obtain the channel code of the current user and determine the corresponding channel orchestration configuration table based on the channel code; determine a candidate skill interface list in the channel orchestration configuration table based on the target business domain and the question intent; perform relevance filtering on the candidate skill interface list based on the target business elements to obtain relevance filtering results, and determine the target skill interface based on the relevance filtering results.

[0175] In one embodiment, the interface determination module 40 is further configured to: when no corresponding channel orchestration configuration table is found based on the channel code, obtain a public database and use the general skill interface corresponding to the public database as a candidate skill interface; when no filtering requirement is detected in the candidate skill interface, obtain a backup skill interface, wherein the backup skill interface includes basic question and answer interfaces for each business domain; when no filtering requirement is detected in the backup skill interface, obtain a basic skill interface and use the basic skill interface as a target skill interface; when the filtering requirement is detected in the backup skill interface, use the backup skill interface as the target skill interface.

[0176] In one embodiment, the question-answering module 50 is further configured to: search the financial Q&A knowledge according to the target skill interface to obtain search results; determine whether the search results are valid answers, wherein the criteria for a valid answer are that it contains detailed business information, the answer information is unambiguous, and it correctly matches the question intent; when the search results are valid answers, format the valid answers to obtain a target answer; when the search results are empty or not valid answers, obtain a preset basic word set and determine whether the financial question matches the keywords in the preset basic word set; when the financial question matches the keywords in the preset basic word set, determine the target answer based on the keywords.

[0177] This application provides a financial question answer generation device based on question intent. The financial question answer generation device based on question intent 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 to enable the at least one processor to execute the financial question answer generation method based on question intent in the above embodiment 1.

[0178] The following is for reference. Figure 8 This document illustrates a structural diagram suitable for implementing the financial question answer generation device based on question intent in the embodiments of this application. The financial question answer generation device based on question intent in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 8 The financial question answer generation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0179] like Figure 8As shown, the financial question answer generation device based on question intent may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the financial question answer generation device based on question intent. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the question-intention-based financial question answer generation device to exchange data with other devices wirelessly or via wired communication. Although the figure shows a question-intention-based financial question answer generation device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0180] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0181] The financial question answer generation device based on question intent provided in this application, employing the financial question answer generation method based on question intent in the above embodiments, can solve the technical problem of difficulty in ensuring the accuracy of answers in financial question-and-answer scenarios due to the complexity of professional terminology. Compared with the prior art, the beneficial effects of the financial question answer generation device based on question intent provided in this application are the same as those of the financial question answer generation method based on question intent provided in the above embodiments, and other technical features in this financial question answer generation device based on question intent are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0182] It should be understood that the various parts disclosed in this application can be implemented using 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 suitable manner in one or more embodiments or examples.

[0183] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0184] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon. These computer-readable program instructions are used to execute the financial question answer generation method based on question intent described in the above embodiments. The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), Erasable Programmable Read Only Memory (EPROM), optical fiber, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. 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 conjunction with an instruction execution system, system, or device. Program code contained on a computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0185] The aforementioned computer-readable storage medium may be included in a financial question answer generation device based on question intent; or it may exist independently and not assembled into the financial question answer generation device based on question intent. The aforementioned computer-readable storage medium carries one or more programs that, when executed by the financial question answer generation device based on question intent, cause the financial question answer generation device to: acquire a preset financial vocabulary set, a financial question-and-answer knowledge base, and a financial question raised by the user; perform intent recognition and business element extraction based on the financial question and the preset financial vocabulary set, respectively obtaining the question intent and the target business element; determine the target business domain based on the target business element; determine the target skill interface based on the question intent, the target business element, and the target business domain; and determine the target answer in the financial question-and-answer knowledge base based on the target skill interface.

[0186] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0187] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0188] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the accuracy of the unit itself. The readable storage medium provided in this application is a computer-readable storage medium storing computer-readable program instructions (i.e., a computer program) for executing the above-described method for generating answers to financial questions based on question intent. This solves the technical problem of difficulty in ensuring the accuracy of answers in financial question-and-answer scenarios due to the complexity of technical terminology. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the method for generating answers to financial questions based on question intent provided in the above embodiments, and will not be repeated here. This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for generating answers to financial questions based on question intent. The computer program product provided in this application solves the technical problem of difficulty in ensuring the accuracy of answers in financial question-and-answer scenarios due to the complexity of technical terminology. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the method for generating answers to financial questions based on question intent provided in the above embodiments, and will not be repeated here. The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for generating answers to financial questions based on question intent, characterized in that, The method includes: Acquire a pre-defined financial vocabulary set, a financial Q&A knowledge base, and financial questions submitted by users; Based on the financial question and the preset financial vocabulary set, intent recognition and business element extraction are performed to obtain the question intent and target business element, respectively. The target business area is determined based on the target business elements; The target skill interface is determined based on the questioning intent, the target business elements, and the target business domain. The target answer is determined in the financial Q&A knowledge base based on the target skill interface.

2. The method as described in claim 1, characterized in that, The step of performing intent recognition and business element extraction based on the financial question and the preset financial vocabulary set to obtain the question intent and target business elements respectively includes: Obtain a standard business element library, preset similarity thresholds, and preset edit distance thresholds; The financial issue is segmented into words to obtain product slots, and the product slots are matched with words according to the preset financial vocabulary set to obtain subdivided business elements. Based on the standard business element library, the preset similarity threshold, the preset edit distance threshold, and the subdivided business elements, business elements are extracted to obtain the target business elements; Based on the financial question and the target business element, the intent of the question is identified.

3. The method as described in claim 2, characterized in that, The step of extracting target business elements based on the standard business element library, the preset similarity threshold, the preset edit distance threshold, and the subdivided business elements includes: Calculate the business similarity between the segmented business element and the elements in the standard business element library; When the business similarity is greater than the preset similarity threshold, the standard business element corresponding to the business similarity is taken as the target business element. When the business similarity is less than or equal to the preset similarity threshold, the edit distance between the pinyin of the subdivided business element and the pinyin of the element in the standard business element library is calculated; If the edit distance is less than the preset edit distance threshold, the standard business element corresponding to the edit distance is taken as the target business element.

4. The method as described in claim 1, characterized in that, The step of determining the target business domain based on the target business elements includes: Obtain a preset business domain classification table and a preset business domain quantity threshold, wherein the preset business domain classification table contains a mapping relationship between business domains, corresponding product types, and business scenario keywords; Based on the target business elements, determine product type information, business scenario keywords, and business context information; The product type information and the business scenario keywords are matched with the contents of the preset business domain classification table to obtain the business domain matching results, and the number of successfully matched business domains is determined. When the number of matched business domains is less than or equal to the preset threshold for the number of business domains, the business domains whose matching results are successful are taken as the target business domains; When the number of matched business domains is greater than the preset threshold for the number of business domains, the target business domain is determined from the business domain matching results based on the business context information.

5. The method as described in claim 1, characterized in that, The step of determining the target skill interface based on the question intent, the target business element, and the target business domain includes: Obtain the current user's channel code and determine the corresponding channel orchestration configuration table based on the channel code; Based on the target business area and the question intent, a list of candidate skill interfaces is determined in the channel orchestration configuration table; The candidate skill interface list is filtered for relevance based on the target business elements to obtain the relevance filtering results, and the target skill interface is determined based on the relevance filtering results.

6. The method as described in claim 5, characterized in that, After the steps of obtaining the current user's channel code and determining the corresponding channel orchestration configuration table based on the channel code, the method further includes: When no corresponding channel orchestration configuration table is found based on the channel code, the public database is obtained, and the general skill interface corresponding to the public database is used as a candidate skill interface. If no filtering requirement is detected in the candidate skill interface, an alternative skill interface is obtained, wherein the alternative skill interface includes basic question and answer interfaces for each business domain. If the filtering requirement is not detected in the backup skill interface, the basic skill interface is obtained and the basic skill interface is used as the target skill interface. When the filtering requirement is detected to be included in the backup skill interface, the backup skill interface is used as the target skill interface.

7. The method as described in claim 1, characterized in that, The step of determining the target answer in the financial question-and-answer knowledge base based on the target skill interface includes: The search results are obtained by performing a search on the financial Q&A knowledge based on the target skill interface. The criteria for determining whether the search results are valid answers are: containing detailed business information, having unambiguous answer information, and correctly matching the question intent. When the search result is a valid answer, the valid answer is formatted to obtain the target answer; When the search results are empty or not a valid answer, a preset basic word set is obtained, and it is determined whether the financial question matches the keywords in the preset basic word set. When the financial question is matched with keywords in the preset basic word set, the target answer is determined based on the keywords.

8. A device for generating answers to financial questions based on question intent, characterized in that, The device includes: The data acquisition module is used to acquire a preset financial vocabulary set, a financial Q&A knowledge base, and financial questions raised by users. The intent recognition module is used to perform intent recognition and business element extraction based on the financial question and the preset financial vocabulary set, so as to obtain the question intent and the target business element respectively. The domain determination module is used to determine the target business domain based on the target business elements. The interface determination module is used to determine the target skill interface based on the question intent, the target business element, and the target business domain. The question-answering module is used to determine the target answer in the financial question-answering knowledge base based on the target skill interface.

9. A device for generating answers to financial questions based on question intent, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the financial question answer generation method based on question intent as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium 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, it implements the steps of the financial question answer generation method based on question intent as described in any one of claims 1 to 7.

Images