Index table retrieval and question answering method and system based on large model identification

Abstract

The application discloses an index table retrieval and question and answer method and system based on large model identification, wherein the method comprises the following steps: obtaining a to-be-processed text containing a query intention; performing date identification on the to-be-processed text by using a large model to obtain target date information; performing index identification on the to-be-processed text by using the large model to obtain target index information; performing data retrieval in a preset index data table according to the target date information and the target index information to obtain target business data; and performing text generation according to the to-be-processed text and the target business data to obtain reply information corresponding to the query intention.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to an indicator table retrieval and question answering method based on large model recognition, an indicator table retrieval and question answering system based on large model recognition, an electronic device, and a readable storage medium. Background Technology

[0002] As the digitalization of banking services deepens, a large amount of structured indicator data has been generated in retail business scenarios, distributed in the form of data tables across various business systems. When making business decisions or conducting data analysis, it is often necessary to obtain multiple business indicators over a specific time dimension.

[0003] Traditional automated query methods primarily rely on keyword matching or predefined logical templates. When processing natural language commands, they struggle to accurately map query intent to data table dimensions. This is especially true when faced with queries containing diverse time expressions and industry terminology, often failing to pinpoint the target business item within the data table. Furthermore, existing query responses are mostly presented as single numerical values ​​or raw table fragments, lacking descriptive text that matches the query context. This results in cumbersome data understanding and limited interaction. Summary of the Invention

[0004] The purpose of this application is to provide a method for index table retrieval and question answering based on large model recognition, a system for index table retrieval and question answering based on large model recognition, an electronic device, and a readable storage medium. These methods can solve the problems of existing technologies in handling unstructured query intents, such as difficulty in achieving accurate semantic mapping between date and index dimensions, and low retrieval accuracy, lack of factual basis and language organization ability in scenarios with missing data or semantic ambiguity.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide a method for index table retrieval and question answering based on large model identification, the method comprising: Retrieve the text to be processed containing the query intent; The target date information is obtained by using a large model to identify the date in the text to be processed. The large model is used to identify indicators in the text to be processed to obtain target indicator information. Based on the target date information and the target indicator information, data is retrieved from the preset indicator data table to obtain the target business data. Text is generated based on the text to be processed and the target business data to obtain response information corresponding to the query intent.

[0006] Optionally, the step of using a large model to perform date recognition on the text to be processed to obtain target date information includes: The large model is used to extract temporal semantic elements from the text to be processed; If the time dimension is missing or relative time representation is included in the time semantic elements, obtain the reference time corresponding to the text to be processed. Based on the reference time, the time semantic elements are augmented with dimensions or offsets are calculated to obtain the target date information that conforms to a preset format.

[0007] Optionally, the step of using the large model to identify indicators in the text to be processed to obtain target indicator information includes: The text to be processed is vectorized with candidate indicators in a preset indicator library, and several preliminary indicators are determined from the candidate indicators based on vector similarity. The large model is used to determine whether the initially selected indicators match the query intent, thus obtaining the intent indicators. The target indicator information is determined based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library.

[0008] Optionally, determining the target indicator information based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library includes: Identify whether there are any preset related indicators among the intent indicators; If the existence of the associated indicators is confirmed, corresponding supplementary indicators are added to the intent indicators according to preset association rules. The intention indicator, after adding the supplementary indicator, is mapped to the corresponding original indicator to obtain the target indicator information.

[0009] Optionally, the step of retrieving data from a preset indicator data table based on the target date information and the target indicator information to obtain the target business data includes: Determine whether the original data corresponding to the target date information exists in the preset indicator data table; If it is determined that the original data does not exist, alternative date information is determined based on the time sequence position of the target date information relative to the current system time; The target business data is obtained by retrieving the data corresponding to the alternative date information and the target indicator information from the preset indicator data table.

[0010] Optionally, determining the alternative date information based on the time sequence position of the target date information relative to the current system time includes: If the month to which the target date information belongs is the same as the month to which the current system time belongs, the date of the most recent existing data in the preset indicator data table that is earlier than the target date information is determined as the alternative date information; If the month to which the target date information belongs is earlier than the month to which the current system time belongs, the last day of the month to which the target date information belongs, or the date with the most recently existing data in the month to which the target date information belongs, shall be determined as the alternative date information.

[0011] Optionally, the step of generating text based on the text to be processed and the target business data to obtain response information corresponding to the query intent includes: The text to be processed and the target business data are input into the large model to obtain the natural language description text generated by the large model that matches the query intent; The natural language description text is associated with the target business data used as a reference and displayed to obtain the response information.

[0012] Secondly, embodiments of this application provide an indicator table retrieval and question-answering system based on large model recognition, the system comprising: The text acquisition module is used to acquire the text to be processed, which contains the query intent; The date recognition module is used to perform date recognition on the text to be processed using a large model to obtain the target date information; The indicator recognition module is used to use the large model to recognize indicators in the text to be processed and obtain target indicator information. The data retrieval module is used to retrieve data from a preset indicator data table based on the target date information and the target indicator information to obtain the target business data. The response determination module is used to generate text based on the text to be processed and the target business data, and obtain response information corresponding to the query intent.

[0013] Optionally, the date recognition module includes: The temporal semantic element extraction module is used to extract temporal semantic elements from the text to be processed using the large model. The reference time acquisition module is used to acquire the reference time corresponding to the text to be processed when the time semantic element is missing a time dimension or contains a relative time expression. The target date information determination module is used to perform dimensional completion or offset calculation on the time semantic elements according to the reference time to obtain the target date information that conforms to a preset format.

[0014] Optionally, the indicator recognition module includes: The initial selection indicator determination module is used to vectorize the text to be processed and the candidate indicators in the preset indicator library, and determine a number of initial selection indicators from the candidate indicators based on the vector similarity. The intent indicator determination module is used to determine whether the initial selected indicators match the query intent using the large model, and to obtain the intent indicator. The target indicator information determination module is used to determine the target indicator information based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library.

[0015] Optionally, the target indicator information determination module includes: The associated indicator identification module is used to identify whether there are preset associated indicators among the intent indicators; The supplementary indicator addition module is used to add corresponding supplementary indicators to the intent indicator according to preset association rules when it is determined that the associated indicator exists. The intent indicator mapping module is used to map the intent indicator after adding the supplementary indicator to the corresponding original indicator to obtain the target indicator information.

[0016] Optionally, the data retrieval module includes: The raw data judgment module is used to determine whether the raw data corresponding to the target date information exists in the preset indicator data table; The alternative date information determination module is used to determine alternative date information based on the time sequence position of the target date information relative to the current system time when it is determined that the original data does not exist; The target business data determination module is used to retrieve data corresponding to the alternative date information and the target indicator information from the preset indicator data table to obtain the target business data.

[0017] Optionally, the alternative date information determination module is used to determine the date of the most recently existing data in the preset indicator data table that is earlier than the target date information as the alternative date information when the month to which the target date information belongs is the same as the month to which the current system time belongs; and to determine the last day of the month to which the target date information belongs, or the date of the most recently existing data in the month to which the target date information belongs, as the alternative date information when the month to which the target date information belongs is earlier than the month to which the current system time belongs.

[0018] Optionally, the response determination module includes: The natural language description text acquisition module is used to input the text to be processed and the target business data into the large model, and obtain the natural language description text generated by the large model that matches the query intent; The response information determination module is used to associate and display the natural language description text with the target business data used as a reference to obtain the response information.

[0019] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0020] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0021] In this embodiment, a large model is used to identify the date and indicator dimensions of the text to be processed, directly mapping the query intent in natural language form to the physical retrieval dimensions of a preset indicator data table. This avoids the failure of traditional rule matching in parsing fuzzy contexts. After obtaining the target business data based on the identified dimensions, the large model is used to fuse the scattered indicator values ​​with the original query intent, so that the final output response information is presented as logically descriptive natural language. This establishes a direct connection from the underlying data to the high-level semantics, ensuring the consistency between the feedback result and the user's actual query intent. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the steps of an indicator table retrieval and question-answering method based on large model identification, according to an embodiment of this application. Figure 2 This is a schematic diagram of the overall logical architecture and processing flow of a retail indicator table retrieval and question-answering scheme based on large model recognition according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an indicator table retrieval and question-answering system based on large model recognition according to an embodiment of this application; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0025] This application provides a query and question-answering scheme based on a large model for indicator table retrieval. First, it acquires the text to be processed, containing the query intent, and then uses a large model to perform date recognition and indicator recognition to determine the target date and indicator information. During the recognition process, ambiguous dates are handled by performing dimensional completion or offset calculation on time semantic elements, and the intent indicators are accurately extracted and supplemented by vectorization processing, large model filtering, and association rules. Subsequently, data is retrieved from a preset indicator data table to obtain the target business data. If the original data is missing, alternative date information is determined based on the chronological position for compensation retrieval. Finally, the large model is used to combine the text to be processed with the retrieved target business data to generate text. The generated natural language description text is then associated with the data source and displayed, thereby providing users with factually based answers.

[0026] The following, in conjunction with the accompanying drawings, provides a detailed description of an indicator table retrieval and question-answering scheme based on large model recognition, through specific embodiments and application scenarios.

[0027] Reference Figure 1 The diagram illustrates a flowchart of the steps of an indicator table retrieval and question-answering method based on large model recognition, according to an embodiment of this application.

[0028] Step 101: Obtain the text to be processed containing the query intent.

[0029] The text to be processed originates from various unstructured expressions accumulated in the banking retail business scenario, typically involving requests for information on existing or incremental changes in core indicators such as financial assets, personal deposit and loan status, and the number of various retail customers. Since business personnel often rely on manual retrieval and filtering to obtain data from daily reports, system reports, and other channels in their daily work, the acquisition of the text to be processed involves converting natural language query inputs into the raw data source for subsequent automated processing.

[0030] The text to be processed contains business and time-related information relevant to the user, serving as the foundational input for all subsequent recognition modules. At the technical implementation level, upon receiving the text, it is passed as an instruction stream to the large-scale model recognition component. Obtaining the external input text sequence through a standardized interface ensures that subsequent processing logic can specifically interpret the user's intent. In the context of retail business, this text encompasses diverse query logic, ranging from macro-level business understanding to targeted indicator analysis. As a carrier of information, it provides crucial contextual material for the subsequent extraction of target date and target indicator information.

[0031] Step 102: Use a large model to identify the date in the text to be processed and obtain the target date information.

[0032] After acquiring the text to be processed, a large model is used to perform deep semantic analysis on the text to perform date recognition tasks and ultimately obtain the target date information. The large model can accurately extract the temporal semantic elements in the text to be processed, and can recognize the complete set of dates even if the text contains multiple time nodes. The recognition results are output in a preset structured format, such as the standard string "XXXX year XX month XX day".

[0033] To address the frequent use of ambiguous date descriptions in retail metric queries, the large model leverages its built-in reasoning capabilities in conjunction with the current system time for logical processing. For queries mentioning "today" or "currently," the large model determines the date based on the system time at the time of the query; for relative time descriptions such as "yesterday," "the day before yesterday," or "last Friday," the date is determined through offset calculations. In scenarios where time elements are missing, the large model executes dimension completion logic: if the text only contains "15th," it is completed to a specific year, month, and day based on the current month and year; if the text only mentions a month or year, it is completed to the last day of that time period according to business rules, such as completing "August" as "August 31, 2025."

[0034] Step 103: Use the large model to identify indicators in the text to be processed and obtain target indicator information.

[0035] Building upon date recognition, this paper further utilizes a large model and vectorization techniques to identify indicators in the text to be processed, thereby determining the target indicator information. This process first maps the original indicators and their similar indicators from a pre-defined indicator library to a high-dimensional continuous vector space, converting them into fixed-dimensional numerical vectors. Simultaneously, the text to be processed undergoes the same vectorization encoding process. By calculating the semantic similarity between the text vector and the candidate indicator vectors, several initially selected indicators with the highest similarity are retrieved from the vector space as candidate objects.

[0036] Subsequently, a large-scale model is used to filter and determine the authenticity of the initially selected indicators. For each candidate indicator, the large-scale model considers the context of the text to be processed to determine whether it truly matches the user's real query intent and outputs the determination result. Only indicators that receive a positive determination are considered intent indicators. To ensure that the recognition results can be physically aligned with the underlying preset indicator data table, similar indicator mapping and special indicator association processing are also involved. For example, a preset dictionary is used to map intent indicators back to their original indicator expressions in the table; if the recognition result contains "financial assets", supplementary indicators such as "financial assets (assessment criteria)" are automatically added according to association rules.

[0037] Step 104: Based on the target date information and target indicator information, perform data retrieval in the preset indicator data table to obtain the target business data.

[0038] After obtaining accurate target date and target metric information, a search is performed in a pre-defined metric data table based on these dimensions to extract the target business data. The search process utilizes data search tools to locate rows of data that meet the criteria in the latest table file. If the cell data corresponding to the target date and target metric exists completely, it is directly extracted and returned. The core of this step lies in converting the previously identified semantic elements into data location instructions in the physical table.

[0039] Considering the periodicity of retail data collection and the potential risk of data loss, this step includes logic for handling missing data based on time-series location. When the original data corresponding to the query date is missing, a replacement date is determined based on the logical relationship between the target date and the current system time. Specifically, if the target date belongs to the current month, the nearest valid data date before that date will be searched and returned; if the target date is not in the current month, the last day of the month or the last existing date within the month will be retrieved first.

[0040] Step 105: Generate text based on the text to be processed and the target business data to obtain response information corresponding to the query intent.

[0041] After data retrieval, the language generation capabilities of the large-scale model are utilized to generate text based on the text to be processed and the retrieved target business data, thus obtaining the final response information. In this stage, the large-scale model semantically fuses the original query intent with objective indicator values, generating descriptive natural language responses to the user's questions. Through summarizing and analyzing indicator data, the large-scale model can answer complex questions, such as the increasing or decreasing trends of indicators and numerical comparisons.

[0042] To further enhance the rigor of the responses and address the illusion problem of large models, the generated response information not only includes natural language descriptions but also explicitly displays the retrieved target business data rows as a reference for users.

[0043] In this embodiment, a large model is used to identify the date and indicator dimensions of the text to be processed, directly mapping the query intent in natural language form to the physical retrieval dimensions of a preset indicator data table. This avoids the failure of traditional rule matching in parsing fuzzy contexts. After obtaining the target business data based on the identified dimensions, the large model is used to fuse the scattered indicator values ​​with the original query intent, so that the final output response information is presented as logically descriptive natural language. This establishes a direct connection from the underlying data to the high-level semantics, ensuring the consistency between the feedback result and the user's actual query intent.

[0044] In one exemplary embodiment of this application, one implementation method for using a large model to identify the date in the text to be processed and obtain the target date information is as follows: extracting time semantic elements from the text to be processed using a large model; obtaining the reference time corresponding to the text to be processed when the time semantic elements are missing a time dimension or contain relative time expressions; and performing dimension completion or offset calculation on the time semantic elements according to the reference time to obtain the target date information that conforms to a preset format.

[0045] Specifically, upon receiving text containing the query intent, the large model is invoked to perform deep analysis of the text content, extracting all time-related semantic information. These time semantic elements may be expressed as explicit dates (such as "October 15, 2025") or as implicit or relative time descriptions (such as "yesterday," "last Friday," "August," or "15th"). During the extraction process, the large model will initially identify the time dimension represented by these elements and determine whether they satisfy the complete year-month-day structure required for subsequent retrieval.

[0046] During the recognition process, if a key time dimension is missing from the time semantic elements (e.g., only mentioning the month or date), or if it contains relative time expressions that require calculation based on a specific benchmark, the reference time corresponding to the text to be processed will be automatically obtained. This reference time is usually the current system time when the user initiated the query. Based on this reference time, corresponding logical calculations are performed: for words such as "yesterday" and "the day before yesterday," the exact date coordinates are calculated using offsets; for expressions such as "15th" that lack the year and month, the dimension is completed by combining the year and month information from the reference time. In particular, for broad queries that only ask for a specific month or year, the query is completed to the last day of that time period according to preset business logic, for example, "August" is completed to "August 31, 2025," or "2024" is completed to "December 31, 2024." Finally, all recognized and completed results are converted into a unified preset format, such as returning target date information conforming to the "XXXX year XX month XX day" specification.

[0047] This implementation method combines reference time with the reasoning capabilities of large models to achieve automated completion and accurate conversion of non-standard, ambiguous, and relative time expressions, significantly improving the flexibility and fault tolerance of date recognition. It also solves the problem of retrieval failure caused by incomplete date information or semantic ambiguity when processing natural language queries using traditional technologies.

[0048] In one exemplary embodiment of this application, one implementation method for using a large model to identify indicators in the text to be processed and obtain target indicator information is as follows: the text to be processed is vectorized with candidate indicators in a preset indicator library, and several preliminary indicators are determined from the candidate indicators based on vector similarity; the large model is used to determine whether the preliminary indicators meet the query intent to obtain the intent indicator; and the target indicator information is determined based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library.

[0049] Specifically, all original indicators covered in the preset indicator library, as well as candidate indicators with semantic similarity to each original indicator, are converted into numerical vectors in a high-dimensional space. Simultaneously, the acquired text to be processed undergoes the same vectorization operation, converting it into a numerical vector. By calculating the semantic distance or cosine similarity between the user query vector and each candidate indicator vector in the vector space, the degree of semantic overlap between the two can be quantified, and the top five similarity indicators are selected as initial indicators. These initial indicators not only include items with the same names as the original indicators but also semantically highly related synonyms.

[0050] To further eliminate semantic confusion and improve recognition accuracy, a large-scale model is used to perform secondary filtering on the initial selected indicators. The large-scale model considers the context of the text to be processed and determines whether each initial selected indicator truly represents the user's actual query intent, returning the result in a specific format (e.g., "Indicator Name: Yes / No"). Only indicators judged as "Yes" are confirmed as intent indicators. Finally, using a corresponding dictionary from a pre-defined indicator library, these filtered intent indicators are uniformly mapped back to the standard original indicator names in the table, where the original indicators directly correspond to themselves.

[0051] This implementation method utilizes vectorization to achieve rapid filtering of massive indicators in a large-scale semantic space, and eliminates irrelevant interference items through secondary intent judgment of a large model, thus ensuring both the coverage of recognition and the accuracy of results, and solving the problem of difficulty in accurately matching complex business terms in natural language queries.

[0052] In one exemplary embodiment of this application, one way to determine target indicator information based on the mapping relationship between intent indicators and original indicators in a preset indicator library is as follows: identify whether there are preset associated indicators in the intent indicators; if it is determined that there are associated indicators, add corresponding supplementary indicators to the intent indicators according to preset association rules; map the intent indicators after adding supplementary indicators to the corresponding original indicators to obtain target indicator information.

[0053] After initial intent filtering, a deep business logic scan is performed on the intent metric list. The purpose is to identify whether the intent metrics include pre-defined related metrics that require linked queries in specific business scenarios. For example, in the retail business, user query intent is often not limited to metrics with a single literal meaning, but also involves supporting data under specific statistical definitions. A series of predefined association rules are used to guide the expansion of such special business logic.

[0054] In practice, if "financial assets" are identified among the intent indicators, the association rules determine whether supplementary data is needed in the current context. If an association is established, the corresponding supplementary indicator, such as "financial assets," is automatically added to the current intent indicator list; conversely, if "financial assets" are identified, they are also added accordingly. Similar association logic is applied between indicators such as "third-party custody margin" and "third party." After all associations are completed, the expanded intent indicator set is then uniformly converted into standard original indicators in the preset indicator library through mapping relationships.

[0055] This implementation method, by introducing association rules and supplementary indicator mechanisms, achieves deep alignment of specific business logic. It can automatically complete the associated data dimensions that are not explicitly mentioned by the user but are required for business analysis, thereby improving the system's business adaptability and professional service capabilities.

[0056] In one exemplary embodiment of this application, one implementation method for obtaining target business data by retrieving data from a preset indicator data table based on target date information and target indicator information is as follows: determining whether there is original data corresponding to the target date information in the preset indicator data table; if it is determined that there is no original data, determining alternative date information based on the time sequence position of the target date information relative to the current system time; retrieving data in the preset indicator data table that corresponds to the alternative date information and the target indicator information to obtain the target business data.

[0057] Once the target date and target indicator information are determined, the actual data extraction task is executed. The retrieval logic first uses the target date information as a benchmark, and uses data processing tools to perform a matching query in the date index column of the preset indicator data table to determine whether the original data for that specific date exists.

[0058] Upon determining that the original data corresponding to the target date does not exist in the table, the system does not directly report a query failure. Instead, it triggers a missing data handling mechanism. This mechanism compares the target date with the current system time or the time period it belongs to to determine the date's temporal position and uses different logical paths to determine the most reasonable alternative date. After determining the alternative date, the retrieval process will be re-executed based on this alternative date and the original target indicator information. The final target business data obtained through retrieval provides solid factual evidence for subsequent responses, effectively avoiding system response interruptions caused by incomplete underlying data.

[0059] This implementation establishes a time-series logic-based data fault-tolerant retrieval mechanism that can automatically locate the most relevant alternative data source when the original data is missing, thereby enhancing the robustness and data availability of the system and ensuring the continuity of the user query process and the reference value of the feedback results.

[0060] In one exemplary embodiment of this application, one implementation of determining alternative date information based on the temporal position of the target date information relative to the current system time is as follows: when the month to which the target date information belongs is the same as the month to which the current system time belongs, the date of the most recently existing data in the preset index data table that is earlier than the target date information is determined as the alternative date information; when the month to which the target date information belongs is earlier than the month to which the current system time belongs, the last day of the month to which the target date information belongs, or the date of the most recently existing data in the month to which the target date information belongs, is determined as the alternative date information.

[0061] First, a month consistency check is performed to determine if the target date falls within the current system's calendar month. If the months match, it indicates the user is querying recent dynamic data. In this case, the system will backtrack along the timeline in the preset indicator data table to find and return the nearest valid date record before the target date, and use that as the replacement date. For example, if the data for the current day has not yet been synchronized, the system will automatically retrieve and display the data from the previous day.

[0062] If the month of the target date is earlier than the month of the current system time, it indicates that the user is querying historical data. In this scenario, we prioritize searching for and attempting to return the data for the last day of that historical month, as the final value of the month is usually the most representative in retail analysis. If the data for the last day of that month still does not exist in the table, we search for the nearest existing date within that month's range as a replacement date. If no corresponding data is found after trying both of these approaches, a message indicating that the data was not found will be displayed.

[0063] This implementation method achieves differentiated processing of recent dynamic and historical data by using different substitution logics for the current month and historical months. This ensures the business rationality of the substitution date information, highly simulates the manual table lookup logic of business personnel, and improves the intelligence level of automated retrieval.

[0064] In one exemplary embodiment of this application, one implementation method for generating text based on the text to be processed and the target business data to obtain response information corresponding to the query intent is as follows: input the text to be processed and the target business data into a large model, obtain the natural language description text generated by the large model that matches the query intent; associate and display the natural language description text with the target business data used as a reference to obtain the response information.

[0065] The original text to be processed (i.e., the user's question) is combined with the retrieved specific target business data rows (including the values ​​of various indicators and their corresponding dates) and input as prompt information into the large model. Based on these real data facts and combined with the user's query intent (such as comparison, summary, or simple query), the large model generates a natural language description text that matches the user's query intent.

[0066] To ensure the rigor of the response and address potential illusions that may arise during long-term generation of large models, the generated text is not simply output. The final response is presented in a contextualized manner: the natural language description generated by the large model serves as a summary, while the original target business data line upon which it is based is displayed directly below or in a prominent position. This presentation method not only provides users with an intuitive conclusion but also allows them to verify the authenticity of the generated content in real time by showcasing the underlying real data, significantly enhancing user trust in the system and improving the user experience.

[0067] This implementation effectively combines the linguistic expressiveness of the large model with the objective authenticity of the underlying retrieval. Through the associated display mode of "text summary + factual data", it not only improves the readability of feedback information, but also avoids data distortion caused by model illusion from a mechanism perspective, and significantly optimizes the efficiency of information acquisition for users in complex retail indicator analysis scenarios.

[0068] Based on the above description of an embodiment of an indicator table retrieval and question-answering method based on large model recognition, the following introduces a retail indicator table retrieval and question-answering scheme based on large model recognition. (Refer to...) Figure 2 , Figure 2 This diagram illustrates the overall logical architecture and processing flow of a retail indicator table retrieval and question-answering scheme based on large model recognition, according to an embodiment of this application. User questions are sequentially processed through a date recognition component and an indicator recognition component to obtain the target date and target indicator. The target date and target indicator are then used to search within a preset indicator data table, ultimately leading to an answer generation module where the response information is obtained.

[0069] During the date recognition process, if the date is determined to be clear, the question date is directly output; if the date is determined to be ambiguous, the question date is determined by the pre-set rule completion logic.

[0070] In the indicator recognition component, the original indicators and their meanings are pre-converted into vectorized indicators and stored using offline embedding. For user questions, embedding-based vectorized question processing is performed, and the questions are matched with the vectorized indicators based on similarity, selecting the top n candidate indicators. Subsequently, a large model is used for refined screening of the candidate indicators, and an indicator meaning dictionary is used to convert the screened candidate indicators back into the original indicators used in the question.

[0071] A table search is performed based on the determined question date and the original question metrics. If the search results show that the data exists, the large model is directly triggered to generate the answer; if the data is missing, a replacement date is determined according to business logic rules, and the corresponding data is obtained before entering the large model to generate the answer, ultimately outputting the response information corresponding to the user's question.

[0072] It should be noted that the indicator table retrieval and question-answering method based on large model recognition provided in this application embodiment can be executed by an indicator table retrieval and question-answering system based on large model recognition, or by a control module within that system for executing the loading of the indicator table retrieval and question-answering method based on large model recognition. This application embodiment uses the execution of the loading of the indicator table retrieval and question-answering method based on large model recognition by an indicator table retrieval and question-answering system as an example to illustrate the indicator table retrieval and question-answering method based on large model recognition provided in this application embodiment.

[0073] Reference Figure 3 This diagram illustrates the structure of an index table retrieval and question-answering system based on large model recognition, according to an embodiment of this application. The system may specifically include the following modules: The text acquisition module 31 is used to acquire the text to be processed containing the query intent; Date recognition module 32 is used to perform date recognition on the text to be processed using a large model to obtain target date information; The indicator recognition module 33 is used to use the large model to recognize indicators in the text to be processed and obtain target indicator information. The data retrieval module 34 is used to retrieve data from a preset indicator data table based on the target date information and the target indicator information to obtain the target business data. The response determination module 35 is used to generate text based on the text to be processed and the target business data to obtain response information corresponding to the query intent.

[0074] In one exemplary embodiment of this application, the date recognition module 32 includes: The temporal semantic element extraction module is used to extract temporal semantic elements from the text to be processed using the large model. The reference time acquisition module is used to acquire the reference time corresponding to the text to be processed when the time semantic element is missing a time dimension or contains a relative time expression. The target date information determination module is used to perform dimensional completion or offset calculation on the time semantic elements according to the reference time to obtain the target date information that conforms to a preset format.

[0075] In one exemplary embodiment of this application, the indicator recognition module 33 includes: The initial selection indicator determination module is used to vectorize the text to be processed and the candidate indicators in the preset indicator library, and determine a number of initial selection indicators from the candidate indicators based on the vector similarity. The intent indicator determination module is used to determine whether the initial selected indicators match the query intent using the large model, and to obtain the intent indicator. The target indicator information determination module is used to determine the target indicator information based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library.

[0076] In one exemplary embodiment of this application, the target indicator information determination module includes: The associated indicator identification module is used to identify whether there are preset associated indicators among the intent indicators; The supplementary indicator addition module is used to add corresponding supplementary indicators to the intent indicator according to preset association rules when it is determined that the associated indicator exists. The intent indicator mapping module is used to map the intent indicator after adding the supplementary indicator to the corresponding original indicator to obtain the target indicator information.

[0077] In one exemplary embodiment of this application, the data retrieval module 34 includes: The raw data judgment module is used to determine whether the raw data corresponding to the target date information exists in the preset indicator data table; The alternative date information determination module is used to determine alternative date information based on the time sequence position of the target date information relative to the current system time when it is determined that the original data does not exist; The target business data determination module is used to retrieve data corresponding to the alternative date information and the target indicator information from the preset indicator data table to obtain the target business data.

[0078] In an exemplary embodiment of this application, the alternative date information determination module is configured to, when the month to which the target date information belongs is the same as the month to which the current system time belongs, determine the date of the most recently existing data in the preset index data table that is earlier than the target date information as the alternative date information; and when the month to which the target date information belongs is earlier than the month to which the current system time belongs, determine the last day of the month to which the target date information belongs, or the date of the most recently existing data in the month to which the target date information belongs, as the alternative date information.

[0079] In one exemplary embodiment of this application, the response determining module 35 includes: The natural language description text acquisition module is used to input the text to be processed and the target business data into the large model, and obtain the natural language description text generated by the large model that matches the query intent; The response information determination module is used to associate and display the natural language description text with the target business data used as a reference to obtain the response information.

[0080] The indicator table retrieval and question-answering system based on large model recognition in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, Ultra-Mobile Personal Computers (UMPCs), netbooks, or Personal Digital Assistants (PDAs), etc., while non-mobile electronic devices can be servers, Network Attached Storage (NAS), Personal Computers (PCs), TeleVision (TVs), ATMs, or self-service machines, etc., and this application embodiment does not impose specific limitations.

[0081] The indicator table retrieval and question-answering system based on large model recognition in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0082] The indicator table retrieval and question-answering system based on large model recognition provided in this application embodiment can achieve... Figure 1 The method embodiments include various processes for implementing an indicator table retrieval and question-answering system based on large model recognition. To avoid repetition, these will not be described in detail here.

[0083] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the various processes of the above-described embodiments of the index table retrieval and question-answering method based on large model recognition, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0084] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0085] Figure 4 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of this application.

[0086] The electronic device 1000 includes, but is not limited to, components such as: a radio frequency unit 1001, a network module 1002, an audio output unit 1003, an input unit 1004, a sensor 1005, a display unit 1006, a user input unit 1007, an interface unit 1008, a memory 1009, and a processor 1010. The input unit 1004 may include a graphics processor 10041 and a microphone 10042. The display unit 1006 may include a display panel 10061. The user input unit 1007 may include a touch panel 10071 and other input devices 10072. The memory 1009 may include applications and an operating system.

[0087] Those skilled in the art will understand that the electronic device 1000 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 1010 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 4 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0088] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described embodiment of the index table retrieval and question-answering method based on large model recognition, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0089] The processor mentioned above is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0090] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and systems in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0092] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for index table retrieval and question answering based on large model recognition, characterized in that, The method includes: Retrieve the text to be processed containing the query intent; The target date information is obtained by using a large model to identify the date in the text to be processed. The large model is used to identify indicators in the text to be processed to obtain target indicator information. Based on the target date information and the target indicator information, data is retrieved from the preset indicator data table to obtain the target business data. Text is generated based on the text to be processed and the target business data to obtain response information corresponding to the query intent.

2. The method according to claim 1, characterized in that, The process of using a large model to perform date recognition on the text to be processed to obtain target date information includes: The large model is used to extract temporal semantic elements from the text to be processed; If the time dimension is missing or relative time representation is included in the time semantic elements, obtain the reference time corresponding to the text to be processed. Based on the reference time, the time semantic elements are augmented with dimensions or offsets are calculated to obtain the target date information that conforms to a preset format.

3. The method according to claim 1, characterized in that, The step of using the large model to identify indicators in the text to be processed to obtain target indicator information includes: The text to be processed is vectorized with candidate indicators in a preset indicator library, and several preliminary indicators are determined from the candidate indicators based on vector similarity. The large model is used to determine whether the initially selected indicators match the query intent, thus obtaining the intent indicators. The target indicator information is determined based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library.

4. The method according to claim 3, characterized in that, The step of determining the target indicator information based on the mapping relationship between the intent indicator and the original indicators in the preset indicator library includes: Identify whether there are any preset related indicators among the intent indicators; If the existence of the associated indicators is confirmed, corresponding supplementary indicators are added to the intent indicators according to preset association rules. The intention indicator, after adding the supplementary indicator, is mapped to the corresponding original indicator to obtain the target indicator information.

5. The method according to claim 1, characterized in that, The step of retrieving target business data from a preset indicator data table based on the target date information and the target indicator information includes: Determine whether the original data corresponding to the target date information exists in the preset indicator data table; If it is determined that the original data does not exist, alternative date information is determined based on the time sequence position of the target date information relative to the current system time; The target business data is obtained by retrieving the data corresponding to the alternative date information and the target indicator information from the preset indicator data table.

6. The method according to claim 5, characterized in that, The step of determining the alternative date information based on the time sequence position of the target date information relative to the current system time includes: If the month to which the target date information belongs is the same as the month to which the current system time belongs, the date of the most recent existing data in the preset indicator data table that is earlier than the target date information is determined as the alternative date information; If the month to which the target date information belongs is earlier than the month to which the current system time belongs, the last day of the month to which the target date information belongs, or the date with the most recently existing data in the month to which the target date information belongs, shall be determined as the alternative date information.

7. The method according to claim 1, characterized in that, The step of generating text based on the text to be processed and the target business data to obtain response information corresponding to the query intent includes: The text to be processed and the target business data are input into the large model to obtain the natural language description text generated by the large model that matches the query intent; The natural language description text is associated with the target business data used as a reference and displayed to obtain the response information.

8. A retrieval and question-answering system for indicator tables based on large model recognition, characterized in that, The system includes: The text acquisition module is used to acquire the text to be processed, which contains the query intent; The date recognition module is used to perform date recognition on the text to be processed using a large model to obtain the target date information; The indicator recognition module is used to use the large model to recognize indicators in the text to be processed and obtain target indicator information. The data retrieval module is used to retrieve data from a preset indicator data table based on the target date information and the target indicator information to obtain the target business data. The response determination module is used to generate text based on the text to be processed and the target business data, and obtain response information corresponding to the query intent.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the index table retrieval and question-answering method based on large model recognition as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the index table retrieval and question-answering method based on large model recognition as described in any one of claims 1-7.

Images