Table retrieval method and device and electronic equipment

Abstract

The invention discloses a table retrieval method and device and electronic equipment. The method comprises the steps of obtaining a table retrieval request of a target object; in response to the table retrieval request, a target table retrieval library with the same category information as the to-be-retrieved table is determined, and the target table retrieval library comprises a plurality of tables with the same category information as the to-be-retrieved table; at least one candidate table is retrieved from a target table retrieval library, and a first matching degree between table description information of the candidate table and the table retrieval request meets a first matching degree threshold value; based on the table description information corresponding to the at least one candidate table, a target table is determined from the at least one candidate table, and the first similarity between the table description information corresponding to the target table and the table retrieval request meets a first similarity threshold value; and pushing the target table to the target object. According to the method and the device, the technical problem of inaccurate table retrieval in related technologies is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence question answering technology, and in particular to a table retrieval method, device and electronic device. Background Art

[0002] At present, table retrieval technology mainly relies on converting natural language queries into structured query statements (Structured Query Language, referred to as SQL) or determining target slots based on semantic similarity. These methods have improved the retrieval efficiency and accuracy of table data to a certain extent. However, in practical applications, due to the complexity of natural language understanding and the uncertainty of table data quality, traditional methods face many challenges. For example, complex query statements or professional terms in specific fields may be beyond the full understanding of current natural language processing technology, resulting in inaccurate retrieval results; at the same time, errors or incompleteness of table data will also directly affect the retrieval effect, especially when converting natural language queries into structured query statements such as SQL. If the conversion is inaccurate, the query statement may not be executed correctly, thereby affecting the answer output. Therefore, there is a technical problem of inaccurate table retrieval.

[0003] Currently, no effective solution has been proposed for the above-mentioned technical problem of inaccurate table retrieval. Summary of the invention

[0004] The embodiments of the present invention provide a table retrieval method, device and electronic device to at least solve the technical problem of inaccurate table retrieval.

[0005] According to one aspect of an embodiment of the present invention, a table retrieval method is provided. The method may include: obtaining a table retrieval request of a target object, wherein the table retrieval request includes category information of the table to be retrieved requested by the target object; in response to the table retrieval request, determining a target table retrieval library with the same category information as the table to be retrieved, wherein the target table retrieval library includes multiple tables with the same category information as the table to be retrieved; retrieving at least one candidate table from the target table retrieval library, wherein the first matching degree between the table description information of the candidate table and the table retrieval request satisfies a first matching degree threshold; based on the table description information corresponding to the at least one candidate table, determining a target table from the at least one candidate table, wherein the first similarity between the table description information corresponding to the target table and the table retrieval request satisfies a first similarity threshold; and pushing the target table to the target object.

[0006] Optionally, the table retrieval method also includes: obtaining row texts contained in the target table; using a fuzzy matching method, selecting row texts from the row texts whose second matching degree with the table retrieval request satisfies a second matching threshold, to obtain candidate row texts; calculating a second similarity between the table retrieval request and the candidate row texts, wherein the second similarity is used to characterize the degree of similarity between the table retrieval request and the candidate row texts; based on the second similarity and the second similarity threshold, selecting target row texts from the candidate row texts that meet the second similarity threshold with the table retrieval request; and pushing the target row text to the target object.

[0007] Optionally, in response to a form retrieval request, a target form retrieval library having the same category information as the form to be retrieved is determined, including: in response to the form retrieval request, performing semantic recognition on the form retrieval request to obtain the category information to which the form to be retrieved belongs; based on the category information to which the form to be retrieved belongs, determining a target form retrieval library from multiple form retrieval libraries, wherein the multiple form retrieval libraries correspond to different category information.

[0008] Optionally, retrieving at least one candidate table from a target table retrieval library includes: using a fuzzy matching method to respectively determine a first degree of match between a plurality of table description information included in the target table retrieval library and the table retrieval request; respectively comparing the first degrees of match between the plurality of table description information and the table retrieval request with a first degree of match threshold to obtain a first comparison result; and based on the first comparison result, retrieving at least one candidate table from the target table retrieval library.

[0009] Optionally, based on table description information corresponding to at least one candidate table, a target table is determined from at least one candidate table, including: determining a first similarity between the table description information corresponding to at least one candidate table and the table retrieval request; comparing the first similarity between the table description information corresponding to at least one candidate table and the table retrieval request with a first similarity threshold to obtain a second comparison result; based on the second comparison result, determining the target table from at least one candidate table.

[0010] Optionally, the table retrieval method further includes: acquiring a plurality of tables; and dividing the plurality of tables into a plurality of categories according to application fields, wherein tables of different categories correspond to different category information.

[0011] Optionally, the table retrieval method further includes: adding table description information to the multiple tables respectively according to key information contained in the multiple tables, wherein the table description information is used to describe the key information of the corresponding table.

[0012] Optionally, the table retrieval method further includes: establishing a table retrieval library of the same number according to the number of category information corresponding to the multiple tables; and storing the tables with the same category information in one table database.

[0013] Optionally, the table retrieval method further includes: splitting the merged cells in the multiple tables into independent cells respectively; and converting the content cells of the independent cells included in the multiple tables into text form.

[0014] According to another aspect of an embodiment of the present invention, a table retrieval device is also provided. The device may include: an acquisition unit, used to acquire a table retrieval request of a target object, wherein the table retrieval request includes category information to which the table to be retrieved requested by the target object belongs; a first determination unit, used to determine a target table retrieval library with the same category information as the table to be retrieved in response to the table retrieval request, wherein the target table retrieval library includes multiple tables with the same category information as the table to be retrieved; a retrieval unit, used to retrieve at least one candidate table from the target table retrieval library, wherein the first matching degree between the table description information of the candidate table and the table retrieval request satisfies a first matching degree threshold; a second determination unit, used to determine a target table from at least one candidate table based on the table description information corresponding to the at least one candidate table, wherein the first similarity between the table description information corresponding to the target table and the table retrieval request satisfies a first similarity threshold; and a push unit, used to push the target table to the target object.

[0015] According to another aspect of an embodiment of the present invention, there is further provided an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the method in each embodiment of the present invention is executed when the program is running.

[0016] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0018] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0019] According to another aspect of the embodiments of the present invention, a computer program is provided. When the computer program is executed by a processor, the methods in the embodiments of the present invention are implemented.

[0020] In an embodiment of the present invention, a table retrieval request of a target object is obtained, wherein the table retrieval request includes the category information of the table to be retrieved requested by the target object; in response to the table retrieval request, a target table retrieval library with the same category information as the table to be retrieved is determined, wherein the target table retrieval library includes multiple tables with the same category information as the table to be retrieved; at least one candidate table is retrieved from the target table retrieval library, wherein the first matching degree between the table description information of the candidate table and the table retrieval request satisfies the first matching degree threshold; based on the table description information corresponding to at least one candidate table, a target table is determined from at least one candidate table, wherein the first similarity between the table description information corresponding to the target table and the table retrieval request satisfies the first similarity threshold; and the target table is pushed to the target object. That is to say, in this embodiment, the category information of the table to be retrieved included in the table retrieval request of the target object can be used to quickly locate the table retrieval library related to it, thereby avoiding blind searching in a large and diverse set of tables, greatly shortening the retrieval time, and improving the user experience. Moreover, through step-by-step retrieval, potential candidate tables are screened out first, and then the target table is further confirmed, avoiding indiscriminate similarity calculations for all tables, effectively reducing the consumption of computing resources and improving system performance. Furthermore, by adopting the dual threshold mechanism of the first matching degree and the first similarity, it is possible to screen out tables that truly match the retrieval request, and even in the face of ambiguous or complex query statements, it is possible to provide accurate retrieval results, thereby enhancing the robustness and reliability of the system, achieving the purpose of accurately identifying the target table that matches the table retrieval request of the target object, and thus achieving the technical effect of improving the accuracy of table retrieval and solving the technical problem of low table retrieval accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0022] Figure 1 is a flow chart of a table retrieval method according to an embodiment of the present invention;

[0023] Figure 2 is a flow chart of a table content retrieval method based on intent recognition and similarity retrieval according to an embodiment of the present application;

[0024] Figure 3It is a schematic diagram of a table retrieval device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, functional component or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, functional components or devices.

[0027] According to an embodiment of the present invention, an embodiment of a table retrieval method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] Figure 1 is a flow chart of a table retrieval method according to an embodiment of the present invention. Figure 1 As shown, the method may include the following steps:

[0029] Step S101, obtaining a table search request of a target object.

[0030] In the technical solution provided in the above step S101 of the present invention, the target object is a user who performs table retrieval, and the table retrieval request includes category information to which the table to be retrieved requested by the target object belongs.

[0031] In this embodiment, the form retrieval request can also be regarded as a query statement in natural language input by the user on the interactive interface of the form retrieval system. The statement can be an inquiry for product information, an inquiry for service details, a demand for specific data, or any other form of information retrieval request, without specific limitation here.

[0032] Optionally, after obtaining the form retrieval request of the target object, the intention recognition method can be used to extract the retrieval intent of the target object from the retrieval statement of the target object, and then extract the category information of the to-be-retrieved form requested to be retrieved by the target object from the retrieval intent, wherein the category information to which the form belongs is at least one of the following: consultation category, complaint category, repair category, and maintenance category.

[0033] Optionally, one of the methods of intent recognition is to use the natural language processing capabilities of a large language model, take the category information corresponding to the pre-determined table and the search sentence of the target object as input, control it through prompt words, and infer the search intent of the target object. In addition, a pre-trained intent recognition model can also be used to analyze the intent of the target object. The specific method of intent recognition is not limited here.

[0034] Step S102, in response to the table search request, determining a target table search library having the same category information as the table to be searched.

[0035] In the technical solution provided in the above step S102 of the present invention, the target table retrieval library includes multiple tables with the same category information as the table to be retrieved. It should be noted that the multiple tables are divided into different categories in advance according to the application fields of the multiple tables included in the database, wherein one category corresponds to one category information, and the tables belonging to one category can be clustered into one table retrieval library to obtain multiple table retrieval libraries, each of which corresponds to different category information.

[0036] Optionally, various types of forms are divided in advance according to the application fields of multiple forms. For example, forms of product information, service content, operation instructions, promotional activities, and purchase process are divided into consulting categories, and then the forms belonging to the consulting category are clustered into a table retrieval library, and the category information corresponding to the table retrieval library is consulting. Forms of product quality, service issues, logistics distribution, and after-sales experience are divided into complaints, and then the forms belonging to the complaints are clustered into a table retrieval library, and the category information corresponding to the table retrieval library is complaints. Forms of fault diagnosis, troubleshooting, maintenance services, and parts replacement are divided into repair reporting categories, and then the forms belonging to the repair reporting category are clustered into a table retrieval library, and the category information corresponding to the table retrieval library is repair reporting. The tables of maintenance consultation, warranty service, and warranty status query are classified as maintenance, and the tables belonging to the maintenance are clustered into a table search library. The category information corresponding to the table search library is maintenance. The tables of account management, account problems, and personal information are classified as account management. The tables belonging to the account management are clustered into a table search library. The category information corresponding to the table search library is account management. This is only an illustrative example, and does not limit the clustering method of the table search library and the category information corresponding to the table search library.

[0037] In this embodiment, after obtaining a table retrieval request for the target object, since the table retrieval request includes category information to which the table to be retrieved belongs, based on this, a table retrieval library with the same category information as the table to be retrieved can be searched from multiple table retrieval libraries according to the category information to which the table to be retrieved included in the table retrieval request, and then the found table retrieval library is determined as the target table retrieval library, wherein the category information to which the target table retrieval library belongs is the same as the category information to which the table to be retrieved in the table retrieval request belongs.

[0038] Optionally, since the category information of the target form retrieval library is the same as the category information of the form to be retrieved in the form retrieval request, searching the form to be retrieved in the target form retrieval library can narrow the scope of the search and improve the effectiveness and accuracy of subsequent retrieval steps.

[0039] Step S103: retrieve at least one candidate table from the target table retrieval library.

[0040] In the technical solution provided in the above step S103 of the present invention, the first matching degree between the table description information of the candidate table and the table retrieval request satisfies the first matching degree threshold, wherein the table description information of the candidate table is used to indicate the description text of the table, the name of the table, the category to which the table belongs, and other information.

[0041] In this embodiment, after determining the target table retrieval library, you can further search for tables that match the table to be retrieved in the target table retrieval library, and then calculate the matching degree between the table description information corresponding to the table retrieved from the target table retrieval library and the table retrieval request. For the convenience of explanation, this matching degree is called the first matching degree, and then compare the first matching degree corresponding to each table in the target table retrieval library with the first matching degree threshold. According to the comparison result, at least one candidate table is determined from the tables retrieved from the target table retrieval library. Among them, the first matching threshold is a pre-set value used to determine whether the first matching degree is high enough to determine whether the table retrieved from the target retrieval library is a candidate table. If the first matching degree is higher than the first matching degree threshold, the corresponding table is considered to be sufficiently relevant to the table retrieval request of the target object, and is thus selected as a "candidate table".

[0042] Optionally, data such as table name, table category, and table description text in the table description information corresponding to the table retrieved in the target table retrieval library are pre-converted into a row text format to facilitate calculation of a first matching degree with the table retrieval request.

[0043] Optionally, since the fuzzy matching algorithm can handle imperfect matching situations, even if there are spelling errors, synonym differences or differences in expression between the query statement and the table description information, the similarity between the two can be calculated. Based on this, the first matching degree between the row text corresponding to the table description information corresponding to the table retrieved in the target table retrieval library and the table retrieval request is calculated by the fuzzy matching algorithm.

[0044] In this step, based on the fuzzy matching algorithm and the matching degree threshold, tables that are highly relevant to the table retrieval request of the target object can be screened out from the target table retrieval library of Jiaotong University as candidate tables, thereby greatly narrowing the scope of subsequent retrieval and improving the retrieval efficiency and the accuracy of the results.

[0045] Optionally, if no table that meets the threshold is found during the candidate screening process, a fallback strategy is usually adopted, such as relaxing the threshold or providing a list of all potentially relevant tables for further selection by the target object, to ensure that useful information can be provided to the target object.

[0046] Step S104: determining a target table from at least one candidate table based on the table description information corresponding to at least one candidate table.

[0047] In the technical solution provided in the above step S104 of the present invention, the first similarity between the table description information corresponding to the target table and the table search request satisfies a first similarity threshold.

[0048] In this embodiment, at least one candidate table that is highly relevant to the table retrieval request has been retrieved from the target table retrieval library in the above step S103. Based on this, a table that is most similar to the table retrieval request can be determined from at least one candidate table as the target table according to the table description information of the at least one candidate table.

[0049] Optionally, the description information of the candidate table includes: the name of the table, the category to which the table belongs, the description text of the table, etc. This information has been converted into a line text format in the preprocessing stage to facilitate the subsequent similarity calculation of the system. Based on this, the vector similarity method can be used to calculate the similarity between the table retrieval request of the target object and the description information of the candidate table. For the sake of convenience, this similarity can be referred to as the first similarity. This step is a deeper analysis based on fuzzy matching screening. By converting the text into a vector and calculating the distance between the vectors, the semantic similarity between the texts can be more accurately judged. Among them, the vector similarity method can adopt a variety of algorithms, such as cosine similarity, text similarity evaluation based on deep learning, etc., which are not specifically limited here.

[0050] Optionally, a first similarity threshold is pre-set for determining whether the first similarity obtained by the vector similarity calculation is high enough. Based on this, after the first similarity between the table retrieval request and the table description information of the candidate table is calculated using the vector similarity method, the first similarity can be compared with the first similarity threshold, and then the target table can be determined from the candidate tables based on the comparison result.

[0051] Optionally, a table whose first similarity reaches or exceeds a first similarity threshold is selected from the candidate tables as the final target table. If more than one candidate table meets the condition, the Top_k most relevant tables are selected, where k is a preset value representing the number of tables that the user may be interested in.

[0052] In this step, the scope can be further narrowed down from the multiple candidate tables initially screened out based on more accurate vector similarity calculations to determine one or more target tables that best meet the user's query requirements. The target table can be used for the next step of detailed row text retrieval, thereby providing the specific information required by the user.

[0053] Optionally, if no table that meets the similarity threshold is found at this stage, other solutions are usually provided, such as lowering the similarity threshold, displaying all candidate tables for users to manually select, etc., to ensure that the most accurate and satisfactory retrieval results can be provided to users.

[0054] Step S105: Push the target table to the target object.

[0055] In the technical solution provided in the above step S105 of the present application, after the target form is determined through the above step S104, the retrieved target form can be displayed on the interactive interface of the form retrieval system for the target object to consult, or a link to the target form can be provided to the target object for user access, thereby achieving the purpose of pushing the retrieved target form to the target object.

[0056] The above steps S101 to S105 of the present invention can quickly locate the table retrieval library related to the target object through the category information of the table to be retrieved included in the table retrieval request of the target object, avoiding blind search in a large and diverse table collection, greatly shortening the retrieval time, and improving the user experience. Moreover, by step-by-step retrieval, potential candidate tables are preferentially screened out, and then the target table is further confirmed, avoiding indiscriminate similarity calculations for all tables, effectively reducing the consumption of computing resources and improving system performance. Furthermore, by adopting the dual threshold mechanism of the first matching degree and the first similarity, the table that truly matches the retrieval request can be screened out, and even in the face of ambiguous or complex query statements, accurate retrieval results can be provided, thereby enhancing the robustness and reliability of the system, achieving the purpose of accurately identifying the target table that matches the table retrieval request of the target object, and then achieving the technical effect of improving the accuracy of table retrieval, and solving the technical problem of low table retrieval accuracy.

[0057] The above method of this embodiment is further introduced below.

[0058] As an optional implementation method, the table retrieval method also includes: obtaining the row text contained in the target table; using a fuzzy matching method, selecting from the row text the row text whose second matching degree with the table retrieval request satisfies a second matching threshold, to obtain candidate row text; calculating the second similarity between the table retrieval request and the candidate row text, wherein the second similarity is used to characterize the degree of similarity between the table retrieval request and the candidate row text; based on the second similarity and the second similarity threshold, selecting from the candidate row text the target row text that satisfies the second similarity threshold with the table retrieval request; and pushing the target row text to the target object.

[0059] In this embodiment, after the target table is retrieved, all the row texts in the target table can be further obtained, and then the second matching degree between the table search request and all the row texts in the target table can be calculated using the fuzzy matching method. It should be noted that the row texts in the target table are converted through the preprocessing step, ensuring that each row contains the column name and the corresponding value, and the format is unified and easy to retrieve.

[0060] Optionally, a second matching degree threshold is set in advance. Based on this, after the second matching degree between each row text in the target table and the table retrieval request is calculated by the fuzzy matching method, the second matching degree corresponding to each row text in the target table can be compared with the second matching degree threshold, and then the candidate row text with a second matching degree greater than the second matching degree threshold is selected according to the comparison result, wherein the candidate row text is used to indicate the row text in the target table that is highly relevant to the table retrieval request.

[0061] Optionally, after obtaining the candidate line text, the vector similarity method can be further used to calculate the similarity between the candidate line text and the user search request. For the convenience of explanation, this similarity can be called the second similarity. Among them, the vector similarity method can more accurately capture the semantic relationship of the text. By converting the text into a vector and calculating the distance or angle between the vectors, the obtained similarity value can more accurately reflect the semantic similarity between the texts.

[0062] Optionally, after calculating the second similarity between each candidate line text and the table retrieval request, each second similarity can be compared with a preset second similarity threshold, and then the line text of the table retrieval request that best matches the target object is selected from the candidate text as the target line text.

[0063] Optionally, if multiple lines of text in the target table meet the conditions (the second similarities corresponding to the multiple lines of text are greater than the second similarity threshold), the Top_n most similar lines of text can be selected, where n is a preset value representing the number of lines of text that the target object may be interested in.

[0064] Optionally, after determining the target line text, the target line text can be pushed to the target object. The push form can be to directly display the line text content on the user interface, provide a download link for the line text, or present it in the form of a summary or keyword list, depending on the specific design of the system and the requirements of the user interface.

[0065] Through the above steps, the row text most relevant to the user's search request can be found from the target table, which not only improves the accuracy and relevance of the search results, but also ensures that the user can obtain the most direct and valuable answers, thereby greatly improving the user experience and search efficiency.

[0066] As an optional implementation method, step S102, in response to a table retrieval request, determines a target table retrieval library having the same category information as the table to be retrieved, including: in response to the table retrieval request, performing semantic recognition on the table retrieval request to obtain the category information to which the table to be retrieved belongs; based on the category information to which the table to be retrieved belongs, determining a target table retrieval library from multiple table retrieval libraries, wherein the multiple table retrieval libraries correspond to different category information.

[0067] In this embodiment, the form retrieval request may be a question or query statement raised by the target object (user), which may involve multiple fields such as product information, service content, and operation instructions. After receiving the form retrieval request of the target object, the form retrieval request of the target object may be semantically recognized to determine the category information of the form to be retrieved requested by the target object. Among them, semantic recognition of the form retrieval request can be achieved through natural language processing technology, especially a large language model or a pre-trained intent recognition model. The purpose of semantic recognition is to understand the intent of the user's query and the category information to which it belongs, so as to provide direction for subsequent form retrieval.

[0068] Optionally, semantic recognition can be used to determine the category information to which the table to be retrieved should belong. These category information are preset when the table is hierarchically classified and descriptive text is added in the early stage, such as consultation, complaint, repair, maintenance, account management, etc. Each category corresponds to a specific field and type of table data.

[0069] Optionally, multiple table retrieval libraries are pre-stored, each table retrieval library corresponds to a category information, and contains all relevant tables under the category. These retrieval libraries are pre-processed and information integrated to ensure the consistency and searchability of table data.

[0070] Optionally, based on the category information of the to-be-searched form requested by the form search request of the target object, a target form search library can be determined from multiple form search libraries, and the target form search library contains all forms that match the category information of the to-be-searched form. By determining the target form search from multiple form search libraries, the search scope can be greatly narrowed, and the efficiency and accuracy of subsequent searches can be improved.

[0071] As an optional implementation method, step S103, retrieving at least one candidate table from the target table retrieval library, includes: using a fuzzy matching method to respectively determine the first matching degrees between multiple table description information included in the target table retrieval library and the table retrieval request; respectively comparing the first matching degrees between the multiple table description information and the table retrieval request with a first matching degree threshold to obtain a first comparison result; based on the first comparison result, retrieving at least one candidate table from the target table retrieval library.

[0072] In this embodiment, for each table description information in the target table retrieval library, fuzzy matching is performed with the table retrieval request raised by the target object. Fuzzy matching is a technology that can handle incomplete or inaccurate data matching, and can accept a certain degree of spelling errors, synonym changes or differences in expression methods. By calculating the similarity between the two texts, it is determined whether the two are sufficiently matched. For the convenience of explanation, the result of the calculation can be called the first matching degree, which reflects the matching degree between each table description information and the target object retrieval request.

[0073] Optionally, a first matching degree threshold is preset to determine whether the first matching degree reaches a certain standard. By comparing the first matching degree between each table description information and the user search request with the first matching degree threshold, the table description information with the first matching degree higher than the first matching degree threshold can be screened out. If the first matching degree is lower than the first matching degree threshold, it is considered that the table has a low relevance to the needs of the target object and may not be further considered.

[0074] Optionally, based on the comparison result between the first matching degree and the first matching degree threshold, all tables corresponding to the table description information having a first matching degree higher than the first matching degree threshold are determined as candidate tables. These candidate tables are preliminarily matched with the user's search request and represent tables that may contain the information required by the user.

[0075] Optionally, if the matching degree of all tables does not reach the threshold, you can choose to regard all tables in the target table retrieval library as candidate tables, or adjust the threshold to re-match to ensure that the retrieval results can be provided to the user.

[0076] As an optional implementation method, step S104, based on the table description information corresponding to at least one candidate table, determines the target table from at least one candidate table, including: determining a first similarity between the table description information corresponding to at least one candidate table and the table retrieval request; comparing the first similarity between the table description information corresponding to at least one candidate table and the table retrieval request with a first similarity threshold to obtain a second comparison result; based on the second comparison result, determines the target table from at least one candidate table.

[0077] In this embodiment, after the candidate tables are determined, more accurate matching techniques such as vector similarity can be further used to screen again from these candidate tables to finally determine one or more target tables. If there are a large number of candidate tables, the screening can be further optimized based on the user's historical search behavior, the update frequency of the table, or other related factors to improve the efficiency of the search and the accuracy of the results.

[0078] For example, a vector similarity calculation method can be used to calculate the first similarity between each candidate table and the table retrieval request of the target object and the candidate table description information, and then the first similarity between the table description information corresponding to at least one candidate table and the table retrieval request is compared with a first similarity threshold; based on the comparison result, a target table is selected from at least one candidate table to obtain a second comparison result; based on the second comparison result, a target table is determined from at least one candidate table, and the determined target table is the table that best meets the user's query requirements.

[0079] As an optional implementation, the table retrieval method further includes: acquiring multiple tables; and dividing the multiple tables into multiple categories according to application fields, wherein tables of different categories correspond to different category information.

[0080] In this embodiment, a large number of tables need to be collected and stored. These tables may come from different data sources, such as internal company databases, public online resources, industry standard documents, etc., and contain a wide range of information types and application fields, such as product information, service content, operation guides, financial data, customer feedback, etc.

[0081] Optionally, after obtaining multiple forms, the obtained multiple forms can be divided into different categories according to their application fields. For example, in an intelligent customer service system, forms may be divided into multiple categories such as product consultation, service issues, account management, and promotion activities, each of which corresponds to specific field information and user needs.

[0082] Optionally, after categorizing a plurality of tables, category information may be added for each table.

[0083] As an optional implementation, the table retrieval method further includes: adding table description information to the multiple tables respectively according to the key information contained in the multiple tables, wherein the table description information is used to describe the key information of the corresponding table.

[0084] In this embodiment, after the multiple tables are classified, table description information may be added to each table according to the key information included in each table, wherein the table description information is a concise and general text for describing the main content and purpose of the table.

[0085] For example, for each table, add descriptive text based on the content it contains, briefly summarize the core content, enrich the table information, support the subsequent retrieval process, ensure that the target table can be located more accurately, and improve retrieval efficiency.

[0086] As an optional implementation, the table retrieval method further includes: establishing the same number of table retrieval libraries according to the number of category information corresponding to the multiple tables; and storing the tables with the same category information in one table retrieval library.

[0087] In this embodiment, after a plurality of tables are classified, the category information of the plurality of tables can be counted to determine how many different categories there are. The statistics of the category information are based on the previous classification work of the tables, and each table has one or more category labels that describe its application field. According to the number of types of different category information counted, a corresponding number of table retrieval libraries can be established. Each retrieval library corresponds to a specific category information and is used to store all tables under this category. Doing so can ensure that the system can effectively manage and retrieve table data of a specific category, thereby improving the efficiency and accuracy of retrieval.

[0088] Optionally, through classified storage, a large amount of table data can be organized according to application fields to form a special search library. In this way, when a user asks a question or search request in a specific field, he can directly access the table search library corresponding to the category information, avoiding indiscriminate search in all table data, greatly improving the search speed and efficiency.

[0089] As an optional implementation, the table search method further includes: splitting the merged cells in the multiple tables into independent cells respectively; and converting the content cells of the independent cells included in the multiple tables into text form.

[0090] In this embodiment, merged cells are common in tables, and merged cells are used to represent cell information across columns or rows. However, when performing retrieval and data analysis, merged cells may cause missing or confusion of data items, affecting the efficiency and accuracy of retrieval. Therefore, splitting the merged cells into independent cells can ensure that each data item exists independently, which is convenient for subsequent retrieval and processing.

[0091] For example, by analyzing the table structure, all merged cells are identified and split. For merged cells across columns, their contents can be copied to all merged columns; for merged cells across rows, their contents will be copied to all merged rows. In this way, each cell contains independent and complete information, which is convenient for subsequent text conversion and retrieval operations.

[0092] Alternatively, tabular data is usually stored in a table-specific format, for example, in Excel or a database. However, natural language processing technology or text retrieval algorithms are better at processing plain text data. Therefore, the table content can be converted into text form to ensure that all data can be correctly processed by the retrieval algorithm, thereby improving the comprehensiveness and efficiency of the retrieval.

[0093] For example, the content of each independent cell is read out and converted into a text string according to a certain format. For example, for a table containing column names and corresponding data, the system may generate a text format such as "column name 1: value 1, column name 2: value 2, column name 3: value 3..." In this way, each row of data in the table is converted into an independent text row, which is convenient for subsequent text matching and similarity calculation.

[0094] Optionally, the above preprocessing steps not only solve the data processing problem caused by merging cells, but also ensure that the table data exists in a unified text form, providing a standardized data format for subsequent retrieval steps such as intent recognition, fuzzy matching, and vector similarity calculation, thereby improving the performance and user experience of the entire table retrieval method. These preprocessing operations are indispensable technical means in modern table retrieval and data management.

[0095] The technical solution of the embodiment of the present invention is illustrated below in conjunction with preferred implementation modes.

[0096] At present, when using artificial intelligence technology for table retrieval, it mainly relies on converting natural language queries into structured query statements (such as SQL queries) or determining target slots based on semantic similarity. These methods have improved the retrieval efficiency and accuracy of table data to a certain extent. However, in practical applications, due to the complexity of natural language understanding and the uncertainty of table data quality, traditional methods face many challenges. Specifically, complex query statements or professional terms in specific fields may be beyond the full understanding of current natural language processing technology, resulting in inaccurate retrieval results; at the same time, errors or incompleteness of table data will also directly affect the retrieval effect, especially when converting natural language queries into structured query statements such as SQL. If the conversion is inaccurate, the query statement may not be executed correctly, thereby affecting the answer output. In addition, the existing technology also has certain technical difficulties in dealing with merged cells and the diversity of table formats, which affects the retrieval and analysis of data.

[0097] In response to the above technical problems, this application proposes a table content retrieval method based on intent recognition and similarity retrieval, aiming to improve the robustness and operability of table retrieval. By analyzing the intention of the user's question, the category of the required table can be more accurately located, and then the fuzzy matching and vector similarity methods are used to determine the target table and its internal target rows step by step to achieve effective retrieval and question-answering of the table content. This method can not only overcome the limitations of the existing technology in semantic understanding and data quality, but also deal with the problems of merged cells and format preprocessing to ensure the accuracy and efficiency of the retrieval process. A more intelligent and user-friendly table retrieval technology can be implemented to meet the growing needs of data retrieval and processing, especially in application scenarios such as intelligent customer service and digital healthcare, to provide more accurate and timely question-and-answer services.

[0098] Figure 2 FIG. 1 is a flow chart of a table content retrieval method based on intent recognition and similarity retrieval according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0099] Step S201 , hierarchically classify the table to be searched and add description text.

[0100] In this embodiment, for all the tables to be retrieved, hierarchical classification is performed according to the application field segmentation to determine the category to which each table belongs. For example, when handling customer consultations, the intelligent customer service system usually performs field segmentation according to the nature and purpose of the problem (consultation: product information, service content, operation instructions, promotional activities, purchase process; complaint: product quality, service problems, logistics distribution, after-sales experience; repair: fault diagnosis, troubleshooting, repair service, parts replacement; maintenance: maintenance consultation, warranty service, warranty status inquiry; account management: account problems, personal information; etc.).

[0101] Optionally, add description text for all tables to be searched. For each table, add description text based on the content it contains, briefly summarize the core content, enrich the table information, support the subsequent search process, ensure that the target table can be located more accurately, and improve the search efficiency.

[0102] Step S202, summarizing the classification, description text and other information of all tables to form a general table.

[0103] In this embodiment, the basic information of all the tables to be searched is summarized to form a comprehensive master table, which is used as a search library for the process of determining the target table. The master table integrates the information obtained in the above steps, including the table name, category, and description text of each table to be searched.

[0104] Step S203: perform format preprocessing on the table to be searched and the total table.

[0105] In this embodiment, preprocessing is performed for all tables to be retrieved. First, the merged cells are split to ensure that each data item in the table can be displayed independently. Then, it is converted into a unified text form according to a specific format. Each table is processed according to the following specific format so that each row of content is converted into row text:

[0106] Column name 1: value 11, column name 2: value 12, column name 3: value 13, ...;

[0107] Column name 1: value 21, column name 2: value 22, column name 3: value 23, ...;

[0108] …

[0109] This preprocessing method can ensure the consistency and searchability of table data, thereby improving retrieval accuracy.

[0110] Step S204: obtaining the search statement input by the user.

[0111] In this embodiment, a search sentence input by a user is obtained.

[0112] Step S205: perform intent recognition on the user's search sentence and infer the category to which it belongs.

[0113] In this embodiment, the user's search sentence is obtained, and the intention recognition method is used to infer which sub-category of the current application field the user's intention belongs to. One method of intention recognition is to use the natural language processing capability of a large language model, take the pre-determined field sub-category and user search sentence as input, control through prompt words, and infer the user's intention. In addition, a pre-trained intention recognition model can also be used to analyze user intentions, and the specific method of intention recognition is not limited here.

[0114] Optionally, after performing intent recognition, an intent recognition result may be obtained, and if the intent recognition is successful, step S206 is executed.

[0115] Step S206: Use the table corresponding to the category to which the user's intention belongs in the total table as a table search library.

[0116] In this embodiment, if the intention recognition is successful, the row text under the corresponding category of the general table is used as the table search library. If the intention recognition fails, step S207 is executed.

[0117] Step S207: if the intent recognition fails, all tables in the total table are used as the search library.

[0118] In this embodiment, if the intention recognition fails, all the rows of text in the total table are used as the table retrieval library.

[0119] Step S208: using a fuzzy matching method to determine candidate tables in the table search library that meet a matching degree threshold.

[0120] In this embodiment, the fuzzy matching method is used to calculate the matching degree between the user's search statement and the row text in the table search library, and the row text in the table search library that meets the preset matching degree threshold with the user's search statement is obtained as the candidate row text, and the corresponding table is used as the candidate table. If no row text that meets the threshold is retrieved, step S209 is executed. Then all the row texts in the table search library are used as candidate row texts, and the corresponding tables are used as candidate tables.

[0121] Optionally, the fuzzy matching method refers to a technology that recognizes and processes information by calculating the similarity between character strings under incomplete or inaccurate data conditions, which can help users find relevant search results even when there are spelling errors or ambiguities in the input, thereby improving the user experience. The core of fuzzy matching technology includes a variety of algorithms, such as Levenshtein distance, Jaccard similarity coefficient, etc., which can quantify the similarity between two character strings. For example, Levenshtein distance measures the similarity of two character strings by calculating the minimum number of editing operations required to convert one character string into another. The specific method is not limited here.

[0122] Step S209: if no relevant form is found, a table list is output for the user to select a target form.

[0123] In this embodiment, if no row text meeting the threshold is found, all row texts in the table search library are used as candidate row texts, and their corresponding tables are used as candidate tables.

[0124] Step S210 , using a vector similarity method, determining Top_k target tables that meet a similarity threshold from the candidate tables.

[0125] In this embodiment, the similarity between the user's query statement and the candidate line text is calculated using a vector similarity method. First, the user's query statement and the candidate line text are vectorized, and then their vector similarity is calculated to obtain a preset number of Top_k line texts that meet the preset similarity threshold as the target line text, and the corresponding table is used as the target table. If no line text meeting the threshold is retrieved, step S211 is executed.

[0126] Step S211, obtaining all the rows of text in the target table, and using a fuzzy matching method to determine candidate rows of text that meet a matching threshold.

[0127] In this embodiment, if no row text meeting the threshold is retrieved, all candidate row texts are pushed to the user for the user to select a target row text, and the corresponding table is used as the target table.

[0128] Step S212: using a vector similarity method, determine the Top_n target lines of text that meet a similarity threshold from the candidate lines of text.

[0129] In this embodiment, the row texts contained in the target table are obtained as the row search library. The row texts in the row search library that meet the preset matching threshold with the user's search statement are obtained by using the fuzzy matching method as candidate row texts. If no row text meeting the threshold is retrieved, all the row texts in the row search library are used as candidate row texts.

[0130] Optionally, the similarity between the user's query statement and the candidate line text is calculated using a vector similarity method. First, the user's query statement and the candidate line text are vectorized, and then their vector similarity is calculated to obtain a preset number of Top_n line texts that meet a preset similarity threshold as target line texts. If no line text meeting the threshold is retrieved, step S213 is executed.

[0131] Step S213: If no relevant row is found, all row texts of the target table are used as target row texts.

[0132] In this embodiment, if no line text satisfying the threshold is retrieved, all candidate line texts are taken as target line texts.

[0133] Step S214, obtain the target line text and output it directly to the user / provide it to the generated model user to output an answer.

[0134] In this embodiment, the retrieved target line text is obtained and can be directly pushed to the user as an answer; or it can be used as a reference and provided to the large model together with the user's search statement to extract the answer and push it to the user.

[0135] Through the above steps, a table retrieval process is completed. The target table and the target row in it are determined step by step by using the similarity retrieval method, and effective retrieval and question-answering for the table content are achieved. This method can better understand the user's intention, improve the robustness of the retrieval results, and achieve more accurate question-answering results.

[0136] According to an embodiment of the present invention, a table retrieval device is also provided. It should be noted that the table retrieval device can be used to execute the table retrieval method in the embodiment.

[0137] Figure 3 FIG. 1 is a schematic diagram of a table retrieval device according to an embodiment of the present invention. Figure 3 As shown, the table retrieval device 300 may include: an acquisition unit 301 , a first determination unit 302 , a retrieval unit 303 , a second determination unit 304 and a push unit 305 .

[0138] The acquisition unit 301 is used to acquire a table search request of a target object, wherein the table search request includes category information to which the table to be searched requested by the target object belongs.

[0139] The first determining unit 302 is configured to determine, in response to a table search request, a target table search library having the same category information as the table to be searched, wherein the target table search library includes a plurality of tables having the same category information as the table to be searched.

[0140] The retrieval unit 303 is configured to retrieve at least one candidate table from the target table retrieval library, wherein a first matching degree between table description information of the candidate table and the table retrieval request satisfies a first matching degree threshold.

[0141] The second determining unit 304 is configured to determine a target table from at least one candidate table based on table description information corresponding to at least one candidate table, wherein a first similarity between the table description information corresponding to the target table and the table search request satisfies a first similarity threshold.

[0142] The pushing unit 305 is used to push the target table to the target object.

[0143] Optionally, the retrieval device of the table is also used to: obtain the row text contained in the target table; use the fuzzy matching method to select from the row text the row text whose second matching degree with the table retrieval request satisfies a second matching threshold, and obtain candidate row text; calculate the second similarity between the table retrieval request and the candidate row text, wherein the second similarity is used to characterize the degree of similarity between the table retrieval request and the candidate row text; based on the second similarity and the second similarity threshold, select the target row text that satisfies the second similarity threshold with the table retrieval request from the candidate row text; and push the target row text to the target object.

[0144] Optionally, the first determination unit 302 is also used to: respond to a table retrieval request, perform semantic recognition on the table retrieval request, and obtain category information to which the table to be retrieved belongs; based on the category information to which the table to be retrieved belongs, determine a target table retrieval library from multiple table retrieval libraries, wherein the multiple table retrieval libraries correspond to different category information.

[0145] Optionally, the retrieval unit 303 is also used to: use a fuzzy matching method to respectively determine the first matching degrees between multiple table description information included in the target table retrieval library and the table retrieval request; respectively compare the first matching degrees between the multiple table description information and the table retrieval request with a first matching degree threshold to obtain a first comparison result; based on the first comparison result, retrieve at least one candidate table from the target table retrieval library.

[0146] Optionally, the second determination unit 304 is also used to: determine a first similarity between the table description information corresponding to at least one candidate table and the table retrieval request; compare the first similarity between the table description information corresponding to at least one candidate table and the table retrieval request with a first similarity threshold to obtain a second comparison result; and determine a target table from at least one candidate table based on the second comparison result.

[0147] Optionally, the table retrieval device is further used to: obtain multiple tables; and divide the multiple tables into multiple categories according to application fields, wherein tables of different categories correspond to different category information.

[0148] Optionally, the table retrieval device is further used to: add table description information to the multiple tables respectively according to the key information contained in the multiple tables, wherein the table description information is used to describe the key information of the corresponding table.

[0149] Optionally, the table retrieval device is further used to: establish the same number of table retrieval libraries according to the number of category information corresponding to the multiple tables; and store the tables with the same category information in one table database.

[0150] Optionally, the table retrieval device is further used to: split the merged cells in the multiple tables into independent cells respectively; and convert the content cells of the independent cells included in the multiple tables into text form.

[0151] In the device, the category information of the table to be retrieved included in the table retrieval request of the target object can be used to quickly locate the table retrieval library related to it, avoiding blind search in a large and diverse table collection, greatly shortening the retrieval time, and improving the user experience. Moreover, through step-by-step retrieval, potential candidate tables are preferentially screened out, and then the target table is further confirmed, avoiding indiscriminate similarity calculations for all tables, effectively reducing the consumption of computing resources and improving system performance. Furthermore, by adopting the dual threshold mechanism of the first matching degree and the first similarity, the table that truly matches the retrieval request can be screened out, and even in the face of ambiguous or complex query statements, accurate retrieval results can be provided, enhancing the robustness and reliability of the system, achieving the purpose of accurately identifying the target table that matches the table retrieval request of the target object, and then achieving the technical effect of improving the accuracy of table retrieval, and solving the technical problem of low table retrieval accuracy.

[0152] An embodiment of the present application further provides an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein when the program is running, the method for determining the off-site pickup mode in each embodiment of the present invention is executed.

[0153] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the table retrieval method in each embodiment of the present invention.

[0154] An embodiment of the present application further provides a computer program product, including a computer program, which implements the table retrieval method in each embodiment of the present invention when executed by a processor.

[0155] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the table retrieval method in each embodiment of the present invention is implemented.

[0156] The embodiments of the present application further provide a computer program, which, when executed by a processor, implements the table retrieval method in each of the embodiments of the present invention described above.

[0157] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0158] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0160] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0161] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0162] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, Read-Only Memory (ROM), Random Access Memory (RAM), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program codes.

[0163] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A table retrieval method, characterized in that: include: Obtaining a table search request of a target object, wherein the table search request includes category information to which the table to be searched requested by the target object belongs; In response to the table search request, determining a target table search library having the same category information as the table to be searched, wherein the target table search library includes a plurality of tables having the same category information as the table to be searched; Retrieving at least one candidate table from the target table retrieval library, wherein a first matching degree between table description information of the candidate table and the table retrieval request satisfies a first matching degree threshold; Based on the table description information corresponding to at least one of the candidate tables, determining a target table from at least one of the candidate tables, wherein a first similarity between the table description information corresponding to the target table and the table retrieval request satisfies a first similarity threshold; The target table is pushed to the target object.

2. The method according to claim 1, characterized in that The method further comprises: Get the row text contained in the target table; Using a fuzzy matching method, selecting from the line texts a line text whose second matching degree with the table search request satisfies a second matching degree threshold, to obtain a candidate line text; Calculating a second similarity between the table retrieval request and the candidate row text, wherein the second similarity is used to represent a degree of similarity between the table retrieval request and the candidate row text; Based on the second similarity and the second similarity threshold, selecting a target line of text that satisfies the second similarity threshold with the table search request from the candidate lines of text; The target line of text is pushed to the target object.

3. The method according to claim 1, characterized in that: In response to the table search request, determining a target table search library having the same category information as the table to be searched, comprising: In response to the table search request, semantic recognition is performed on the table search request to obtain category information to which the table to be searched belongs; Based on the category information to which the to-be-searched table belongs, the target table search library is determined from a plurality of table search libraries, wherein the plurality of table search libraries correspond to different category information.

4. The method according to claim 1, characterized in that Retrieving at least one candidate form from the target form retrieval library includes: Determine the first matching degree between the plurality of table description information included in the target table retrieval library and the table retrieval request respectively by using a fuzzy matching method; Comparing first matching degrees between the plurality of table description information and the table search request with the first matching degree threshold respectively to obtain a first comparison result; Based on the first comparison result, at least one candidate table is retrieved from the target table retrieval library.

5. The method according to claim 1, characterized in that Determining a target table from at least one of the candidate tables based on table description information corresponding to at least one of the candidate tables includes: determining the first similarity between the table description information corresponding to at least one of the candidate tables and the table retrieval request; respectively comparing the first similarity between the table description information corresponding to the at least one candidate table and the table search request with the first similarity threshold to obtain a second comparison result; Based on the second comparison result, the target table is determined from at least one of the candidate tables.

6. The method according to claim 1, characterized in that The method further comprises: Get multiple tables; The multiple tables are divided into multiple categories according to application fields, wherein tables of different categories correspond to different category information.

7. The method according to claim 6, characterized in that The method further comprises: According to the key information included in the multiple tables, table description information is added to the multiple tables respectively, wherein the table description information is used to describe the key information of the corresponding table.

8. The method according to any one of claims 6 to 7, characterized in that The method further comprises: According to the number of category information corresponding to the plurality of tables, establishing the same number of table retrieval libraries; Store tables with the same category information in one table database.

9. The method according to claim 6, characterized in that The method further comprises: Splitting the merged cells in the multiple tables into independent cells respectively; The content cells of each of the independent cells included in the plurality of tables are converted into a text form.

10. A table retrieval device, characterized in that: include: An acquisition unit, configured to acquire a table search request of a target object, wherein the table search request includes category information to which the table to be searched requested by the target object belongs; A first determining unit is used to determine, in response to the table search request, a target table search library having the same category information as that of the table to be searched, wherein the target table search library includes a plurality of tables having the same category information as that of the table to be searched; A retrieval unit, configured to retrieve at least one candidate table from the target table retrieval library, wherein a first matching degree between table description information of the candidate table and the table retrieval request satisfies a first matching degree threshold; A second determining unit, configured to determine a target table from at least one of the candidate tables based on table description information corresponding to at least one of the candidate tables, wherein a first similarity between the table description information corresponding to the target table and the table retrieval request satisfies a first similarity threshold; A pushing unit is used to push the target table to the target object.

11. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 9 when running.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 9.

13. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images