Multi-form processing method and device, electronic equipment and storage medium

By analyzing the tables entered by the user and recalling and interpreting the target form, the problem of time-consuming LSTM network and increasing complexity of entity extraction models is solved, and the efficiency of table analysis and chart drawing is improved.

CN119089876APending Publication Date: 2024-12-06NANJING LINGXING TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411177246.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

During human-computer interaction, since the LSTM network is time-consuming and cannot be processed in parallel, the table analysis efficiency is low, and additional models of entity extraction classes are required for table retrieval, which increases complexity and reduces the drawing effect.

Method used

By analyzing the tables entered by the user, determining the target form related to the problem, recalling the target form using cosine similarity, and interpreting it, generating structured query language or drawing code to improve the efficiency of table analysis and chart drawing.

Benefits of technology

It reduces the complexity of table analysis, improves analysis efficiency, and reduces the complexity of chart drawing, and improves the overall table processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089876A_ABST
    Figure CN119089876A_ABST
Patent Text Reader

Abstract

The invention provides a multi-form processing method and device, electronic equipment and a storage medium, and the method comprises the steps: analyzing forms in a form input by a user, and obtaining form names and column name text information of each form; determining a target form related to the question based on the question input by the user; paraphrasing column name text information of the target form and proper nouns in the question to obtain first paraphrased knowledge; when the first paraphrasing knowledge can solve the problem input by the user, if it is determined that the text intention of the user is a table analysis problem, generating a structured query language corresponding to data analysis based on the target form and the first paraphrasing knowledge; and executing the structured query language to obtain a corresponding table analysis result. According to the table problem in the man-machine interaction process, the target form related to the user problem is recalled, the model focuses on key tables and problems, redundant information is reduced, the complexity of table analysis is reduced, and the analysis efficiency of the table problem is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multi-form processing method, device, electronic equipment and storage medium. Background Art

[0002] In daily life and work, many scenarios require analysis and processing of tables.

[0003] In the related technology, during the human-computer interaction process, table analysis can help users automatically generate SQL queries through tool platforms (such as text2sql algorithm, etc.); it can also use long short-term memory models (LSTM, long-short termmemory) to predict the FROM clause of structured query language (SQL) statements; or based on the multi-table situation, an entity extraction model can be supplemented to assist in retrieving the table corresponding to the question.

[0004] However, in the above method, during the human-computer interaction process, the LSTM network is very time-consuming and cannot be processed in parallel, which reduces the efficiency of table analysis problems. In addition, an additional entity extraction model is required for table retrieval, which not only increases the complexity of table retrieval, but also reduces the drawing effect of the table.

[0005] Therefore, in the process of human-computer interaction, how to improve the analysis efficiency of tables and the poor effect of chart drawing is a technical problem that needs to be solved. Summary of the invention

[0006] The present invention provides a multi-form processing method, device, electronic device and computer-readable storage medium, so as to at least solve the problem in the related art that in the process of human-computer interaction, due to the time-consuming LSTM network and the inability to perform parallel processing, an entity extraction model needs to be added for table retrieval, resulting in reduced efficiency of table analysis or chart drawing. The technical solution of the present invention is as follows:

[0007] According to a first aspect of an embodiment of the present invention, a multi-form processing method is provided, comprising:

[0008] Parse the form in the user input table to obtain the form name and column name text information of each form;

[0009] Based on the question input by the user, determining a target form related to the question;

[0010] Interpreting the column name text information of the target form and the proper noun in the question to obtain first interpretation knowledge;

[0011] When the first interpretation knowledge can solve the problem input by the user, if it is determined that the text intention of the user is to analyze a form problem, then a structured query language for corresponding data analysis is generated based on the target form and the first interpretation knowledge;

[0012] Execute the structured query language to obtain corresponding table analysis results.

[0013] Optionally, determining a target form related to the question based on the question input by the user includes:

[0014] Based on the question input by the user, determining the cosine similarity of the forms related to the question;

[0015] A target form related to the question is recalled based on the cosine similarity.

[0016] Optionally, determining the cosine similarity of forms related to the question based on the question input by the user includes:

[0017] Mapping the text of the question input by the user into a first low-dimensional dense vector;

[0018] Extract the form name of each form and the column name text information of the corresponding form;

[0019] splicing the form name of each form and the corresponding column name text information according to a standard format to obtain a spliced ​​text of each form;

[0020] Mapping the concatenated text of each form into a second low-dimensional dense vector;

[0021] Based on the first low-dimensional dense vector and the second low-dimensional dense vector, the cosine similarity between the question and each form is calculated.

[0022] Optionally, recalling a target form related to the question based on the cosine similarity includes:

[0023] Sort all forms by cosine similarity from high to low;

[0024] The first n ranked forms are selected for recall to obtain a target form, where n is a natural number greater than or equal to 1.

[0025] Optionally, interpreting the column names of the target form and the proper nouns in the question to obtain first interpretation knowledge includes:

[0026] Load the content of the target form into dataframe or polardb format;

[0027] Extract the column names of all target tables after loading;

[0028] By using the prompt word engineering in the large language model, the column names of all target forms and the professional names appearing in the questions input by the user are interpreted to obtain the first interpretation knowledge.

[0029] Optionally, the method further includes:

[0030] When the first interpretation knowledge can solve the problem input by the user, if it is determined that the text intention of the user is to draw a chart, then based on the target form and the first interpretation knowledge, a corresponding pandas drawing code is generated;

[0031] The corresponding pandas drawing code is executed through the Python interpreter to obtain the corresponding chart.

[0032] Optionally, the method further includes:

[0033] When the first interpretation knowledge cannot solve the question raised by the user, determining uncertain items in the question;

[0034] Interacting with the user based on the uncertain entry, updating the first interpretation knowledge according to the interaction result, and obtaining the second interpretation knowledge;

[0035] When it is determined that the user's text intention is to analyze a form question, generating a structured query language corresponding to data analysis based on the target form and the second interpretation knowledge;

[0036] Execute the structured query language to obtain corresponding table analysis results.

[0037] Optionally, the method further includes:

[0038] When it is determined that the user's text intention is to draw a chart, a corresponding pandas drawing code is generated based on the target form and the second interpretation knowledge;

[0039] The corresponding pandas drawing code is executed through the Python interpreter to obtain the corresponding chart.

[0040] According to a second aspect of an embodiment of the present invention, there is provided a multi-form processing device, comprising:

[0041] The parsing module is used to parse the form in the user input table to obtain the form name and the column name text information of each form;

[0042] A determination module, used for determining a target form related to the question based on the question input by the user;

[0043] A first interpretation module, used for interpreting the column name text information of the target form and the proper noun in the question to obtain first interpretation knowledge;

[0044] A first generating module is used for generating a structured query language for corresponding data analysis based on the target form and the first interpretive knowledge when the first interpretive knowledge can solve the problem input by the user and if it is determined that the text intention of the user is to analyze a form problem;

[0045] The first execution module is used to execute the structured query language to obtain corresponding table analysis results.

[0046] Optionally, the determining module includes:

[0047] A similarity determination module, configured to determine, based on the question input by the user, the cosine similarity of the forms related to the question;

[0048] A recall module is used to recall target forms related to the question based on the cosine similarity.

[0049] Optionally, the similarity determination module includes:

[0050] A first mapping module, configured to map the text of the question input by the user into a first low-dimensional dense vector;

[0051] The first extraction module is used to extract the form name of each form and the column name text information of the corresponding form;

[0052] A splicing module, used for splicing the form name of each form and the corresponding column name text information according to a standard format to obtain the spliced ​​text of each form;

[0053] A second mapping module, used for mapping the concatenated text of each form into a second low-dimensional dense vector;

[0054] A similarity calculation module is used to calculate the cosine similarity between the question and each form based on the first low-dimensional dense vector and the second low-dimensional dense vector.

[0055] Optionally, the recall module includes:

[0056] Sorting module, used to sort all forms from high to low according to cosine similarity;

[0057] The form recall module is used to select the first n forms in the sorting for recall to obtain a target form, wherein n is a natural number greater than or equal to 1.

[0058] Optionally, the first interpretation module includes:

[0059] A loading module, used to load the content of the target form into a dataframe or polardb format;

[0060] The second extraction module is used to extract the column names of all target tables after loading;

[0061] The first interpretation knowledge module is used to interpret the column names of all target forms and the professional names appearing in the questions input by the user by using the prompt word engineering in the large language model to obtain the first interpretation knowledge.

[0062] Optionally, the device further comprises:

[0063] A second generation module is used for generating corresponding pandas drawing code based on the target form and the first interpretation knowledge when the first interpretation knowledge can solve the problem input by the user and if it is determined that the text intention of the user is to draw a chart;

[0064] The second execution module is used to execute the corresponding pandas drawing code through the Python interpreter to obtain the corresponding chart.

[0065] Optionally, the device further comprises:

[0066] An item determination module, used for determining uncertain items in the question when the first interpretation knowledge cannot solve the question raised by the user;

[0067] An updating module, configured to interact with the user based on the uncertain item, and continuously update the first interpretation knowledge according to the interaction result to obtain the second interpretation knowledge;

[0068] A third generation module is used to generate a structured query language corresponding to data analysis based on the target form and the second interpretation knowledge when it is determined that the user's text intention is to analyze a form problem;

[0069] The third execution module is used to execute the structured query language to obtain corresponding table analysis results.

[0070] Optionally, the device further comprises:

[0071] A fourth generation module is used to generate corresponding pandas drawing code based on the target form and the second interpretation knowledge when it is determined that the user's text intention is to draw a chart;

[0072] The fourth execution module is used to execute the corresponding pandas drawing code through the Python interpreter to obtain a corresponding chart.

[0073] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, including:

[0074] processor;

[0075] a memory for storing instructions executable by the processor;

[0076] The processor is configured to execute the instructions to implement the multi-form processing method as described above.

[0077] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device can execute the multi-form processing method as described above.

[0078] According to a fifth aspect of an embodiment of the present invention, there is provided a computer program product, including a computer program or instructions, which implement the multi-form processing method as described above when executed by a processor of an electronic device.

[0079] The technical solution provided by the embodiments of the present invention brings at least the following beneficial effects:

[0080] In an embodiment of the present invention, the form in the user input form is parsed to obtain the form name and the column name text information of each form; based on the question input by the user, the target form related to the question is determined; the column name text information of the target form and the proper nouns in the question are interpreted to obtain the first interpretation knowledge; when the first interpretation knowledge can solve the question input by the user, if it is determined that the user's text intention is to analyze the table problem, then based on the target form and the first interpretation knowledge, a structured query language for corresponding data analysis is generated; the structured query language is executed to obtain the corresponding table analysis result. That is to say, in an embodiment of the present invention, for the table problem in the human-computer interaction process, by recalling the target form related to the user's question, the model focuses on key tables and questions, reduces redundant information, reduces the complexity of table analysis, and improves the analysis efficiency of table problems.

[0081] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention, and do not constitute an improper limitation of the present invention. In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0083] Figure 1 The figure is a flow chart of a multi-form processing method provided by an embodiment of the present invention.

[0084] Figure 2 It is an application flow chart of a multi-form processing method provided by an embodiment of the present invention.

[0085] Figure 3 It is a block diagram of a multi-table processing device provided by an embodiment of the present invention.

[0086] Figure 4 It is a block diagram of a determination module provided by an embodiment of the present invention.

[0087] Figure 5 It is a block diagram of a similarity determination module provided by an embodiment of the present invention.

[0088] Figure 6 It is a block diagram of a recall module provided in an embodiment of the present invention.

[0089] Figure 7 It is a block diagram of a first interpretation module provided by an embodiment of the present invention.

[0090] Figure 8 It is a block diagram of an electronic device provided by an embodiment of the present invention.

[0091] Fig. 9 It is a block diagram of a device for multi-table processing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0092] In order to enable ordinary persons in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings.

[0093] 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. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0094] Figure 1 is a flowchart of a multi-form processing method provided by an embodiment of the present invention. Figure 1 As shown, the multi-form processing method includes the following steps:

[0095] Step 101: Parse the form in the user input table to obtain the form name and column name text information of each form.

[0096] Step 102: Based on the question input by the user, determine a target form related to the question.

[0097] Step 103: interpreting the column names of the target form and the proper nouns in the question to obtain first interpretation knowledge.

[0098] Step 104: When the first interpretation knowledge can solve the problem input by the user, if it is determined that the user's text intention is to analyze a form problem, a structured query language for corresponding data analysis is generated based on the target form and the first interpretation knowledge.

[0099] Step 105: Execute the structured query language to obtain corresponding table analysis results.

[0100] The multi-form processing method described in the present invention can be applied to terminals, servers, etc., without limitation herein. The terminal implementation device can be electronic devices such as smart phones, laptops, tablet computers, desktop computers, personal digital assistants (PDAs) and wearable devices. The server can be an independent server or a server cluster, or a server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, intermediate services, domain name services, security services, content distribution networks, or big data and artificial intelligence platforms, etc., without limitation herein.

[0101] Combine the following Figure 1 , the specific implementation steps of a multi-form processing method provided by an embodiment of the present invention are described in detail.

[0102] In step 101, the form in the user input table is parsed to obtain the form name and column name text information of each form.

[0103] In this step, the user inputs a form through the client, that is, the server receives the form input by the user, and parses the form in the form to obtain the form name and the column name text information of each form. For example, 10 forms in the input form are parsed to obtain the form name (sheet name) and the column name text information of each form, such as column name 1, column name 2..., column name N, etc., where N is a natural number greater than or equal to 1.

[0104] Among them, in the embodiment of the present invention, the process of parsing the form is the process of loading or reading the form. For a standard form, it can be directly parsed through the standard form parsing process to obtain the form name and the column name text information of each form. If the parsing process reports an error (i.e., the parsing fails), it means that the form uploaded by the user is a non-standard form.

[0105] For non-standard tables, if the file is not damaged at the physical level, the error will be told to the large language model (LLM) through the prompt word project, allowing the large language model to generate parsing code step by step according to the error for background parsing to solve the parsing of non-standard format tables. Through parsing, the form name of the non-standard form and the column name text information of each form can be obtained.

[0106] The form in this embodiment is an area containing form elements, wherein the form elements are elements that allow users to input information in the form (such as text fields, drop-down lists, radio buttons, check boxes, etc.).

[0107] It should be noted that in this embodiment, the forms uploaded by the user include but are not limited to: multiple forms, a single form, multiple forms with multiple tables, a single form with multiple forms, etc. After receiving the form, the server parses each form in the form and can store it in a dictionary form.

[0108] In step 102, based on the question input by the user, a target form related to the question is determined.

[0109] The step specifically includes:

[0110] 1) Based on the question input by the user, determine the cosine similarity of the forms related to the question.

[0111] In this step, the text of the question input by the user is mapped to a first low-dimensional dense vector; the form name of each form and the column name text information of the corresponding form are extracted; the form name of each form and the corresponding column name text information are spliced ​​according to a standard format to obtain the spliced ​​text of each form; the spliced ​​text of each form is mapped to a second low-dimensional dense vector; based on the first low-dimensional dense vector and the second low-dimensional dense vector, the cosine similarity between the question and each form is calculated.

[0112] That is, in the embodiment of the present invention, the user question and the text information of the form name and column name of each form are concatenated, and the concatenation results are respectively used through a text encoding (embedding) model to obtain corresponding text embeddings, and the cosine similarity is calculated for each pair. Specifically, it includes:

[0113] Based on the parsed table data, combined with the query given by the user, firstly, the form is recalled according to the user's question to determine which form is related to the question. Then, a text encoding (embedding) model (including but not limited to BGE, GTE, M3E, etc.) is used to map the text of the user's question (query) into a low-dimensional dense vector (called the first low-dimensional dense vector), and x is used to represent the first low-dimensional dense vector. q Then, the form name of each form and the text information of the column name of the corresponding form are extracted, and they can be spliced ​​according to the standard format to obtain the spliced ​​text of each form. The spliced ​​text of each form is also mapped into a low-order dense vector (called the second low-order dense vector) through the text encoding (embedding) model, and respectively used Represented; then, the target form is recalled by calculating the similarity between the text embedding of the question and the text embedding of the form concatenation. Specifically, based on the first low-dimensional dense vector and the second low-dimensional dense vector, the cosine similarity between the question and each form is calculated. Specifically, the first low-dimensional dense vector x is calculated. q and the second least dense vector The inner product of can be used to obtain the cosine similarity, as shown in the following formula. In the following formula, the left side of the equal sign is the cosine similarity, and the right side of the equal sign is the first low-order dense vector x q and the second least dense vector The ratio of the inner product to the modulus.

[0114]

[0115] Among them, x q is the first low-order dense vector, is the second least dense vector.

[0116] 2) Recalling target forms related to the question based on the cosine similarity.

[0117] In this step, all forms are sorted from high to low according to cosine similarity; the top n forms in the sorting are selected for recall to obtain a target form, wherein n is a natural number greater than or equal to 1.

[0118] That is, all the calculated cos similarity values ​​are sorted from high to low, and the top n forms are selected for recall to obtain the target form. Of course, the recalled forms can also be formed into a table, that is, the pre-parsed target table is obtained. In this embodiment, for different scenarios, the value of n is also different, and this embodiment does not limit it.

[0119] In step 103, the column names of the target form and the proper nouns in the question are interpreted to obtain first interpretation knowledge.

[0120] In this step, the content of the target form is loaded into a dataframe or polardb format; the column names of all the target forms after loading are extracted; and the prompt word engineering in the large language model is used to interpret the column names of all the target forms and the professional names appearing in the questions input by the user to obtain the first interpretation knowledge, wherein the first interpretation knowledge is called knowledge q.

[0121] Among them, in this embodiment, the content of the target form is read and loaded into a dataframe or polardb format. All column names of the loaded target form and the corresponding 10 rows (take this as an example) of sample data are selected, and relevant prompt words are written through the prompt word engineering of the large language model, and user questions are spliced, so that the large language model can parse non-Chinese column names (for example, it can be non-Chinese, Chinese with English, or Chinese with professional terms, etc.), and interpret some proper nouns in the non-Chinese column names and user questions, and finally output the interpreted knowledge q.

[0122] It should be noted that, in this embodiment, since the input context length of the large language model is not infinite, a memory maintenance module is proposed in this embodiment. Through the memory maintenance module, important information such as relevant knowledge, dialogue context, etc. in the target form or table can be maintained.

[0123] In step 104, when the first interpretation knowledge can solve the problem input by the user, if it is determined that the user's text intention is to analyze a form problem, a structured query language for corresponding data analysis is generated based on the target form and the first interpretation knowledge.

[0124] In this step, the first interpretation knowledge q after the large language model's interpretation is the result of the large language model's understanding based on the input context, and is a guess based on the knowledge of the large language model itself. In practical applications, the large language model may not be clear about the business terms and jargon of many scenarios, so after obtaining the knowledge q, the large language model first stores the knowledge q in the memory maintenance module.

[0125] There are many kinds of questions input by users. For normal table questions, they are divided into two intentions: table data analysis and chart drawing. Abnormal table questions are uniformly divided into small talk intentions. Therefore, the embodiment of the present invention describes the two intentions of table data analysis and chart drawing.

[0126] Usually, a large language model is needed to determine whether the user's intention is to analyze a table, draw a chart, or chat based on the question input by the user. This is very important for user experience. Similarly, this embodiment will also write prompt words based on the prompt word project in the large language model to allow the large language model to identify the relevant intentions.

[0127] In this embodiment, different process execution methods are provided for different intentions. If it is determined that the user's intention is a table analysis question, a structured query language (SQL) for corresponding data analysis is generated based on the target form and the first interpretation knowledge. In other words, the branch of generating SQL is triggered, that is, an SQL statement corresponding to the result is generated according to the target form content and the user's question (query), and then the original table data and the generated SQL are input, and the corresponding SQL statement is executed using related library functions such as pandassql to obtain the final result.

[0128] In step 105, the structured query language is executed to obtain corresponding table analysis results.

[0129] In this step, if the final result is a numerical value, the large language model will rewrite the numerical value into a complete reply sentence according to the user's question and output it. For example, if the user asks "how many of something are counted", after the above series of operations, the result is a numerical value of 50, then the large language model will rewrite the numerical value into "the number of something is 50" according to the user's question. Otherwise, if the final result is a table, the corresponding table result will be directly output.

[0130] In an embodiment of the present invention, the form in the user input form is parsed to obtain the form name and the column name text information of each form; based on the question input by the user, the target form related to the question is determined; the column name text information of the target form and the proper nouns in the question are interpreted to obtain the first interpretation knowledge; when the first interpretation knowledge can solve the question input by the user, if it is determined that the user's text intention is to analyze the table problem, then based on the target form and the first interpretation knowledge, a structured query language for corresponding data analysis is generated; the structured query language is executed to obtain the corresponding table analysis result. That is to say, in an embodiment of the present invention, for the table problem in the human-computer interaction process, by recalling the target form related to the user's question, the model focuses on key tables and questions, reduces redundant information, reduces the complexity of table analysis, and improves the analysis efficiency of table problems.

[0131] Optionally, in another embodiment, based on the above embodiment, the method may also include: when the first interpretation knowledge can solve the problem input by the user, if it is determined that the user's text intention is to draw a chart, then based on the target form and the first interpretation knowledge, generating a corresponding pandas drawing code; executing the corresponding pandas drawing code through the Python interpreter to obtain a corresponding chart.

[0132] That is to say, when the first interpretation knowledge can solve the problem input by the user, if it is determined that the user's intention is to draw a chart, this embodiment uses pandas to allow the large language model to generate corresponding pandas drawing code based on the existing information (i.e., the first interpretation knowledge), and then in the background, the corresponding code is executed through the Python interpreter to obtain the final generated chart.

[0133] The embodiment of the present invention is based on a large language model and does not require fine-tuning. It solves the problem of users drawing charts by building an agent. Through form recall, the model focuses on key tables and questions and reduces redundant information. The complexity of drawing charts is reduced and the efficiency of drawing charts is improved.

[0134] Optionally, in another embodiment, based on the above embodiment, the method may also include: when the first interpretation knowledge cannot solve the question raised by the user, determining the uncertain items in the question; interacting with the user based on the uncertain items, updating the first interpretation knowledge according to the interaction results, and obtaining the second interpretation knowledge; when it is determined that the user's text intention is to analyze a table problem, generating a structured query language for corresponding data analysis based on the target form and the second interpretation knowledge Q; executing the structured query language to obtain corresponding table analysis results.

[0135] In an embodiment of the present invention, the large language model can be allowed to first determine whether the knowledge is sufficient to solve the question (query) raised by the user based on the sample data of 10 lines of the target form (this is taken as an example, and is not limited to this in actual applications) and the knowledge obtained by the large language model's own interpretation. For the question raised by the user, the large language model interprets whether there are uncertain items. If there are uncertain items, the user is asked related questions. Afterwards, the memory maintenance module will update and maintain the first interpretation knowledge q according to the user's return results, until all uncertain items are replied and explained by the user, and the interpretation knowledge maintained at this time is recorded as the second interpretation knowledge, represented by Q.

[0136] Afterwards, when it is determined that the user's text intention is to analyze a table problem, a structured query language for corresponding data analysis is generated based on the target form and the second interpretation knowledge Q; the structured query language is executed to obtain a corresponding table analysis result; or

[0137] When it is determined that the user's text intention is to draw a chart, a corresponding pandas drawing code is generated based on the target form and the second interpretation knowledge Q; the corresponding pandas drawing code is executed through the Python interpreter to obtain a corresponding chart.

[0138] The specific implementation process is similar to the above process of generating and executing the structured query language for corresponding data analysis or generating and executing the corresponding pandas drawing code based on the target form and the first interpretation knowledge q. For details, please see the above and will not be repeated here.

[0139] The embodiment of the present invention solves the problem of users analyzing tables and drawing charts by building an agent based on a large language model. By recalling forms related to the problem, the large language model focuses on key forms and problems and reduces redundant information. A memory maintenance module is further introduced to store information about human-computer interaction based on problems and form knowledge, solve the context length limit of the large language model, and correct and improve the existing knowledge through multiple rounds of interaction between the large language model and the user, so as to solve the user's current problem in a more targeted manner.

[0140] See also Figure 2 , which is an application example diagram of a multi-form processing method provided by an embodiment of the present invention, such as Figure 2 As shown, the embodiment of the present invention involves table parsing, table recall, memory maintenance, intention judgment and execution. The following describes them separately:

[0141] 1) Table analysis

[0142] In this embodiment, at the beginning of the process, the server obtains the form uploaded by the user (i.e., the user form in the figure), which may include but is not limited to: multiple tables, a single table, multiple tables and multiple forms, a single table and multiple forms, and the like. For example, sheet1, sheet2, ..., sheetn in the figure. And the form of each table is parsed out and stored in the memory maintenance (or memory maintenance module) in the form of a dictionary. It should be noted that for standard forms, this embodiment directly parses through the standard form parsing process to obtain the form name and the column name text information of each form. This parsing process is familiar to those skilled in the art and is not repeated here. If the parsing process reports an error, that is, the parsing fails, it means that the form uploaded by the user is a non-standard form, such as non-physical file damage, etc. Based on this, this embodiment will tell the error to the large language model through the prompt word project, so that the large language model can generate parsing code step by step according to the error for background parsing to solve the parsing of non-standard format tables. Through parsing, the form name of the non-standard form and the column name text information of each form can be obtained.

[0143] 2) Form recall

[0144] Based on the parsed table data and the query question given by the user, firstly, the form is recalled according to the user's question to determine which form is related to the question. The specific determination process is described above and will not be repeated here.

[0145] Read and load the content of the recalled target form into dataframe or polardb format. Select all column names and the corresponding 10 rows of sample data, write relevant prompt words through the prompt word project of the large language model, splice user questions, let the large model parse non-Chinese column names, and interpret some proper nouns in non-Chinese column names and user questions, and finally output the interpreted knowledge q, which is called the first interpretation knowledge q.

[0146] 3) Memory maintenance

[0147] In this step, since the input context length of the large language model is not infinite, this embodiment provides memory maintenance, which is to maintain important information such as table-related knowledge, dialogue context, etc., which can also be called a memory maintenance module, etc.

[0148] The first interpretation knowledge q after the interpretation of the large semantic model is the result of the model's understanding based on the input context. It is a guess based on the knowledge of the model itself. In practical applications, the business terms and jargon models of many scenarios are not necessarily clear. Therefore, after the model obtains the first interpretation knowledge q in the first step, it stores q in the memory maintenance module.

[0149] After that, let the large language model judge whether the knowledge is sufficient to solve the query raised by the user based on the 10 rows of sample data in the target table and the knowledge obtained by the large model's own interpretation, and whether there are uncertain items in the model's own interpretation. If there are such uncertain items, ask the user related questions. The memory maintenance module will continuously update the maintained q according to the user's response results until all uncertain items are replied and explained by the user. The maintained knowledge at this time is recorded as the second interpretation knowledge, represented by capital Q.

[0150] 4) Intention Judgment

[0151] There are many kinds of questions asked by users. Normal table questions are divided into two intents: table data analysis and chart drawing. Abnormal table questions are uniformly divided into chat intent. First, a large language model is required to determine whether the user's intention is to analyze table questions, draw charts, or chat based on the user's input questions. Based on the above problems, this embodiment writes prompt words according to the prompt word project, allowing the large language model to identify related intents.

[0152] 5) Execute code generation

[0153] According to different intentions, this embodiment provides different flow execution methods. If it is a table analysis problem (i.e., the chart analysis in the figure), the processing method is to trigger the branch generated by SQL, generate an SQL statement to obtain the corresponding result based on the table content and the user query, and then input the original table data and the generated SQL, and use related library functions such as pandassql to execute the corresponding SQL statement to obtain the final result. If the final result is a numerical value, a large language model will usually rewrite the numerical value into a complete reply statement for output based on the user's question. Otherwise, the corresponding table result will be output directly.

[0154] If it is a chart drawing problem (i.e., the chart drawing in the figure), the solution is to use pandas to let the large language model generate the corresponding pandas drawing code based on the existing information (i.e., calling the knowledge in the memory maintenance module), and then execute the corresponding code through the Python interpreter to get the final generated chart.

[0155] It should be noted that in the embodiment of the present invention, no matter which intent is triggered, the corresponding code needs to be generated. When generating the code, the generation result of the large language model is not completely controllable and there will be jitter. Therefore, in this embodiment, in order to make the generated code executable, the SELECT statement and import statement are extracted by regular expressions to eliminate jitter. In addition, this embodiment also adds a retry mechanism to prevent the code from running incorrectly and resulting in failure to obtain results.

[0156] That is to say, in an embodiment of the present invention, after the user uploads a form, the server parses the form received in the form to obtain the form name of each form and the text information of the column name of each form. Then, the user question and the text information of the form name and column name of each form are concatenated, and the concatenated text is mapped to a corresponding low-dimensional dense vector through a text encoding model (or a text encoder, etc.), and the cosine similarity between the question and each form is calculated. The top n are selected based on the cosine similarity, where n is a natural number greater than or equal to 1, and the target form related to the user question is recalled. The column names of the target form and the user questions are input into the large language model (LLM). Through the prompt word engineering, the model is allowed to infer the column names of the target form and the proper nouns appearing in the user questions for interpretation, and obtain the preliminary interpretation q. If the model still has unclear items for the question, it will interact with the user to continuously update the knowledge interpretation q until all uncertain items are clear, and the interpretation knowledge q is updated to the complete interpretation knowledge Q. At this time, the large language model believes that the maintained knowledge is sufficient to solve the user's problem. Based on the interpretation knowledge Q, the large language model continues to judge the text intention of the user input, whether it is table analysis or drawing a chart, and takes two process branches respectively. If it is a table analysis, the target table and the maintained knowledge Q are input to generate the corresponding data analysis structured query language (SQL) statement, and then execute it in the background to obtain the corresponding result. Otherwise, the pandas code for drawing the chart is generated, and the corresponding chart is obtained after the background Python interpreter executes it.

[0157] Among them, pandas is a tool based on NumPy that was created to solve data analysis tasks. Pandas incorporates a large number of libraries and some standard data models, providing the tools needed to efficiently operate large data sets.

[0158] Python is a high-level programming language that is concise, easy to read, and easy to learn. It can be applied to many fields such as data analysis, network programming, artificial intelligence, scientific computing, etc.

[0159] The embodiment of the present invention is based on a large language model and does not require fine-tuning. It solves the problem of user analysis of tables and drawing icons by building an agent. Through form recall, the model focuses on key tables and questions and reduces redundant information. Memory maintenance is introduced to store interactive information and form knowledge to solve the context length limitation of large language models. Through the process of multiple rounds of interaction between the large language model and the user, the existing knowledge is corrected and improved, so as to solve the user's current problem in a more targeted manner.

[0160] It should be noted that, for the method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that this implementation disclosure is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the present invention.

[0161] Figure 3 3 is a block diagram of a multi-form processing device provided by an embodiment of the present invention. The device comprises: a parsing module 301, a determining module 302, a first interpretation module 303, a first generating module 304 and a first executing module 305, wherein:

[0162] The parsing module 301 is used to parse the form in the user input form to obtain the form name and the column name text information of each form;

[0163] The determination module 302 is used to determine a target form related to the question based on the question input by the user;

[0164] The first interpretation module 303 is used to interpret the column name text information of the target form and the proper noun in the question to obtain first interpretation knowledge;

[0165] The first generation module 304 is configured to generate a structured query language for corresponding data analysis based on the target form and the first interpretation knowledge when the first interpretation knowledge can solve the problem input by the user and if it is determined that the text intention of the user is to analyze a form problem;

[0166] The first execution module 305 is used to execute the structured query language to obtain corresponding table analysis results.

[0167] Optionally, in another embodiment, based on the above embodiment, the determination module 302 includes: a similarity determination module 401 and a recall module 402, and its structural block diagram is as follows: Figure 4 As shown,

[0168] The similarity determination module 401 is used to determine the cosine similarity of the forms related to the question based on the question input by the user;

[0169] The recall module 402 is used to recall the target form related to the question based on the cosine similarity.

[0170] Optionally, in another embodiment, based on the above embodiment, the similarity determination module 401 includes: a first mapping module 501, a first extraction module 502, a splicing module 503, a second mapping module 504 and a similarity calculation module 505, and its structural block diagram is shown as follows: Figure 5 As shown,

[0171] The first mapping module 501 is used to map the text of the question input by the user into a first low-dimensional dense vector;

[0172] The first extraction module 502 is used to extract the form name of each form and the column name text information of the corresponding form;

[0173] The splicing module 503 is used to splice the form name of each form and the corresponding column name text information according to a standard format to obtain the spliced ​​text of each form;

[0174] The second mapping module 504 is used to map the concatenated text of each form into a second low-dimensional dense vector;

[0175] The similarity calculation module 505 is used to calculate the cosine similarity between the question and each form based on the first low-dimensional dense vector and the second low-dimensional dense vector.

[0176] Optionally, in another embodiment, based on the above embodiment, the recall module 402 includes: a sorting module 601 and a form recall module 602, and its structural block diagram is as follows: Figure 6 As shown,

[0177] The sorting module 601 is used to sort all the forms from high to low according to the cosine similarity;

[0178] The form recall module 602 is used to select the first n forms in the sorting process for recall, and obtain a target form, wherein n is a natural number greater than or equal to 1.

[0179] Optionally, in another embodiment, based on the above embodiment, the first interpretation module 303 includes: a loading module 701, a second extraction module 702 and a first interpretation knowledge module 703, and its structural block diagram is as follows: Figure 7 As shown,

[0180] The loading module 701 is used to load the content of the target form into a dataframe or polardb format;

[0181] The second extraction module 702 is used to extract the column names of all target forms after loading;

[0182] The first interpretation knowledge module 703 is used to interpret the column names of all target forms and the professional names appearing in the questions input by the user by using the prompt word engineering in the large language model to obtain the first interpretation knowledge.

[0183] Optionally, in another embodiment, based on the above embodiment, the device further includes: a second generating module and a second executing module, wherein:

[0184] The second generation module is used for generating corresponding pandas drawing code based on the target form and the first interpretation knowledge when the first interpretation knowledge can solve the problem input by the user and if it is determined that the text intention of the user is to draw a chart;

[0185] The second execution module is used to execute the corresponding pandas drawing code through the Python interpreter to obtain a corresponding chart.

[0186] Optionally, in another embodiment, based on the above embodiment, the device further includes: an entry determination module, an update module, a third generation module and a third execution module, wherein:

[0187] An item determination module, used for determining uncertain items in the question when the first interpretation knowledge cannot solve the question raised by the user;

[0188] An updating module, configured to interact with the user based on the uncertain item, and continuously update the first interpretation knowledge according to the interaction result to obtain the second interpretation knowledge;

[0189] A third generation module is used to generate a structured query language corresponding to data analysis based on the target form and the second interpretation knowledge when it is determined that the user's text intention is to analyze a form problem;

[0190] The third execution module is used to execute the structured query language to obtain corresponding table analysis results.

[0191] Optionally, in another embodiment, based on the above embodiment, the device further includes: a fourth generating module and a fourth executing module, wherein:

[0192] A fourth generation module is used to generate corresponding pandas drawing code based on the target form and the second interpretation knowledge when it is determined that the user's text intention is to draw a chart;

[0193] The fourth execution module is used to execute the corresponding pandas drawing code through the Python interpreter to obtain a corresponding chart.

[0194] Optionally, an embodiment of the present invention further provides an electronic device, including:

[0195] processor;

[0196] a memory for storing instructions executable by the processor;

[0197] The processor is configured to execute the instructions to implement the multi-form processing method as described above.

[0198] Optionally, an embodiment of the present invention further provides a computer-readable storage medium, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the multi-table processing method as described above.

[0199] Optionally, an embodiment of the present invention further provides a computer program product, including a computer program or instructions, which implement the multi-table processing method as described above when executed by a processor of an electronic device.

[0200] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0201] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0202] Figure 8 800 is a block diagram of an electronic device 800 provided in an embodiment of the present invention. For example, the electronic device 800 may be a mobile terminal or a server, and the electronic device is a mobile terminal as an example for explanation in the embodiment of the present invention. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a message transceiver device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0203] Reference Figure 8 , the electronic device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0204] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0205] The memory 804 is configured to store various types of data to support operations on the device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0206] The power supply component 806 provides power to the various components of the electronic device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device 800.

[0207] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and the rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0208] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0209] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.

[0210] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the electronic device 800. For example, the sensor assembly 814 can detect the open / closed state of the device 800, the relative positioning of components, such as the display and keypad of the electronic device 800, and the sensor assembly 814 can also detect the position change of the electronic device 800 or a component of the electronic device 800, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and the temperature change of the electronic device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0211] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, a carrier network (such as 2G, 3G, 4G or 5G), or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0212] In an embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to execute the multi-table processing method shown above.

[0213] In an embodiment, a computer-readable storage medium is also provided, and when the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device 800 can perform the multi-form processing method shown above. For example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0214] In an embodiment, a computer program product is also provided, including a computer program or instructions. When the computer program or instructions are executed by the processor 820 of the electronic device 800, the electronic device 800 executes the above-mentioned multi-form processing method.

[0215] Fig. 9 is a block diagram of a device 900 for multi-table processing provided by an embodiment of the present invention. For example, the device 900 may be provided as a server. Fig. 9 , the apparatus 900 includes a processing component 922, which further includes one or more processors, and a memory resource represented by a memory 932 for storing instructions, such as an application, that can be executed by the processing component 922. The application stored in the memory 932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 922 is configured to execute the instructions to perform the above method.

[0216] The device 900 may also include a power supply component 926 configured to perform power management of the device 900, a wired or wireless network interface 950 configured to connect the device 900 to a network, and an input / output (I / O) interface 958. The device 900 may operate based on an operating system stored in the memory 932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or the like.

[0217] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0218] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A multi-form processing method, characterized in that: include: Parse the form in the user input table to obtain the form name and column name text information of each form; Based on the question input by the user, determining a target form related to the question; Interpreting the column name text information of the target form and the proper noun in the question to obtain first interpretation knowledge; When the first interpretation knowledge can solve the problem input by the user, if it is determined that the text intention of the user is to analyze a form problem, then a structured query language for corresponding data analysis is generated based on the target form and the first interpretation knowledge; Execute the structured query language to obtain corresponding table analysis results.

2. The multi-form processing method according to claim 1, characterized in that: The step of determining a target form related to the question based on the question input by the user comprises: Based on the question input by the user, determining the cosine similarity of the forms related to the question; A target form related to the question is recalled based on the cosine similarity.

3. The multi-form processing method according to claim 2, characterized in that: The step of determining the cosine similarity of the forms related to the question based on the question input by the user comprises: Mapping the text of the question input by the user into a first low-dimensional dense vector; Extract the form name of each form and the column name text information of the corresponding form; splicing the form name of each form and the corresponding column name text information according to a standard format to obtain a spliced ​​text of each form; Mapping the concatenated text of each form into a second low-dimensional dense vector; Based on the first low-dimensional dense vector and the second low-dimensional dense vector, the cosine similarity between the question and each form is calculated.

4. The multi-form processing method according to claim 2, characterized in that: The step of recalling a target form related to the question based on the cosine similarity comprises: Sort all forms by cosine similarity from high to low; The first n ranked forms are selected for recall to obtain a target form, where n is a natural number greater than or equal to 1.

5. The multi-form processing method according to claim 1, characterized in that: The interpreting the column names of the target form and the proper nouns in the question to obtain first interpretation knowledge includes: Load the content of the target form into dataframe or polardb format; Extract the column names of all target tables after loading; By using the prompt word engineering in the large language model, the column names of all target forms and the professional names appearing in the questions input by the user are interpreted to obtain the first interpretation knowledge.

6. The multi-table processing method according to any one of claims 1 to 5, characterized in that: The method further comprises: When the first interpretation knowledge can solve the problem input by the user, if it is determined that the text intention of the user is to draw a chart, then based on the target form and the first interpretation knowledge, a corresponding pandas drawing code is generated; The corresponding pandas drawing code is executed through the Python interpreter to obtain the corresponding chart.

7. The multi-form processing method according to any one of claims 1 to 5, characterized in that: The method further comprises: When the first interpretation knowledge cannot solve the question raised by the user, determining uncertain items in the question; Interacting with the user based on the uncertain entry, updating the first interpretation knowledge according to the interaction result, and obtaining the second interpretation knowledge; When it is determined that the user's text intention is to analyze a form question, generating a structured query language corresponding to data analysis based on the target form and the second interpretation knowledge; Execute the structured query language to obtain corresponding table analysis results.

8. The multi-form processing method according to claim 7, characterized in that: The method further comprises: When it is determined that the user's text intention is to draw a chart, a corresponding pandas drawing code is generated based on the target form and the second interpretation knowledge; The corresponding pandas drawing code is executed through the Python interpreter to obtain the corresponding chart.

9. A multi-form processing device, characterized in that: include: The parsing module is used to parse the form in the user input table to obtain the form name and the column name text information of each form; A determination module, used for determining a target form related to the question based on the question input by the user; A first interpretation module, used for interpreting the column name text information of the target form and the proper noun in the question to obtain first interpretation knowledge; A first generating module is used for generating a structured query language for corresponding data analysis based on the target form and the first interpretive knowledge when the first interpretive knowledge can solve the problem input by the user and if it is determined that the text intention of the user is to analyze a form problem; The first execution module is used to execute the structured query language to obtain corresponding table analysis results.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the multi-form processing method as claimed in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the multi-form processing method as claimed in any one of claims 1 to 8.