A text review method, system, terminal and medium

Abstract

The present invention discloses a text review method, system, terminal and medium, the method comprising: obtaining an original text, and processing the original text to obtain a processed text; extracting objects from the processed text to obtain a plurality of extracted objects; performing local evaluation, global evaluation and industry evaluation on a plurality of extracted objects among the plurality of extracted objects through an optimized large language model, and obtaining a local evaluation result, a global evaluation result and an industry evaluation result of each extracted object respectively; obtaining an overall evaluation result of the original text according to the local evaluation result, the global evaluation result and the industry evaluation result. The present invention can improve the review efficiency and the review quality, and simulates the review process of multi-expert collaboration to achieve effective integration and complementarity of knowledge in different fields.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a text review method, system, terminal and medium. Background Art

[0002] With the rapid development of science and technology, the number of peer-reviewed scientific publications such as papers has grown exponentially, making it a major challenge for field experts to conduct high-quality reviews of scientific publications.

[0003] In recent years, researchers have begun to explore the possibility of using natural language processing (NLP) models to achieve preliminary AI-assisted review for scientific papers. For example, by collecting paper datasets in the field of machine learning, targeted summary models were trained to generate paper review opinions.

[0004] However, existing methods, whether traditional manual review or AI-assisted paper review, have problems such as low review efficiency, inability to handle paper review tasks in multiple fields, and lack of a multi-dimensional and multi-angle evaluation framework, which affects the accuracy of the review opinions.

[0005] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0006] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a text review method, system, terminal and medium are provided, aiming to solve the problems existing in the prior art of text review technology and methods, such as low review efficiency, inability to handle paper review tasks in multiple fields, lack of a multi-dimensional and multi-angle evaluation framework, thereby affecting the accuracy of the review opinions, etc.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] In a first aspect, the present invention provides a text review method, wherein the method comprises:

[0009] Obtaining an original text, and processing the original text to obtain a processed text;

[0010] Performing object extraction on the processed text to obtain a number of extracted objects;

[0011] Performing local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects by using the optimized large language model, and obtaining local evaluation results, global evaluation results and industry evaluation results of each extraction object respectively;

[0012] An overall evaluation result of the original text is obtained according to the local evaluation result, the global evaluation result and the industry evaluation result.

[0013] In one implementation, the original text is processed to obtain a processed text, including:

[0014] Segmenting the original text to obtain a plurality of original text paragraphs;

[0015] Converting each of the plurality of original text paragraphs into a text paragraph in a target format to obtain a formatted text paragraph;

[0016] Adding a number index before each of the formatted text paragraphs to obtain a text paragraph containing the number index;

[0017] According to the text paragraphs containing the number indexes, a processed text is obtained.

[0018] In one implementation, object extraction is performed on the processed text to obtain a plurality of extracted objects, including:

[0019] A regular expression is used to match each text paragraph of the processed text to obtain a plurality of extraction objects, the text paragraphs of the plurality of extraction objects, and the serial number indexes of the text paragraphs of the plurality of extraction objects.

[0020] In one implementation, object extraction is performed on the processed text to obtain a plurality of extracted objects, including:

[0021] Based on each text paragraph of the processed text and a preset first prompt, a combined text paragraph is obtained;

[0022] The combined text paragraph is input into the optimized large language model to obtain a plurality of extraction objects, the text paragraphs of the plurality of extraction objects and the number indexes of the text paragraphs of the plurality of extraction objects.

[0023] In one implementation, the optimized large language model is used to perform local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects, and local evaluation results, global evaluation results and industry evaluation results of each extraction object are obtained respectively, including:

[0024] Performing local evaluation on text paragraphs of multiple extraction objects among the plurality of extraction objects to obtain a local evaluation result of each extraction object;

[0025] Acquire structured text catalogs of multiple extracted objects among the plurality of extracted objects, and evaluate at least one text paragraph of the multiple extracted objects in the structured text catalog based on the multiple extracted objects and the structured text catalog to obtain a global evaluation result of each extracted object;

[0026] An industry assessment is performed on text paragraphs of multiple extracted objects among the plurality of extracted objects to obtain an industry assessment result.

[0027] In one implementation, locally evaluating the text paragraphs of multiple extraction objects among the plurality of extraction objects to obtain a local evaluation result of each extraction object includes: for each extraction object among the multiple extraction objects, respectively screening out a corresponding text paragraph in the processed text, and respectively constructing a second prompt for each extraction object;

[0028] The second prompt is input into the optimized large language model to perform local evaluation on each extracted object to obtain a local evaluation result of each extracted object.

[0029] In one implementation, constructing the second hint for each extracted object separately includes:

[0030] Get the preset first prompt template;

[0031] Based on the preset first prompt template and the text paragraph of each extracted object, a second prompt of each extracted object is obtained.

[0032] In one implementation, obtaining structured text directories of multiple extracted objects among the plurality of extracted objects, and evaluating at least one text paragraph of the multiple extracted objects in the structured text directory based on the multiple extracted objects and the structured text directory to obtain a global evaluation result of each extracted object includes:

[0033] Based on the text paragraphs of the processed text, the structured text catalog is obtained, wherein the structured text catalog contains multiple chapters and multiple sections;

[0034] Evaluate at least one text paragraph of the multiple extraction objects in the structured text catalog according to the text paragraphs of the multiple extraction objects and the structured text catalog to obtain cross-chapter evaluation results of the multiple extraction objects;

[0035] Evaluate the contents of each chapter in the structured text catalog of the plurality of extracted objects respectively to obtain evaluation results of each chapter of the plurality of extracted objects;

[0036] Evaluate each section of the structured text catalog of the plurality of extraction objects respectively to obtain evaluation results of each section of the plurality of extraction objects;

[0037] A global evaluation result of each extracted object is obtained according to the per-chapter evaluation results and the per-section evaluation results of the multiple extracted objects.

[0038] In one implementation, obtaining the structured text catalog based on the text paragraphs of the processed text includes:

[0039] Based on all text paragraphs of the processed text and a preset second prompt template, a third prompt is obtained;

[0040] Inputting the third prompt into the optimized large language model to obtain a text directory;

[0041] Based on the main text directory and a preset third prompt template, a fourth prompt is obtained, and the fourth prompt is input into the optimized large language model to obtain a structured main text directory.

[0042] In one implementation, evaluating at least one text paragraph of the multiple extraction objects in the structured text directory according to the text paragraphs of the multiple extraction objects and the structured text directory to obtain cross-chapter evaluation results of the multiple extraction objects includes:

[0043] Based on all text paragraphs of the processed text and a preset fourth prompt template, a fifth prompt is obtained, and the fifth prompt is input into the optimized large language model to obtain a text intention conclusion;

[0044] A sixth prompt is obtained based on at least one text paragraph of the multiple extraction objects, the structured text directory and a preset fifth prompt template, and the sixth prompt is input into the optimized large language model to obtain a cross-chapter evaluation result.

[0045] In one implementation, evaluating the contents of each chapter in the structured text catalog of the plurality of extraction objects respectively to obtain the evaluation result of each chapter includes:

[0046] Based on the structured text catalog, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each chapter, the text paragraphs of each chapter, and the preset sixth prompt template, a seventh prompt is obtained;

[0047] Input the seventh prompt into the optimized large language model to obtain evaluation results for several chapters;

[0048] Based on the evaluation result of each chapter, the title of each chapter, the text paragraph of each chapter, the current system time and the seventh prompt template, an eighth prompt is obtained;

[0049] The eighth prompt is input into the optimized large language model to obtain evaluation results for each chapter of the multiple extraction objects.

[0050] In one implementation, each section of the structured text catalog of the plurality of extraction objects is evaluated respectively to obtain an evaluation result of each section of the plurality of extraction objects, including:

[0051] Based on the structured text catalog, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each section, the text paragraphs of each section, and the preset eighth prompt template, a ninth prompt is obtained;

[0052] Inputting the ninth prompt into the optimized large language model to obtain evaluation results of several sections;

[0053] Based on the evaluation result of each section, the title of each section, the text paragraph of each section, the current system time and the ninth prompt template, a tenth prompt is obtained;

[0054] The tenth prompt is input into the optimized large language model to obtain evaluation results for each section of the multiple extraction objects.

[0055] In one implementation, after obtaining a global evaluation result of each extracted object according to the evaluation results of each chapter and the evaluation results of each section of the plurality of extracted objects, the method further includes:

[0056] Respectively obtaining the number indexes corresponding to the text paragraphs of each chapter and each section of the plurality of extraction objects in the structured text directory;

[0057] Inserting the evaluation result of each chapter into the first numbered index of the corresponding chapter according to the numbered index corresponding to the text paragraph of each chapter;

[0058] The evaluation result of each section is inserted into the first numbered index of the corresponding section according to the numbered index corresponding to the text paragraph of each section.

[0059] In one implementation, industry evaluation is performed on text paragraphs of multiple extracted objects among the plurality of extracted objects to obtain an industry evaluation result of each extracted object, including:

[0060] Obtaining an eleventh prompt according to the text information input by the user and the preset tenth prompt template;

[0061] Inputting the eleventh prompt into the optimized large language model to obtain a preliminary evaluation result of each extracted object;

[0062] Based on the preliminary evaluation result and the preset eleventh prompt template, a twelfth prompt is obtained;

[0063] The twelfth prompt is input into the optimized large language model to obtain the industry evaluation result of each extracted object.

[0064] In one implementation, obtaining an overall evaluation result of the original text according to the local evaluation result, the global evaluation result and the industry evaluation result includes:

[0065] Based on the local evaluation result, the global evaluation result, the industry evaluation result, the text title, the user type and the preset twelfth prompt template, a thirteenth prompt is obtained;

[0066] The thirteenth prompt is input into the optimized large language model to obtain an overall evaluation result of the original text sorted by importance, and the overall evaluation result is inserted at the end of the original text.

[0067] In one implementation, the training process of the optimized large language model includes: classifying texts in a text library and sampling in each text classification to obtain text sampling data;

[0068] For each text in the text sampling data, construct a fine-tuning data pair, wherein the fine-tuning data pair consists of a prompt and output content of a preset large language model;

[0069] Obtain a fine-tuning data set according to the fine-tuning data pair;

[0070] Constructing a first fine-tuning dataset based on the fine-tuning dataset and a preset general dataset;

[0071] The fine-tuning data set in the first fine-tuning data set is processed to obtain a second fine-tuning data set, and the second fine-tuning data set and a preset large language model are fine-tuned to obtain the optimized large language model.

[0072] In a second aspect, an embodiment of the present invention further provides a text review system, the system comprising:

[0073] A text processing module, used to obtain the original text and process the original text to obtain a processed text;

[0074] An object extraction module, used to extract objects from the processed text to obtain a number of extracted objects;

[0075] A text evaluation module, used to perform local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects through the optimized large language model, and obtain local evaluation results, global evaluation results and industry evaluation results of each extraction object respectively;

[0076] An overall evaluation module is used to obtain an overall evaluation result of the original text according to the local evaluation result, the global evaluation result and the industry evaluation result.

[0077] In a third aspect, an embodiment of the present invention further provides a terminal, wherein the terminal includes a memory, a processor, and a text review program stored in the memory and executable on the processor, and when the processor executes the text review program, the steps of the text review method of any one of the above-mentioned schemes are implemented.

[0078] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein a text review program is stored on the computer-readable storage medium, and when the text review program is executed by a processor, the steps of the text review method described in any one of the above schemes are implemented.

[0079] Beneficial effects: Compared with the prior art, the present invention provides a text review method. The present invention first obtains the original text, processes the original text to obtain the processed text, and then extracts the object from the processed text to obtain several extracted objects. Then, the optimized large language model is used to perform local evaluation, global evaluation and industry evaluation on multiple extracted objects among the several extracted objects, and local evaluation results, global evaluation results and industry evaluation results are obtained respectively. It can be seen that the present invention can perform local evaluation, global evaluation and industry evaluation on the original text, thereby achieving a comprehensive evaluation of the original text in multiple dimensions, improving the evaluation quality, and facilitating the output of accurate review opinions. Finally, the present invention obtains the overall evaluation result of the original text based on the local evaluation results, the global evaluation results and the industry evaluation results. The present invention greatly improves the review efficiency, and through multi-dimensional and fine-grained evaluation, it ensures the comprehensiveness and depth of the review, significantly surpassing the traditional review method. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 A flowchart of a preferred embodiment of the text review method provided in the embodiment of the present invention.

[0081] Figure 2 A flowchart of preliminary evaluation in the text review method provided in an embodiment of the present invention.

[0082] Figure 3 A flow chart of chapter evaluation in a text review method provided in an embodiment of the present invention.

[0083] Figure 4 A schematic diagram of multi-dimensional evaluation and suggestions in a text review method provided in an embodiment of the present invention.

[0084] Figure 5 A schematic diagram of the architecture of a text review system provided in an embodiment of the present invention.

[0085] Figure 6 A functional block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0086] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0087] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations or steps, nor must they be executed in the order described. For example, some operations or steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0088] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0089] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish between identical or similar items with substantially identical functions and effects. For example, the first control information and the second control information are only used to distinguish different control information, and their order is not limited.

[0090] Those skilled in the art can understand that the words "first", "second", etc. do not limit the quantity and execution order, and the words "first", "second", etc. do not necessarily limit the differences.

[0091] It should be further understood that the term “and / or” used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0092] The existing paper review technologies and methods, whether traditional manual review or preliminary AI-assisted review, basically have the problem of low efficiency. In addition, the manual review method is limited by the scope of personal knowledge and it is difficult to fully cover all aspects of the paper, while AI-assisted review often only focuses on specific aspects and is also difficult to conduct comprehensive and in-depth content analysis. In addition, both manual review and AI-assisted review have the problem of difficulty in determining unified review standards. Since manual review is conducted by multiple experts, the review process is often opaque and opinions are difficult to unify, while the opinions of AI-assisted review are often lacking in constructiveness, and the credibility of the review results is not high. Therefore, the existing manual review and AI-assisted review have seriously restricted the efficiency and quality of academic review, affecting the rapid dissemination of scientific research results and the effectiveness of academic exchanges.

[0093] This embodiment provides a text review method, firstly obtaining the original text, and processing the original text to obtain the processed text, then extracting the object from the processed text to obtain several extracted objects. Then, the optimized large language model is used to perform local evaluation, global evaluation and industry evaluation on multiple extracted objects among the several extracted objects, and local evaluation results, global evaluation results and industry evaluation results are obtained respectively. It can be seen that the present invention can perform local evaluation, global evaluation and industry evaluation on the original text, thereby realizing a comprehensive evaluation of the original text in multiple dimensions, improving the evaluation quality, and facilitating the output of accurate review opinions. Finally, this embodiment obtains the overall evaluation result of the original text based on the local evaluation results, the global evaluation results and the industry evaluation results. This embodiment can comprehensively review a text of tens of thousands of words in just a few minutes, greatly improving the review efficiency, and this embodiment can ensure the comprehensiveness and depth of the review through multi-dimensional and multi-angle evaluation, significantly surpassing the traditional review method. This embodiment gives the review results by the same model, avoiding the subjective factors of inconsistency among multiple experts. The review process can also be granularized to the point of sentences in a paragraph and phrases in a keyword. The opinions provided are practical and can support the addition of more detailed tasks for reviews at different stages. Each review will point out deficiencies and provide suggestions for improvement, thus improving the quality of the review.

[0094] The text review method of this embodiment can be applied to a terminal, which can be a computer, a smart TV, a mobile phone or other intelligent product terminal. Figure 1 As shown in , the text review method of this embodiment includes the following steps:

[0095] Step S100: obtaining an original text, and processing the original text to obtain a processed text.

[0096] Specifically, this embodiment takes a paper as an example. After the original text of the paper is obtained, the original text is first segmented to obtain several original text paragraphs. Then, each of the several original text paragraphs is converted into a text paragraph in the target format to obtain a formatted text paragraph. In this embodiment, each text paragraph can be converted into JSON format to obtain a text paragraph in JSON format. JSON format is an easy-to-process data format, which facilitates the processing of text paragraphs in JSON format in subsequent steps. Further, this embodiment adds a numbered index before each text paragraph in JSON format to obtain a text paragraph in JSON format containing a numbered index, so as to locate each text paragraph. Then, this embodiment obtains a processed text based on the text paragraph in JSON format containing a numbered index, and the processed text is the text paragraph and the numbered index corresponding to the text paragraph.

[0097] Step S200: extract objects from the processed text to obtain a number of extracted objects.

[0098] This embodiment takes a paper as an example, and the extraction objects of the paper include paper abstract, keywords, paper title, Chinese and English abstracts, Chinese and English keywords, table of contents, chapter titles, paragraphs, main text / section, references, etc.

[0099] In specific applications, this embodiment can use regular expressions to match each text paragraph of the processed text to obtain a number of extraction objects, the text paragraphs of the extraction objects, and the number indexes of the text paragraphs of the extraction objects, such as performing string or character rule matching to obtain a number of extraction objects, the text paragraphs of the extraction objects, and the number indexes of the text paragraphs of the extraction objects.

[0100] For example, the string matching for paper abstracts is as follows:

[0101] abstract_pattern = re.compile(r"Abstract[::]\s*") # Use regular expressions to match "Abstract:" or "Abstract:" and any space after it.

[0102] In another implementation, this embodiment can also obtain a combined text paragraph based on each text paragraph of the processed text and a preset first prompt. Then, the combined text paragraph is input into the optimized large language model to obtain a number of extraction objects, the text paragraphs of the several extraction objects, and the number indexes of the text paragraphs of the several extraction objects. Specifically, in this embodiment, the first prompt is used to extract key information in the text, such as the title, Chinese and English abstracts, Chinese and English keywords, table of contents, chapter titles, paragraphs and other key information. In actual application, each text paragraph in JSON format containing a number index and the preset first prompt can be spliced ​​to obtain a combined text paragraph. Then, the combined text paragraph is input into the optimized large language model, and the text paragraphs of the several extraction objects and the number indexes corresponding to the text paragraphs of the several extraction objects are output, such as, output: "1. Abstract", "2. Introduction".

[0103] In this embodiment, the optimized large language model is pre-trained. In the actual training process, this embodiment can classify the text in the text library and sample in each text classification to obtain text sampling data. Then, for each text in the text sampling data, a fine-tuning data pair is constructed. The fine-tuning data pair consists of a prompt and the output content of the preset large language model, which reflects the mapping relationship between the prompt and the output content of the preset large language model. The prompt can be any prompt in the first prompt, the second prompt, ..., and the twelfth prompt. The output content is the output obtained by the preset large language model according to the prompt. Then, a fine-tuning data set is constructed based on the fine-tuning data pair and the preset general data set. Then, a second fine-tuning data set is obtained according to the fine-tuning data in the fine-tuning data set, and the second fine-tuning data set and the preset model are fine-tuned to obtain the optimized large language model. Specifically, this embodiment can construct a first fine-tuning data set based on the fine-tuning data pair, and splice the first fine-tuning data set and the general data set to obtain a spliced ​​fine-tuning data set. Finally, the fine-tuning data pairs in the spliced ​​fine-tuning data set are randomly shuffled to obtain a second fine-tuning data set, and the second fine-tuning data set and the base large model are fine-tuned by SFT (Supervised Fine-Tuning) technology to obtain the optimized large language model. By pre-training the model, the text paragraphs of several extraction objects and the number indexes corresponding to the text paragraphs of the several extraction objects can be automatically output according to the input spliced ​​text paragraphs.

[0104] Step S300: Perform local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects through the optimized large language model, and obtain local evaluation results, global evaluation results and industry evaluation results of each extraction object respectively.

[0105] Specifically, this embodiment can locally evaluate the text paragraphs of multiple extracted objects among the several extracted objects to obtain the local evaluation results of each extracted object. Then, the structured text directory of the multiple extracted objects among the several extracted objects is obtained, and at least one text paragraph of the multiple extracted objects in the structured text directory is evaluated based on the multiple extracted objects and the structured text directory to obtain the global evaluation results of each extracted object, wherein the text directory contains the title of each chapter of the text body and the numbered index corresponding to the text paragraphs of the several extracted objects, and the title of each section and the numbered index corresponding to the text paragraphs of the several extracted objects. The global evaluation results are at least used to reflect the logic of the chapter content of each extracted object, the consistency with the title, the research questions and the text structure. Then, the text paragraphs of multiple extracted objects among the several extracted objects are evaluated by industry to obtain industry evaluation results, which are at least used to reflect the value of the paper selection, the professional matching, the innovation, the adequacy of the literature review and the logic of the paper.

[0106] Specifically, combined Figure 2 As shown, when performing local evaluation on the text paragraphs of several extraction objects, it includes performing polishing evaluation on the text paragraphs to determine whether it is necessary to perform polishing processing on the text paragraphs of several extraction objects, such as summary polishing evaluation, paragraph content polishing evaluation, etc. Specifically, for each of the multiple extraction objects, this embodiment selects the corresponding text paragraphs from the processed text, and constructs the second prompt for each of the extraction objects. When constructing the second prompt, this embodiment first obtains a preset first prompt template (such as Figure 2 As shown). Then, based on the preset first prompt template and the text paragraphs related to each extracted object, the second prompt of each extracted object is obtained. Specifically, the first prompt template and the text paragraphs related to each extracted object can be spliced ​​to obtain the second prompt of each extracted object. After obtaining the second prompt, this embodiment inputs the second prompt into the optimized large language model to perform polishing evaluation on each extracted object to obtain the local evaluation results of each extracted object. The local evaluation results include "no need for polishing" and "polishing results and existing problems". If the local evaluation result is "no need for polishing", the task flow ends; otherwise, each problem and polishing result is pointed out and stored as input for subsequent tasks. If annotations of the local evaluation results need to be inserted in the original text, the annotations will be stored in the numbered indexes corresponding to the text paragraphs of the several extracted objects.

[0107] In practical applications, the local evaluation results also include other evaluation results. Taking the paper text as an example, when performing local evaluation on multiple extraction objects of the paper text, such as title evaluation, Chinese and English keyword optimization evaluation, reference quality inspection evaluation, etc., this embodiment can set the specific content of the corresponding first prompt template according to different extraction objects. The first prompt template contains: task, role, background, goal, constraint, skill, workflow, input, output, initialization and other field contents. The specific content contained in each field content is specifically set according to different extraction objects; then it is necessary to splice the text paragraph related to the extraction object with the first prompt template as the second prompt, and the second prompt can be used to reflect whether the text paragraph related to the object needs to be optimized / polished. Since the text paragraphs related to each extraction object are generally different, this embodiment can screen the related text paragraphs according to the specific extraction object. If the final output local evaluation result is "no optimization / polishing", the task flow is terminated. Otherwise, each problem and optimization / polishing result is pointed out and stored as the input of the subsequent task. If it is necessary to insert annotations of the local evaluation results, the annotations will be stored at the numbered indexes corresponding to the text paragraphs of the several extraction objects.

[0108] For example, when performing a partial evaluation of a title, the input is: paper title, user major, nature of the paper, paper table of contents, Chinese abstract, and introduction; the text paragraph content corresponding to the input is spliced ​​with the first prompt template as the second prompt. The preset first prompt template includes: tasks, roles, backgrounds, goals, constraints, skills, workflows, inputs, outputs, initialization, and other content related to the title; then the optimized large language model is input; the output partial evaluation results include the conclusion that the title needs to be polished, and the partial evaluation results will be inserted into the corresponding annotations.

[0109] For another example, when performing a partial evaluation of an abstract, the input is: paper title, Chinese abstract, English abstract; the text paragraph content corresponding to the input is spliced ​​with the preset first prompt template as the second prompt. Similarly, the preset first prompt template contains: tasks, roles, background, goals, constraints, skills, workflows, inputs, outputs, initialization and other content related to the abstract. Then the input is given to the optimized large language model, and the output partial evaluation result is: whether the Chinese abstract needs to be polished. At this time, the partial evaluation result will be inserted into the corresponding annotation.

[0110] For example, when evaluating the language of a paragraph, it is divided into two stages:

[0111] Stage 1: Input: paper title, content of this paragraph; the corresponding paper paragraph content is spliced ​​with the preset first prompt template as the second prompt. Similarly, the preset first prompt template contains: tasks, roles, backgrounds, goals, constraints, skills, workflows, inputs, outputs, initialization and other content related to the content of this paragraph. Then it is input to the optimized large language model, and the output local evaluation results are the main problems with the language quality of the content of this paragraph, and whether the paragraph language needs to be polished and evaluated. If polishing is required, enter stage 2. At this time, the input is: paper title, local evaluation results of the content of this paragraph in stage 1, and the entire paragraph. The corresponding text paragraph content is spliced ​​with the first prompt template as the second prompt, and then input to the optimized large language model. The output local evaluation results include the polished content of this paragraph. At this time, the polishing evaluation results are inserted into the corresponding annotations.

[0112] For another example, when the quality of references is evaluated locally, it is divided into two stages. Stage 1 has two tasks. At this time, the first task of stage 1 is the quality detection of references. The input is: current time, nature of the paper, paper title, paper introduction, and reference list. When the corresponding paper paragraph content is spliced ​​with the first prompt template as the second prompt, the preset first prompt template includes: tasks, roles, background, goals, constraints, skills, workflows, inputs, outputs, initialization and reference-related content. Then it is input into the optimized large language model, and the output local evaluation result is: the core problems in the quality of references. At this time, the local evaluation result will not be inserted into the annotation. Next, the second task of stage 1 is the evaluation of the adequacy of the literature review. At this time, the input is the current time, paper title, paper introduction, paper introduction, the main research questions or hypotheses of the paper, and the reference list. At this time, the first prompt template includes: tasks, roles, background, goals, constraints, skills, workflows, inputs, outputs, initialization and other content related to the input content. Then, the corresponding paper paragraph content of the input is spliced ​​with the first prompt template as the second prompt, and then input into the optimized large language model. The output local evaluation result is the problem of the reference review. At this time, the local evaluation result will not be inserted into the annotation. Then, proceed to stage two: overall evaluation of references. At this time, the input is: current time, nature of the paper, title of the paper, reference list, reference quantity check result, reference quality check result, reference review check result. Then, the corresponding text paragraph content of the input is spliced ​​with the first prompt template as the second prompt, and then input into the optimized large language model. The output local evaluation result is the review results and suggestions of the summary references. At this time, the local evaluation result is inserted into the annotation.

[0113] It can be seen that this embodiment implements local evaluation of each extracted object based on the optimized large language model, thereby outputting a local evaluation result.

[0114] When evaluating each of the multiple objects of several extracted objects globally, combined Figure 3 As shown, this embodiment evaluates at least one text paragraph of the multiple extracted objects in the structured text directory according to the text paragraphs of the multiple extracted objects and the structured text directory to obtain cross-chapter evaluation results of the multiple extracted objects. Then, the content of each chapter in the structured text directory of the multiple extracted objects is evaluated respectively to obtain the evaluation results of each chapter of the multiple extracted objects. Next, the content of each section in the structured text directory of the multiple extracted objects is evaluated respectively to obtain the evaluation results of each section of the multiple extracted objects. Then, based on the evaluation results of each chapter and the evaluation results of each section of the multiple extracted objects, a global evaluation result of each extracted object is obtained.

[0115] Specifically, the present embodiment can obtain a third prompt based on all text paragraphs of the processed text and the preset second prompt template. Specifically, the full text paragraph with a numbered index can be spliced ​​with the preset second prompt template to obtain the third prompt. The third prompt is input into the optimized large language model to obtain a text directory, and each chapter title and each section title in the text directory are stored according to their order in the text body. It should be noted that each prompt template of the present embodiment is pre-set, and each prompt template is used to process the data input into the optimized large language model, and the corresponding prompt can be used to reflect the output data of the optimized large language model, that is, corresponding to the output data. For example, the second prompt template mentioned above is used to process the full text paragraph of the processed text so that it can be better processed when input into the optimized model. The obtained third prompt can be used to reflect the output data of the optimized model, that is, the text directory in the full text paragraph. Further, the present embodiment can obtain a fourth prompt based on the text directory and the preset third prompt template. Specifically, the text directory and the preset third prompt template can be spliced ​​to obtain the fourth prompt. The third prompt template is a template for associating with the text directory, and the fourth prompt can be used to reflect the structured text directory in the full text paragraph. Therefore, the fourth prompt is input into the optimized large language model to output the structured text directory.

[0116] Further, this embodiment obtains the fifth prompt based on all the text paragraphs of the processed text and the preset fourth prompt template. Similarly, the text paragraph in JSON format containing the number index can be spliced ​​with the preset fourth prompt template to obtain the fifth prompt. The fourth prompt template of this embodiment is used to process the text paragraph in JSON format. The fifth prompt can reflect the text intention conclusion in all the text paragraphs of the processed text. For this purpose, the fifth prompt is input into the optimized large language model to obtain the text intention conclusion, which includes the main research questions of the text. Then, the sixth prompt is obtained according to at least one text paragraph of the multiple extracted objects, the structured text directory and the preset fifth prompt template. The fifth prompt template is used to process the above text paragraphs and the structured text directory. The sixth prompt can be a cross-chapter evaluation based on the structured text directory. For this purpose, the sixth prompt is input into the optimized large language model to obtain a cross-chapter evaluation result. The cross-chapter evaluation result of this embodiment also includes optimization suggestions. Finally, the content of each chapter and each section in the structured text directory of the multiple extracted objects are evaluated respectively to obtain the evaluation results of each chapter and each section.

[0117] In specific applications, when evaluating the contents of each chapter in the structured text directory, such as Figure 4 As shown in , this embodiment can obtain the seventh prompt based on the structured text directory, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each chapter, the text paragraph of each chapter, and the preset sixth prompt template. Similarly, the sixth prompt template is also used to process the above-mentioned structured text directory, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each chapter, and the text paragraph of each chapter. The seventh prompt can be used to reflect the evaluation of each chapter. Therefore, the seventh prompt is input into the optimized large language model to obtain the evaluation results of several chapters. Then, based on the evaluation results of each chapter, the title of each chapter, the text paragraph of each chapter, the current system time and the seventh prompt template, the eighth prompt is obtained. The seventh prompt template is used to process the evaluation results of each chapter, the title of each chapter, the text paragraphs of each chapter, and the current system time. The eighth prompt is used to correspond to the evaluation results of each chapter. Therefore, the eighth prompt is input into the optimized large language model to obtain the evaluation results of each chapter of the multiple extraction objects. Similarly, the evaluation results of each chapter at this time also include optimization suggestions.

[0118] In specific application, when evaluating the content of each section in the structured text catalog, the ninth prompt is obtained based on the structured text catalog, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each section, the text paragraph of each section, and the preset eighth prompt template. The eighth prompt template is used to process the structured text catalog, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each section, and the text paragraph of each section, and the ninth prompt is used to obtain the evaluation results of several sections. To this end, the ninth prompt is input into the optimized large language model to obtain the evaluation results of several sections. Then, based on the evaluation result of each section, the title of each section, the text paragraph of each section, the current system time and the ninth prompt template, the tenth prompt is obtained. The ninth prompt template is used to process the evaluation result of each section, the title of each section, the text paragraph of each section, and the current system time, and the tenth prompt is used to correspond to the evaluation result of each section. Therefore, the tenth prompt is input into the optimized large language model to obtain the evaluation results of each section of the multiple extraction objects. Similarly, the evaluation results for each section at this time also include optimization suggestions. Finally, according to the evaluation results for each chapter and the evaluation results for each section of the multiple extracted objects, a global evaluation result for each extracted object is obtained.

[0119] Furthermore, this embodiment can also obtain the number indexes corresponding to the text paragraphs of each chapter and each section of the plurality of extracted objects in the structured text directory. Then, according to the number indexes corresponding to the text paragraphs of each chapter, the evaluation result of each chapter is inserted into the first number index of the corresponding chapter. Then, according to the number indexes corresponding to the text paragraphs of each section, the evaluation result of each section is inserted into the first number index of the corresponding section.

[0120] Taking the paper as an example, this embodiment mainly focuses on the relationship and overall structure within the paper chapter and between each chapter when performing the evaluation within and across chapters. The evaluation is mainly divided into two parts: one part selects text paragraphs, structured text directories, etc. of X extracted objects from all extracted objects to obtain cross-chapter evaluation results, and then evaluates each chapter and each section in the paper text according to the cross-chapter evaluation results and corresponding optimization suggestions to obtain each chapter evaluation result and each section evaluation result and suggestions, and finally summarizes each chapter evaluation result and each section evaluation result and suggestions as the global evaluation result of this chapter, and inserts them into the annotation of the chapter title.

[0121] It can be seen that this embodiment can realize the intra-chapter and cross-chapter evaluation of the text paragraphs of several extracted objects, so as to obtain the global evaluation results of each chapter in each extracted object, and the global evaluation results are at least used to reflect the logic of the chapter content, consistency with the title, research questions and paper structure.

[0122] Furthermore, industry evaluation and optimization include evaluation and optimization results in multiple dimensions, such as topic value, professional matching, innovation, adequacy of literature review, and paper logic. Topic value judgment (topic_value_judgement, topic_value_judgement_cot) is used to evaluate the academic value and practical significance of the paper topic. For example: it is affirmed that the paper topic is of great significance in the field of artificial intelligence, but it is recommended to further emphasize its potential impact in practical applications. Professional matching evaluation (user_profession_matching): used to evaluate the degree of match between the paper content and the author's professional background. For example: it is pointed out that the description of some technical details of the paper may need to be strengthened to reflect the author's professional depth. Innovation judgment (innovation_judge, innovation_judge_cot): used to evaluate the innovation of the paper and provide in-depth chain of thought analysis. For example: it is pointed out that the new algorithm proposed in the paper is innovative in some aspects, but it is recommended to further compare it quantitatively with existing methods. Literature review sufficiency evaluation (review_sufficiency): used to evaluate whether the literature review is comprehensive and whether it covers the latest research progress. For example: it is recommended to add a discussion of an emerging research direction in the literature review to reflect the cutting-edge nature of the paper. Paper logic evaluation (graduation_project_logic_evaluation_v2): used to evaluate the logical flow of the entire paper, including problem raising, method selection, experimental design, and conclusion derivation. For example: it is pointed out that the logical connection between the experimental design and the research question is not close enough, and it is recommended to adjust the experimental design to better answer the research question.

[0123] Taking the value of the topic as an example, this embodiment can splice the text information input by the user and the preset tenth prompt template to obtain the eleventh prompt. The text information input by the user can be professional information, text title, main research question of the text, or text paragraph of a certain extracted object. The tenth prompt template is used for the text information input by the user, and the eleventh prompt is used to correspond to the preliminary evaluation result of each extracted object. Therefore, this embodiment inputs the eleventh prompt into the optimized large language model to obtain the preliminary evaluation result of each extracted object. Then, based on the preliminary evaluation result and the preset eleventh prompt template, the twelfth prompt is obtained. The eleventh prompt template is used to process the preliminary evaluation result, and the twelfth prompt obtains the industry evaluation result of each extracted object. Finally, the twelfth prompt is input into the optimized large language model to obtain the industry evaluation result of each extracted object.

[0124] For the evaluation and optimization of other dimensions, this embodiment can replace the prompt information and prompt template, and use the same method as above to implement the evaluation of other dimensions. The industry evaluation results obtained in this embodiment will appear at the beginning or end of the paper in the form of an overall evaluation, and may also add cross-reference suggestions between various related chapters.

[0125] Step S400: obtaining an overall evaluation result of the original text according to the local evaluation result, the global evaluation result and the industry evaluation result.

[0126] Specifically, this embodiment can obtain the thirteenth prompt based on the local evaluation result, the global evaluation result, the industry evaluation result, the text title, the user type and the preset twelfth prompt template. The twelfth prompt template is used to process the local evaluation result, the global evaluation result, the industry evaluation result, the text title, and the user type. Specifically, the local evaluation result, the global evaluation result, the industry evaluation result, the text title, the user type and the preset twelfth prompt template can be spliced ​​to obtain the thirteenth prompt. The user type can be the user's academic qualifications (such as undergraduate, graduate, etc.). The thirteenth prompt is used to correspond to the overall evaluation result. Therefore, the thirteenth prompt is input into the optimized large language model to obtain the overall evaluation result of the original text sorted by importance, and the overall evaluation result is inserted at the end of the original text. The overall evaluation result of the original text of this embodiment is a fusion of the local evaluation result, the global evaluation result and the industry evaluation result, and the local evaluation result, the global evaluation result and the industry evaluation result are all evaluated from different dimensions. Therefore, this embodiment can combine all the previous evaluation results to provide a comprehensive list of suggestions for improvement after sorting. This multi-level, multi-angle evaluation method can provide comprehensive, in-depth and constructive review opinions on the text, greatly improving the efficiency and quality of text review.

[0127] It should be noted that the optimized large language model of this embodiment constructs a first fine-tuning data set during training, and the fine-tuning data pairs in the first fine-tuning data set reflect the mapping relationship between prompt information and output content. In order to enable the optimized large language model to implement polishing evaluation (i.e., preliminary evaluation), intra-chapter or cross-chapter evaluation, industry evaluation, etc., this embodiment can update the fine-tuning data pairs in the first fine-tuning data set, and update the prompt information and output content according to the steps of the above-mentioned polishing evaluation (i.e., preliminary evaluation), intra-chapter or cross-chapter evaluation, and industry evaluation, so that the data in the first fine-tuning data set is richer, and the model finally trained can meet the evaluation requirements of the paper.

[0128] In summary, the method of this embodiment has at least the following advantages:

[0129] (1) Significantly improve review efficiency: The review time for a single paper was shortened from 1 hour to 3 minutes, greatly improving the processing speed of paper reviews.

[0130] (2) Improving review quality: Through multi-dimensional and fine-grained evaluation, the comprehensiveness and depth of the review are ensured, significantly surpassing traditional review methods.

[0131] (3) Accurate positioning of review opinions: Intelligent annotation technology can accurately insert review opinions into corresponding positions, greatly improving the operability of review opinions.

[0132] (4) Structured analysis capability: It innovatively implements automatic analysis of the chapter structure of a paper, which helps to evaluate the overall logic of the paper.

[0133] (5) Consistency of evaluation criteria: The evaluation system based on the big model ensures the consistency of the evaluation criteria and reduces the evaluation deviation caused by human factors.

[0134] (6) Strong adaptability: The system can handle academic papers of different disciplines and types and has wide applicability.

[0135] (7) Continuous optimization capability: By continuously accumulating review data and feedback, the system can continuously optimize the review model and improve the quality of the review.

[0136] Based on the above embodiments, this embodiment also provides a text review system. Figure 5 As shown in, the system includes: a text processing module 10, an object extraction module 20, a text evaluation module 30 and a text evaluation module 40. The text processing module 10 is used to obtain the original text and process the original text to obtain the processed text. The object extraction module 20 is used to extract objects from the processed text to obtain a plurality of extracted objects. The text evaluation module 30 is used to perform local evaluation, global evaluation and industry evaluation on a plurality of extracted objects among the plurality of extracted objects through an optimized large language model, and obtain a local evaluation result, a global evaluation result and an industry evaluation result for each extracted object respectively. The text evaluation module 40 is used to obtain an overall evaluation result of the original text based on the local evaluation result, the global evaluation result and the industry evaluation result.

[0137] The working principles of each module in the text review system of this embodiment are the same as the principles of each step in the above method embodiment, and will not be repeated here.

[0138] Each module in the above text review system can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in the terminal in the form of hardware, or can be stored in the memory in the terminal in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0139] Based on the above embodiment, the present invention further provides a terminal, the principle block diagram of the terminal can be as follows: Figure 6 The terminal may include one or more processors 100 ( Figure 6 The processor 100 includes a memory 101 and a computer program 102 stored in the memory 101 and executable on one or more processors 100, for example, a sleep analysis program based on multi-sensor data. When the one or more processors 100 execute the computer program 102, the various steps in the embodiment of the sleep analysis method based on multi-sensor data can be implemented. Alternatively, when the one or more processors 100 execute the computer program 102, the functions of the various modules / units in the embodiment of the sleep analysis device based on multi-sensor data can be implemented, which is not limited here.

[0140] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0141] In one embodiment, the memory 101 may be an internal storage unit of an electronic device, such as a hard disk or memory of the electronic device. The memory 101 may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Further, the memory 101 may also include both an internal storage unit of the electronic device and an external storage device. The memory 101 is used to store computer programs and other programs and data required by the terminal. The memory 101 may also be used to temporarily store data that has been output or is to be output.

[0142] Those skilled in the art will understand that Figure 6 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0143] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, operating database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double operational data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A text review method, characterized in that: The method comprises: Obtaining an original text, and processing the original text to obtain a processed text; Performing object extraction on the processed text to obtain a number of extracted objects; Performing local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects by using the optimized large language model, and obtaining local evaluation results, global evaluation results and industry evaluation results of each extraction object respectively; Obtaining an overall evaluation result of the original text according to the local evaluation result, the global evaluation result and the industry evaluation result; The optimized large language model is used to perform local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects, and local evaluation results, global evaluation results and industry evaluation results of each extraction object are obtained respectively, including: Performing local evaluation on text paragraphs of multiple extraction objects among the plurality of extraction objects to obtain a local evaluation result of each extraction object; Acquire structured text catalogs of multiple extracted objects among the plurality of extracted objects, and evaluate at least one text paragraph of the multiple extracted objects in the structured text catalog based on the multiple extracted objects and the structured text catalog to obtain a global evaluation result of each extracted object; Performing industry evaluation on text paragraphs of multiple extracted objects among the plurality of extracted objects to obtain industry evaluation results; Obtaining structured text directories of multiple extracted objects among the plurality of extracted objects, and evaluating at least one text paragraph of the multiple extracted objects in the structured text directory based on the multiple extracted objects and the structured text directory to obtain a global evaluation result of each extracted object, including: Based on the text paragraphs of the processed text, the structured text catalog is obtained, wherein the structured text catalog contains multiple chapters and multiple sections; The step of obtaining the structured text directory based on the text paragraphs of the processed text includes: Based on all text paragraphs of the processed text and a preset second prompt template, a third prompt is obtained; Inputting the third prompt into the optimized large language model to obtain a text directory; Based on the main text directory and a preset third prompt template, a fourth prompt is obtained, and the fourth prompt is input into the optimized large language model to obtain a structured main text directory.

2. The text review method according to claim 1, characterized in that: The original text is processed to obtain a processed text, including: Segmenting the original text to obtain a plurality of original text paragraphs; Converting each of the plurality of original text paragraphs into a text paragraph in a target format to obtain a formatted text paragraph; Adding a number index before each of the formatted text paragraphs to obtain a text paragraph containing the number index; According to the text paragraphs containing the number indexes, a processed text is obtained.

3. The text review method according to claim 2, characterized in that: Object extraction is performed on the processed text to obtain a number of extracted objects, including: A regular expression is used to match each text paragraph of the processed text to obtain a plurality of extraction objects, the text paragraphs of the plurality of extraction objects, and the serial number indexes of the text paragraphs of the plurality of extraction objects.

4. The text review method according to claim 2, characterized in that: Object extraction is performed on the processed text to obtain a number of extracted objects, including: Based on each text paragraph of the processed text and the preset first prompt, a combined text paragraph is obtained; The combined text paragraph is input into the optimized large language model to obtain a plurality of extraction objects, the text paragraphs of the plurality of extraction objects and the number indexes of the text paragraphs of the plurality of extraction objects.

5. The text review method according to claim 1, characterized in that: Performing local evaluation on the text paragraphs of multiple extraction objects among the plurality of extraction objects to obtain local evaluation results of each extraction object, including: for each extraction object among the multiple extraction objects, respectively screening out corresponding text paragraphs in the processed text, and respectively constructing a second prompt for each extraction object; The second prompt is input into the optimized large language model to perform local evaluation on each extracted object to obtain a local evaluation result of each extracted object.

6. The text review method according to claim 5, characterized in that: The step of respectively constructing a second hint for each extracted object comprises: Get the preset first prompt template; Based on a preset first prompt template and a text paragraph of each extracted object, a second prompt of each extracted object is obtained.

7. The text review method according to claim 1, characterized in that: The method further comprises: obtaining structured text directories of multiple extracted objects among the plurality of extracted objects, evaluating at least one text paragraph of the multiple extracted objects in the structured text directory based on the multiple extracted objects and the structured text directory to obtain a global evaluation result of each extracted object; Evaluate at least one text paragraph of the multiple extraction objects in the structured text catalog according to the text paragraphs of the multiple extraction objects and the structured text catalog to obtain cross-chapter evaluation results of the multiple extraction objects; Evaluate the contents of each chapter in the structured text catalog of the plurality of extracted objects respectively to obtain evaluation results of each chapter of the plurality of extracted objects; Evaluate each section of the structured text catalog of the plurality of extraction objects respectively to obtain evaluation results of each section of the plurality of extraction objects; A global evaluation result of each extracted object is obtained according to the per-chapter evaluation results and the per-section evaluation results of the multiple extracted objects.

8. The text review method according to claim 7, characterized in that: Evaluating at least one text paragraph of the multiple extraction objects in the structured text catalog according to the text paragraphs of the multiple extraction objects and the structured text catalog to obtain cross-chapter evaluation results of the multiple extraction objects, including: Based on all text paragraphs of the processed text and a preset fourth prompt template, a fifth prompt is obtained, and the fifth prompt is input into the optimized large language model to obtain a text intention conclusion; A sixth prompt is obtained based on at least one text paragraph of the multiple extraction objects, the structured text directory and a preset fifth prompt template, and the sixth prompt is input into the optimized large language model to obtain a cross-chapter evaluation result.

9. The text review method according to claim 8, characterized in that: Evaluate the contents of each chapter in the structured text catalog of the plurality of extraction objects respectively to obtain the evaluation results of each chapter, including: Based on the structured text catalog, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each chapter, the text paragraphs of each chapter, and the preset sixth prompt template, a seventh prompt is obtained; Input the seventh prompt into the optimized large language model to obtain evaluation results for several chapters; Based on the evaluation result of each chapter, the title of each chapter, the text paragraph of each chapter, the current system time and the seventh prompt template, an eighth prompt is obtained; The eighth prompt is input into the optimized large language model to obtain evaluation results for each chapter of the multiple extraction objects.

10. The text review method according to claim 9, characterized in that: Evaluate each section of the structured text catalog of the plurality of extraction objects respectively to obtain evaluation results of each section of the plurality of extraction objects, including: Based on the structured text catalog, the text intention conclusion, the cross-chapter evaluation result, the title of the processed text, the title of each section, the text paragraphs of each section, and the preset eighth prompt template, a ninth prompt is obtained; Inputting the ninth prompt into the optimized large language model to obtain evaluation results of several sections; Based on the evaluation result of each section, the title of each section, the text paragraph of each section, the current system time and the ninth prompt template, a tenth prompt is obtained; The tenth prompt is input into the optimized large language model to obtain evaluation results for each section of the multiple extraction objects.

11. The text review method according to claim 8, characterized in that: After obtaining a global evaluation result of each extracted object according to the evaluation results of each chapter and the evaluation results of each section of the plurality of extracted objects, the method further includes: Respectively obtaining the number indexes corresponding to the text paragraphs of each chapter and each section of the plurality of extraction objects in the structured text directory; Inserting the evaluation result of each chapter into the first numbered index of the corresponding chapter according to the numbered index corresponding to the text paragraph of each chapter; The evaluation result of each section is inserted into the first numbered index of the corresponding section according to the numbered index corresponding to the text paragraph of each section.

12. The text review method according to claim 1, characterized in that: Perform industry evaluation on text paragraphs of multiple extracted objects among the plurality of extracted objects to obtain industry evaluation results, including: Obtaining an eleventh prompt according to the text information input by the user and the preset tenth prompt template; Inputting the eleventh prompt into the optimized large language model to obtain a preliminary evaluation result of each extracted object; Based on the preliminary evaluation result and the preset eleventh prompt template, a twelfth prompt is obtained; The twelfth prompt is input into the optimized large language model to obtain the industry evaluation result of each extracted object.

13. The text review method according to claim 1, characterized in that: The overall evaluation result of the original text is obtained according to the local evaluation result, the global evaluation result and the industry evaluation result, including: Based on the local evaluation result, the global evaluation result, the industry evaluation result, the text title, the user type and the preset twelfth prompt template, a thirteenth prompt is obtained; The thirteenth prompt is input into the optimized large language model to obtain an overall evaluation result of the original text sorted by importance, and the overall evaluation result is inserted at the end of the original text.

14. The text review method according to any one of claims 1 to 13, characterized in that: The training process of the optimized large language model includes: Classify the texts in the text library and sample each text category to obtain text sampling data; For each text in the text sampling data, construct a fine-tuning data pair, wherein the fine-tuning data pair consists of a prompt and output content of a preset large language model; Obtain a fine-tuning data set according to the fine-tuning data pair; Constructing a first fine-tuning dataset based on the fine-tuning dataset and a preset general dataset; The fine-tuning data set in the first fine-tuning data set is processed to obtain a second fine-tuning data set, and the second fine-tuning data set and a preset large language model are fine-tuned to obtain the optimized large language model.

15. A text review system, characterized in that: The system is used to implement the steps of the text review method according to any one of claims 1 to 14, and the system comprises: A text processing module, used to obtain the original text and process the original text to obtain a processed text; An object extraction module, used to extract objects from the processed text to obtain a number of extracted objects; A text evaluation module, used to perform local evaluation, global evaluation and industry evaluation on multiple extraction objects among the plurality of extraction objects through the optimized large language model, and obtain local evaluation results, global evaluation results and industry evaluation results of each extraction object respectively; An overall evaluation module is used to obtain an overall evaluation result of the original text according to the local evaluation result, the global evaluation result and the industry evaluation result.

16. A terminal, characterized in that: The terminal includes a memory, a processor, and a text review program stored in the memory and executable on the processor. When the processor executes the text review program, the steps of the text review method according to any one of claims 1 to 14 are implemented.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a text review program, and when the text review program is executed by the processor, the steps of the text review method according to any one of claims 1 to 14 are implemented.

Images