An AI-based multi-modal office document intelligent generation method and system

Abstract

The application belongs to the technical field of document intelligent generation, and provides a multi-modal office document intelligent generation method and system based on AI, which comprises the following steps: through coincidence analysis on a meeting minutes and historical office documents contained in a historical office document library, similar office documents are extracted, according to the enterprise personnel review records of the similar office documents, review situation analysis is carried out, main review personnel are screened out, paragraph content similarity analysis is carried out on each paragraph in the similar office documents and the summary report, similar paragraphs are obtained, and similar paragraph content degree proportion analysis is carried out, main review personnel of the paragraphs are obtained, document layout habits of each main review personnel are obtained, the final version of the meeting minutes and the coincident content of the similar office documents are marked as marked content, the final version of the meeting minutes is subjected to multi-modal office document intelligent generation by using an AI intelligent generation system, and the main review personnel are pushed, so that the generation efficiency and adaptability of the office document are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent document generation, in particular to an AI-based multi-modal office document intelligent generation method and system. BACKGROUND

[0002] In multi-modal office document generation, AI-driven multi-modal office document intelligent generation has become the core direction to improve office efficiency. As an important carrier of office documents, meeting minutes directly generate office documents, leading to uniform templates and failing to solve the adaptability problem of multiple participants. Therefore, it is necessary to combine AI to generate office documents intelligently through multi-modal to solve the adaptability of office documents and improve the information acquisition efficiency of reviewers.

[0003] In the prior art, there is a lack of similarity analysis of meeting minutes and historical office documents when generating office documents from meeting minutes, which leads to the inability to obtain document review information of meeting participants from historical office documents. At the same time, there is a lack of systematic arrangement of reviewer behavior data for the retrieved similar documents, which cannot accurately filter out core reviewers and their document layout habits. In addition, in the personalized layout generation link, the association mechanism of "paragraph content-main reviewer layout habits" is not established, only a unified format template is used, which cannot dynamically adjust the layout according to the font preferences and line spacing requirements of different paragraph main reviewers, and there is no targeted push, leading to insufficient adaptability of the generated multi-modal office documents and difficulty in meeting diversified review needs.

[0004] Therefore, the application provides an AI-based multi-modal office document intelligent generation method and system. SUMMARY

[0005] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background art.

[0006] The technical solution adopted by the application to solve its technical problems is: an AI-based multi-modal office document intelligent generation method and system.

[0007] By performing coincidence analysis on the meeting minutes and the historical office documents contained in the historical office document library, similar office documents are extracted from the historical office document library;

[0008] According to the enterprise personnel review records of the similar office documents, the review records of each same meeting enterprise personnel are analyzed to filter out the main reviewers;

[0009] obtaining a summary report of the similar office documents of the main reviewer, and similar pairing and integrating the summary report with the content in the similar office documents to obtain a similar paragraph archive, analyzing the content proportion of each similar paragraph in the similar paragraph archive to obtain a paragraph main reviewer of each paragraph of the similar office document;

[0010] By analyzing the similar paragraph archive of the paragraph main reviewer in different historical meetings, the historical document layout of each paragraph main reviewer is obtained, and the marked content is obtained by marking the final draft of the meeting minutes and the overlapping content of the similar office documents. The final draft of the meeting minutes is generated by using an AI intelligent generation system, and the main reviewer is pushed.

[0011] Further, the similar office documents are extracted from the historical office document library as follows:

[0012] Obtain the meeting minutes and count the list of participating enterprise personnel to obtain a participating enterprise personnel table;

[0013] Pretreat the meeting minutes to obtain a final draft of the meeting minutes;

[0014] Extract the main content of the final draft of the meeting minutes, input the main content of the meeting minutes into the historical office document library, and obtain a document coincidence value by content coincidence comparison. If the document coincidence value is greater than or equal to the coincidence threshold value, the corresponding office document in the historical office document library is marked as a similar office document.

[0015] Further, the main reviewer is selected by analyzing the review records of each participating enterprise personnel as follows:

[0016] Obtain the participating enterprise personnel table of the similar office documents, compare it with the participating enterprise personnel table of this time, mark the personnel participating in both meetings, and obtain the same participating enterprise personnel;

[0017] Statistical the review duration of the same participating enterprise personnel when reviewing the similar office documents, and obtain the effective review duration by pretreatment. If the effective review duration of the same participating enterprise personnel is greater than or equal to the preset review duration threshold, the corresponding same participating enterprise personnel is marked as a suspected main reviewer;

[0018] Statistical the number of highlighted annotations and effective annotations of the suspected main reviewer during the review process, and add them to obtain the interaction frequency. The interaction frequency is processed by ratio with the effective review duration to obtain an effective interaction value. If the effective interaction value is greater than or equal to the effective interaction threshold, the suspected main reviewer is marked as a main reviewer.

[0019] Further, the summary report of the similar office documents of the main reviewer is obtained as follows:

[0020] After the main reviewer completes the review of the office document, the corresponding document content is highlighted and the effective annotations of the corresponding document content are collected and summarized, and a summary report is written to summarize the core content of the office document required by each main reviewer, the findings in the review process, and the personal insights and suggestions.

[0021] Further, the summary report is similar to the content in the similar office document, and the similar paragraph archive is obtained by the following process:

[0022] The paragraph content similarity analysis is performed on the document paragraphs in the similar office document, and the paragraph content similarity value is obtained and compared with the paragraph content similarity threshold;

[0023] If the paragraph content similarity value is greater than or equal to the paragraph content similarity threshold, the corresponding paragraph is marked as a similar paragraph;

[0024] The similar paragraphs are integrated to obtain a similar paragraph archive.

[0025] Further, the content proportion of each similar paragraph in the similar paragraph archive is analyzed as follows:

[0026] The total number of paragraphs in the summary report and the total number of words in the summary report are counted, and paragraph proportion analysis and word proportion analysis are performed respectively;

[0027] The number of similar paragraphs is processed by ratio with the total number of paragraphs in the summary report to obtain a similar paragraph proportion value;

[0028] The total number of words in all similar paragraphs is processed by ratio with the total number of words in the summary report to obtain a similar paragraph word proportion value;

[0029] The similar paragraph proportion value and the similar paragraph word proportion value are added to obtain a similar paragraph content degree proportion value;

[0030] The similar paragraph content degree proportion values of each main reviewer are compared in size, and the main reviewer with the largest similar paragraph content degree proportion value is marked as a paragraph main reviewer.

[0031] Further, the historical document layout of each paragraph main reviewer is obtained, and the marked content is obtained by marking the coincident content of the final draft of the meeting minutes and the similar office document, and the process is as follows:

[0032] The similar paragraph archive in the historical meeting of each paragraph main reviewer is analyzed to obtain the historical document layout of each paragraph main reviewer, and the document layout habits of each paragraph main reviewer are extracted;

[0033] Mark the main content in the similar office documents and the final version of the meeting minutes at the same time to obtain the marked content.

[0034] Further, the AI intelligent generation system generates the final version of the meeting minutes in a multi-modal office document, and the process is as follows:

[0035] The marked content is marked in the paragraph where it is located in the similar office document, and the marked content, the corresponding paragraph, the paragraph editor, and the document layout habit of the paragraph editor of each paragraph of the similar document are all input into the AI intelligent generation system. The AI intelligent generation system determines the paragraph editor of the marked content according to the marked content.

[0036] The AI intelligent generation system performs layout on the final version of the meeting minutes according to the document layout habit of each paragraph editor, and performs office document intelligent generation on the final version of the meeting minutes according to the document layout habit of the paragraph editor.

[0037] Further, the process of pushing the main reviewer is as follows:

[0038] Mark the review time period of each main reviewer reviewing the office document each time on the time axis to obtain a set of review time periods.

[0039] Statistically analyze the set of review time periods to obtain the frequency of occurrence of each review time period, sort the review time periods in descending order of frequency of occurrence, and select the review time period with the highest frequency as the peak time period of the main reviewer.

[0040] Extract the peak time period of each main reviewer reviewing the office document, and push the intelligently generated office document within the peak time period of the main reviewer.

[0041] If there are multiple peak time periods of the main reviewer, the peak time periods are pushed according to the order of the peak time periods.

[0042] If the peak time periods of multiple main reviewers are in the same time period, the multiple main reviewers are pushed simultaneously.

[0043] An AI-based multi-modal office document intelligent generation system includes the following modules:

[0044] Overlap analysis module: by analyzing the overlap between the meeting minutes and the historical office documents contained in the historical office document library, similar office documents are extracted from the historical office document library.

[0045] Filtering module: according to the enterprise personnel review records of the similar office documents, analyze each enterprise personnel review record, and select the main reviewer.

[0046] The proportion analysis module: obtains a summary report of the main reviewer of the similar office document, and integrates the summary report with the content in the similar office document for similar pairing, to obtain a similar paragraph archive, and analyzes the content proportion of each similar paragraph in the similar paragraph archive, to obtain the paragraph main reviewer of each paragraph of the similar office document;

[0047] The intelligent generation module: analyzes the similar paragraph archive of the paragraph main reviewer in different historical meetings to obtain the historical document layout of each paragraph main reviewer, marks the overlapping content between the final draft of the meeting minutes and the similar office document, obtains the marked content, uses an AI intelligent generation system to generate a multi-modal office document from the final draft of the meeting minutes, and pushes the main reviewer.

[0048] The beneficial effects of the present application are as follows:

[0049] (1) By analyzing the overlapping of the meeting minutes and the historical office documents contained in the historical office document library, similar office documents in the historical office document library are extracted, the enterprise personnel review records of the similar office documents are obtained, the effective review time is obtained, the main reviewers are selected according to the effective review time, and the review information of the reviewers is obtained by analyzing the similar office documents in the historical office documents, which provides a personnel basis for subsequent analysis of the paragraph main reviewer, is conducive to the statistics of the main content required by the enterprise personnel, and is convenient for subsequent generation of multi-modal office documents.

[0050] (2) According to the summary report of the main reviewer, the similarity of each paragraph in the similar office document and the summary report is analyzed, the paragraphs with a similarity exceeding the paragraph similarity threshold are marked as similar paragraphs, the paragraph content similarity analysis and the similar paragraph degree content proportion analysis are carried out based on the similar paragraphs, the paragraph main reviewer is confirmed, the document layout habit of the paragraph main reviewer is obtained, the part consistent with the similar document content in the final draft of the meeting minutes is generated into a multi-modal office document by an AI intelligent generation system, the main review time period of the main reviewer is obtained, and the office document is pushed, the main reviewer of each paragraph is screened out by analyzing the content proportion of each paragraph, which is conducive to personalized layout of the main reviewer of each paragraph, improves the readability and pertinence of the office document, in addition, the office document is pushed according to the main review time period of the main reviewer, which not only meets the individual needs of different reviewers, but also improves the generation efficiency and adaptability of the office document. BRIEF DESCRIPTION OF DRAWINGS

[0051] The present application will be further described below with reference to the accompanying drawings.

[0052] Figure 1is a step flow chart of an AI-based multi-modal office document intelligent generation method according to Embodiment 1 of the present application;

[0053] Figure 2 is a step flow chart of an AI-based multi-modal office document intelligent generation method according to Embodiment 2 of the present application;

[0054] Figure 3 is a judgment flow chart in an AI-based multi-modal office document intelligent generation method according to the present application;

[0055] Figure 4 is a program block diagram of an AI-based multi-modal office document intelligent generation system according to Embodiment 3 of the present application. DETAILED DESCRIPTION

[0056] In order to make the technical means, creative features, purposes and effects achieved by the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0057] Embodiment 1: Please refer to Figure 1 - Figure 3 The AI-based multi-modal office document intelligent generation method according to the present application embodiment includes:

[0058] Step 1: Extract similar office documents from the historical office document library by performing coincidence analysis on the meeting minutes and the historical office documents contained in the historical office document library;

[0059] Obtain the preliminary draft of the meeting minutes, count the list of participating enterprise personnel, and obtain the participating enterprise personnel table;

[0060] Preprocess the preliminary draft of the meeting minutes, which includes removing redundancy, filling in missing parts, disambiguating, etc., to obtain the final draft of the meeting minutes;

[0061] Among them, removing redundancy is to remove colloquial meaningless expressions in the preliminary draft of the meeting minutes to ensure that the meeting minutes are clean and clear, such as "um", "right" and the like;

[0062] Filling in missing parts is to combine the meeting audio to correct the missing parts in the preliminary draft of the meeting minutes to ensure the accuracy of the content of the preliminary draft of the meeting minutes, such as "promote project operation next week", which does not clearly indicate whether the project operation time is Monday or Friday, and the missing part is filled in combination with the meeting audio;

[0063] Disambiguation is to clearly define the unclear content in the preliminary draft of the meeting minutes, for example, the expression "this project" lacks a clear direction, if the current meeting involves two projects, ambiguity will occur, causing unclear designation;

[0064] The main content of the meeting minutes is obtained by content extraction, and the main content of the meeting minutes is obtained by content extraction. The main content includes key issues discussed in the meeting, consensus reached, tasks allocated and follow-up action plans, etc.

[0065] The main content of the meeting minutes is input into the historical office document library, and the document coincidence value is obtained by content coincidence comparison, and compared with the coincidence threshold value;

[0066] If the document coincidence value is greater than or equal to the coincidence threshold value, the corresponding office document in the historical office document library is marked as a similar office document;

[0067] If the document coincidence value is less than the coincidence threshold value, no processing is performed;

[0068] It should be noted that the historical office document library is obtained by recording each meeting to obtain the meeting minutes, and generating office documents through the meeting minutes. Each office document can obtain the main content of the corresponding meeting;

[0069] Specifically, the method for obtaining the document coincidence value is to analyze the similarity of the text content by using natural language processing technology, such as TF-IDF (Term Frequency-Inverse Document Frequency) algorithm, cosine similarity calculation, etc. By identifying the key words in the text and evaluating their distribution in different historical office documents, by comparing the key words distribution of the meeting minutes and the historical office documents in the historical office document library, a numerical similarity index can be obtained, which is denoted as the document coincidence value.

[0070] The purpose of obtaining similar office documents is to further mine the potential value of similar office documents by analyzing similar office documents, which can provide strong support for enterprise knowledge management and office document reuse. Specifically, the analysis of similar office documents can reveal the frequently discussed topics, key issues and consensus reached in the document. In addition, by obtaining the main reviewers of similar office documents, the personnel screening basis for subsequent paragraph main reviewer analysis is provided, and the paragraph main reviewer's layout habits statistics provide data support for subsequent intelligent generation of personalized multi-modal office documents that meet the reviewer's needs;

[0071] Step 2: According to the enterprise personnel review records of similar office documents, analyze each same participant enterprise personnel review record, and screen out the main reviewers;

[0072] According to the similar office documents, the corresponding participant enterprise personnel table is obtained, and compared with the participant enterprise personnel table of this time, the participants of the two meetings are marked, and the same participant enterprise personnel are obtained.

[0073] obtain the review records of the co-attending enterprise personnel, count the review time length of the co-attending enterprise personnel when reviewing similar office documents, and perform preprocessing to obtain an effective review time length;

[0074] It should be noted that the effective review time length needs to remove the invalid time length in the review time length, and the invalid time length includes but is not limited to the hanging time length of opening the document board without actually reviewing, and the time length of quickly scrolling the page without actually reviewing, etc.

[0075] According to the effective review time length, the co-attending enterprise personnel are arranged in descending order to obtain the co-attending enterprise personnel effective review time length descending order;

[0076] Based on the co-attending enterprise personnel effective review time length descending order, the co-attending enterprise personnel effective review time length is compared with a preset review time length threshold;

[0077] If the co-attending enterprise personnel effective review time length is greater than or equal to the preset review time length threshold, the corresponding co-attending enterprise personnel is marked as a suspected main reviewer;

[0078] If the co-attending enterprise personnel effective review time length is less than the preset review time length threshold, the corresponding co-attending enterprise personnel is marked as a non-main reviewer;

[0079] Obtain the total interaction frequency of the suspected main reviewer when reviewing similar office documents, count the number of highlight annotations, the number of effective annotations, etc. in the review process of the suspected main reviewer, and add them up to obtain the interaction frequency. The interaction frequency is processed by ratio with the effective review time length to obtain an effective interaction value, and the effective interaction value is compared with a preset effective interaction threshold;

[0080] If the effective interaction value is greater than or equal to the effective interaction threshold, the suspected main reviewer is marked as a main reviewer, and the main reviewer is numbered in descending order of effective review time length, for example, a, b, c, …;

[0081] If the effective interaction value is less than the effective interaction threshold, the suspected main reviewer is marked as a non-main reviewer;

[0082] It should be noted that the highlight annotation includes keyword search in the document, and the effective annotation includes content copying.

[0083] The purpose of screening the main reviewers is to provide basic personnel data for the main reviewers of the subsequent analysis paragraph, so as to individualize the layout for each paragraph of the main reviewer, improve the pertinence and readability of the office document, and the main reviewer usually has a deep understanding and review habit of the document content. By analyzing their review records, the preferences of the main reviewer for the layout of the document, such as font size, line spacing, paragraph format, etc. can be obtained, so that the adaptability and review efficiency of the document can be improved when the intelligent document is generated.

[0084] Embodiment 2: please refer to Figure 2 Figure 3 As shown in the figure, the AI-based multi-modal office document intelligent generation method according to the embodiment of the application comprises:

[0085] Step three: obtaining the summary report of the main reviewer reviewing the similar office document, and similar pairing and integrating the summary report with the content in the similar office document to obtain a similar paragraph archive, and analyzing the content proportion of each similar paragraph in the similar paragraph archive to obtain the main reviewer of each paragraph of the similar office document;

[0086] The summary report of the main reviewer reviewing the similar office document is obtained, wherein the summary report is usually compiled by the main reviewer after completing the review of the office document, and the summary report is compiled by the main reviewer after completing the review of the office document. The summary report is used to summarize the core content of the required office document, the findings in the review process, and the personal opinions and suggestions.

[0087] It should be noted that the union of the highlighted annotations and the effective annotations is arranged (specifically, the document content corresponding to the highlighted annotations and the document content corresponding to the effective annotations are merged), and if the document content corresponding to the highlighted annotations and the document content corresponding to the effective annotations have some overlapping content, the overlapping content is union arranged to avoid repeated description of the same content in the summary report compilation;

[0088] The paragraphs in the similar office document are marked to obtain document paragraph marks, which are denoted as 1, 2, 3, …, n, wherein 1 represents the first paragraph of the similar office document, and n represents the last paragraph of the similar office document.

[0089] Similarly, the paragraphs in the summary report are marked to obtain report paragraph marks, which are denoted as 1, 2, 3, …, m, wherein 1 represents the first paragraph of the summary report, and m represents the last paragraph of the summary report.

[0090] The summary report is similar to the content in the similar office document, and the specific process is as follows:

[0091] ​The paragraph content similarity values Di of the similar office document paragraphs are analyzed respectively, and compared with the paragraph content similarity threshold Dy to screen out the summary report paragraphs similar to the similar office document paragraphs, and the similar paragraphs are obtained;

[0092] If the paragraph content similarity value is greater than or equal to the paragraph content similarity threshold Dy, the corresponding paragraph is marked as a similar paragraph;

[0093] If the paragraph content similarity value is less than the paragraph content similarity threshold Dy, the corresponding paragraph is marked as a non-similar paragraph;

[0094] The similar paragraphs are integrated to obtain a similar paragraph archive;

[0095] It should be noted that the paragraph content similarity value Di represents the similarity of the paragraphs in the summary report and the paragraphs in the similar office document, and the value of i is determined by the document paragraph mark and the report paragraph mark. For example, the paragraph content similarity value D31 of the third paragraph in the summary report and the first paragraph in the similar office document, wherein 3 represents the third paragraph in the summary report, and 1 represents the first paragraph in the similar office document;

[0096] The similar paragraphs in the similar paragraph archive are counted and analyzed for the similar paragraph content degree proportion, to obtain the paragraph main review personnel of each paragraph;

[0097] Similar paragraph content degree proportion analysis: the total number of paragraphs in the summary report and the total number of words in the summary report are counted, and paragraph proportion analysis and word proportion analysis are performed respectively to obtain the paragraph main review personnel of each paragraph in the similar office document;

[0098] The analysis method from the paragraph proportion dimension is as follows:

[0099] The similar paragraph number and the total number of paragraphs in the summary report are processed by ratio to obtain the similar paragraph proportion value;

[0100] The analysis method from the word proportion dimension is as follows:

[0101] The total number of words of all similar paragraphs and the total number of words in the summary report are processed by ratio to obtain the similar paragraph word proportion value;

[0102] The similar paragraph proportion value and the similar paragraph word proportion value are added to obtain the similar paragraph content degree proportion value;

[0103] According to the similar paragraph content degree proportion value, the paragraph main review personnel is confirmed;

[0104] The similar paragraph content degree proportion values of each main review personnel are compared in size, and the main review personnel with the largest similar paragraph content degree proportion value is marked as the paragraph main review personnel;

[0105] For example, the main reviewers of the similar office documents are main reviewers a, b, c, and d. According to the summary report of the main reviewer a, the paragraph content similarity analysis is performed on the document paragraph 1, the paragraph content similarity values between the document paragraph 1 and the report paragraphs in the summary report are calculated respectively, and each paragraph content similarity value is compared with the paragraph similarity threshold value. Through the paragraph content similarity analysis, it is obtained that the report paragraph 1, the report paragraph 2, and the report paragraph 4 in the summary report are similar to the document paragraph 1;

[0106] The paragraph content degree proportion analysis is performed on the report paragraph 1, the report paragraph 2, and the report paragraph 4 in the summary report, the report paragraph 1, the report paragraph 2, and the report paragraph 4 are compared with the total paragraph in the summary report, and the similar paragraph proportion value is obtained;

[0107] The total number of words of the report paragraph 1, the report paragraph 2, and the report paragraph 4 is counted, and the total number of words of the summary report is processed by ratio, and the similar paragraph word proportion value is obtained;

[0108] The similar paragraph proportion value and the similar paragraph word proportion value are added to obtain the similar paragraph content degree proportion value;

[0109] Similarly, the similar paragraph content degree proportion values of the main reviewers b, c, and d are obtained;

[0110] The similar paragraph content degree proportion values of the paragraph main reviewers a, b, c, and d are compared in size. If the similar paragraph content degree proportion value of the main reviewer b is the largest, the main reviewer b is marked as the paragraph main reviewer of the paragraph 1;

[0111] Step four: through the analysis of the similar paragraph archiving of the paragraph main reviewer in different historical meetings, the historical document layout of each paragraph main reviewer is obtained, the overlapping content between the final draft of the meeting minutes and the similar office documents is marked to obtain the marked content, the AI intelligent generation system is used to generate the multi-modal office document intelligently, and the main reviewer is pushed;

[0112] The similar paragraph archiving of each paragraph main reviewer in the historical meetings is analyzed to obtain the historical document layout of each paragraph main reviewer, and the document layout habit of each paragraph main reviewer is extracted;

[0113] For example, the document layout habit of the paragraph main reviewer a is: text format: font is Microsoft Yahei, font size is five, paragraph format: line spacing is 1 times, first line indentation is 2 characters, paragraph spacing is 0.5 lines before paragraph, and 0.5 lines after paragraph, etc.

[0114] The document layout habits of paragraph supervisor b are as follows: text format: font is Kai Ti, font size is four, paragraph format: line spacing is 1 times, first line indent is 2 characters, inter-paragraph spacing is 0 lines before and after the paragraph, and relevant pictures are inserted in the text;

[0115] The main content in similar office documents and meeting minutes is marked to obtain the marked content;

[0116] The paragraph where the marked content is located in the similar office document is marked, and the marked content of each paragraph of the similar document, the corresponding paragraph supervisor, and the document layout habits of the paragraph supervisor are recorded into the AI intelligent generation system. The AI intelligent generation system determines the paragraph supervisor of the marked content according to the marked content;

[0117] According to the document layout habits of each paragraph supervisor, the meeting minutes are processed;

[0118] The marked content is laid out according to the review habits of each paragraph supervisor, including font, font size, line spacing, inter-paragraph spacing, etc. Format adjustment to ensure that the generated office document conforms to the document layout habits of each paragraph supervisor;

[0119] According to the meeting process order, the meeting minutes are laid out, the paragraph supervisor of the similar office document corresponding to the marked content is obtained, and the AI intelligent generation system is used to generate the office document according to the document layout habits of the paragraph supervisor;

[0120] At the same time, the AI intelligent generation system can also adjust the content of the meeting minutes intelligently according to the common words and expressions of each paragraph supervisor, so that the generated office document meets the needs of each paragraph supervisor;

[0121] For the main content difference part, the enterprise document layout specification is used for layout;

[0122] Based on the AI intelligent generation system, the office document is generated, and the office document is pushed according to the main review personnel review office document time period. The specific process is as follows:

[0123] Obtain the time period of each main review personnel historical review office document, analyze the time period of multiple historical review office documents, and obtain the peak time period of each main review personnel review office document. The intelligent generated office document is pushed in the peak time period of the main review personnel;

[0124] The selection process of the peak time period is as follows:

[0125] Mark the review time period of each main review personnel each time on the time axis to obtain a set of review time periods;

[0126] The review time period set is counted to obtain the occurrence frequency of each review time period, the review time periods are sorted according to the occurrence frequency from high to low, and the review time period with the highest frequency in the sorted review time period is marked as the peak time period of the main reviewer;

[0127] If there are multiple review time periods with the same frequency and the highest frequency, the multiple review time periods are marked as the peak time period of the main reviewer;

[0128] When pushing the office document, if there are multiple peak time periods of the main reviewer, the peak time periods are pushed according to the order;

[0129] If multiple peak time periods of the main reviewer exist in the same time period, the multiple main reviewers are pushed simultaneously;

[0130] For example, the review office document time period of the main reviewer a and the main reviewer b is analyzed, the main reviewer a is mainly distributed in the 9.00-11.00 time period in the morning during each review office document time, and the main reviewer b is mainly distributed in the 15.00-17.00 time period in the afternoon during each review office document time, so that the 9.00-11.00 in the morning is the peak time period of the main reviewer a for reviewing the office document, and the 15.00-17.00 in the afternoon is the peak time period of the main reviewer b for reviewing the office document, and the main reviewer a and the main reviewer b are pushed for office document in the 9.00-11.00 in the morning and the 15.00-17.00 in the afternoon, respectively;

[0131] The purpose of the multi-modal office document intelligent generation is to meet the individual needs of different reviewers, improve the review efficiency and effect of the office document, identify the main reviewer and the review habit, and then intelligently generate personalized multi-modal office documents that meet the needs of each reviewer. This intelligent generation method not only improves the generation efficiency of the document, but also ensures the accuracy and pertinence of the document content, provides strong support for the management of the office document of the enterprise, and improves the information acquisition efficiency of the core reviewer;

[0132] The working principle of the present application is:

[0133] By coincidence analysis on the meeting minutes and the historical office documents contained in the historical office document library, similar office documents in the historical office document library are extracted, enterprise personnel review records of the similar office documents are obtained, effective review time length is obtained, main reviewers are screened out according to the effective review time length, through analysis on the similar office documents in the historical office documents, review information of the reviewers can be obtained, which provides a personnel basis for subsequent analysis of paragraph main reviewers, is conducive to statistics of main contents required by enterprise personnel, is convenient for subsequent generation of multi-modal office documents, similarity analysis is performed on each paragraph in the similar office documents and the summary report of the main reviewers according to the summary report of the main reviewers, paragraphs with similarity exceeding a paragraph similarity threshold are marked as similar paragraphs, paragraph content similarity analysis and similar paragraph degree content ratio analysis are performed based on the similar paragraphs, paragraph main reviewers are confirmed, document layout habits of the paragraph main reviewers are obtained, parts consistent with the similar document content in the meeting minutes are generated into multi-modal office documents by an AI intelligent generation system, main review time periods of the main reviewers are obtained, and office documents are pushed, through content ratio analysis of each paragraph, paragraph main reviewers are screened out, which is conducive to personalized layout of the main reviewers of each paragraph, improves readability and pertinence of the office documents, in addition, the office documents are pushed according to main review time periods of the main reviewers, which not only meets individualized needs of different reviewers, but also improves generation efficiency and adaptability of the office documents.

[0134] Embodiment 3: please refer to Figure 4 As shown in the figure, the AI-based multi-modal office document intelligent generation system provided by the embodiment of the application comprises the following modules:

[0135] The coincidence analysis module: through coincidence analysis on the meeting minutes and the historical office documents contained in the historical office document library, similar office documents are extracted in the historical office document library;

[0136] The screening module: according to the enterprise personnel review records of the similar office documents, each enterprise personnel review record is analyzed, and main reviewers are screened out;

[0137] The ratio analysis module: the summary report of the main reviewers of the similar office documents is obtained, the summary report is similar to the content in the similar office documents, similar paragraph archives are obtained, the content ratio of each similar paragraph in the similar paragraph archives is analyzed, and the paragraph main reviewers of each paragraph of the similar office documents are obtained;

[0138] Intelligent generation module: by analyzing the archiving of similar paragraphs of the paragraph main reviewer in different historical meetings, the historical document layout of each paragraph main reviewer is obtained, the final draft of the meeting minutes is marked with the overlapping content of similar office documents, the marked content is obtained, the AI intelligent generation system is used for intelligent generation of the meeting minutes final draft in multiple modal office documents, and the main reviewer is pushed.

[0139] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. An AI-based multi-modal office document intelligent generation method, characterized in that: The method comprises the following steps: extracting similar office documents from the historical office document library by coincidence analysis on the meeting minutes and the historical office documents contained in the historical office document library; analyzing the review records of each same participating enterprise personnel to screen out main reviewers; obtaining a summary report of the similar office documents of the main reviewers, and performing similar pairing and integration of the summary report and the contents in the similar office documents to obtain similar paragraph archives, analyzing the content proportion of each similar paragraph in the similar paragraph archives to obtain the paragraph main reviewers of each paragraph of the similar office documents; obtaining the historical document layout of each paragraph main reviewer by analyzing the similar paragraph archives of the paragraph main reviewers in different historical meetings, and marking the contents of the final version of the meeting minutes and the coincident contents of the similar office documents to obtain marked contents, and using an AI intelligent generation system to generate a multi-modal office document intelligently based on the final version of the meeting minutes, and pushing the main reviewers.

2. The AI-based multi-modal office document intelligent generation method according to claim 1, characterized in that: The process of extracting similar office documents from the historical office document library is as follows: obtain the meeting minutes and the list of participating enterprise personnel to obtain a participating enterprise personnel table; preprocess the meeting minutes to obtain a final version of the meeting minutes; extract the main contents of the final version of the meeting minutes, input the main contents of the meeting minutes into the historical office document library, and obtain a document coincidence degree value through content coincidence comparison, if the document coincidence degree value is greater than or equal to a coincidence degree threshold value, mark the corresponding office document in the historical office document library as a similar office document.

3. The AI-based multi-modal office document intelligent generation method of claim 1, wherein: The process of analyzing the review records of each same participating enterprise personnel to screen out main reviewers is as follows: obtain the participating enterprise personnel table of the similar office documents, compare it with the participating enterprise personnel table of this time, mark the personnel participating in both meetings to obtain same participating enterprise personnel; statistically process the review duration of the same participating enterprise personnel when reviewing the similar office documents to obtain effective review duration, if the effective review duration of the same participating enterprise personnel is greater than or equal to a preset review duration threshold value, mark the corresponding same participating enterprise personnel as a suspected main reviewer; statistically process the number of highlight annotations and the number of effective annotations of the suspected main reviewer during the review process, add them together to obtain the number of interactions, process the number of interactions and the effective review duration by ratio to obtain an effective interaction value, if the effective interaction value is greater than or equal to an effective interaction threshold value, mark the suspected main reviewer as a main reviewer.

4. The AI-based multi-modal office document intelligent generation method of claim 1, wherein: The process of obtaining a summary report of the similar office documents of the main reviewers is as follows: After completing the review of the office documents, the main reviewers perform set integration of the document contents corresponding to the highlight annotations and the document contents corresponding to the effective annotations, and write a summary report of each main reviewer to summarize the core contents of the office documents required by the main reviewer, the findings in the review process, and the personal insights and suggestions.

5. The AI-based multi-modal office document intelligent generation method according to claim 1, characterized in that: The process of performing similar pairing and integration of the summary report and the contents in the similar office documents to obtain similar paragraph archives is as follows: Respectively, the paragraph content similarity of the similar office documents is analyzed to obtain a paragraph content similarity value, and the paragraph content similarity value is compared with a paragraph content similarity threshold value; If the paragraph content similarity value is greater than or equal to the paragraph content similarity threshold value, the corresponding paragraph is marked as a similar paragraph; The similar paragraphs are integrated to obtain a similar paragraph archive.

6. The AI-based multi-modal office document intelligent generation method according to claim 5, characterized in that: The content proportion of each similar paragraph in the similar paragraph archive is analyzed as follows: The total number of paragraphs in the summary report and the total number of words in the summary report are counted, and the total number of words in all similar paragraphs is compared with the total number of words in the summary report to obtain a similar paragraph word proportion value; The number of similar paragraphs is compared with the total number of paragraphs in the summary report to obtain a similar paragraph proportion value; The similar paragraph proportion value and the similar paragraph word proportion value are added to obtain a similar paragraph content degree proportion value, and the main reviewer with the maximum similar paragraph content degree proportion value is selected as the paragraph main reviewer.

7. The AI-based multi-modal office document intelligent generation method according to claim 1, characterized in that: The historical document layout of each paragraph main reviewer is obtained, and the marked content is obtained by marking the overlapping content between the final draft of the meeting minutes and the similar office documents as follows: The historical document layout of each paragraph main reviewer is obtained by analyzing the similar paragraph archive in the historical meeting, and the document layout habit of each paragraph main reviewer is extracted; The main content existing in the similar office documents and the final draft of the meeting minutes is marked to obtain the marked content.

8. The AI-based multi-modal office document intelligent generation method according to claim 1, characterized in that: The AI intelligent generation system generates a multi-modal office document based on the final draft of the meeting minutes as follows: The marked content is marked in the paragraph of the similar office document, and the marked content, the corresponding paragraph main reviewer, and the document layout habit of the paragraph main reviewer of each paragraph of the similar document are input into the AI intelligent generation system. The AI intelligent generation system determines the paragraph main reviewer of the marked content according to the marked content; The AI intelligent generation system arranges the final draft of the meeting minutes according to the document layout habit of each paragraph main reviewer, and generates an office document based on the final draft of the meeting minutes by using the AI intelligent generation system.

9. The AI-based multi-modal office document intelligent generation method according to claim 1, characterized in that: The main reviewer is pushed as follows: The review time period of each main reviewer for reviewing the office document is marked on the time axis to obtain a review time period set; The review time period set is counted to obtain the occurrence frequency of each review time period, and the review time period is sorted in descending order of occurrence frequency, and the review time period with the highest frequency is marked as the peak time period of the main reviewer; The peak time period of each main reviewer for reviewing the office document is extracted, and the intelligently generated office document is pushed in the peak time period of the main reviewer; If the main reviewer has multiple peak time periods, the peak time periods are pushed according to the order of the peak time periods; If the peak time periods of multiple main reviewers are in the same time period, the multiple main reviewers are pushed simultaneously.

10. An AI-based multi-modal office document intelligent generation system, characterized in that: The following modules are included: The coincidence analysis module extracts similar office documents from the historical office document library by performing coincidence analysis on the meeting minutes and the historical office documents contained in the historical office document library; The screening module analyzes each same-enterprise personnel review record and screens out main review personnel according to the enterprise personnel review records of the similar office documents; The proportion analysis module obtains summary reports of the similar office documents of the main review personnel, and performs similar pairing and integration on the summary reports and the contents in the similar office documents to obtain similar paragraph archives, analyzes the content proportion of each similar paragraph in the similar paragraph archives, and obtains the paragraph main review personnel of each paragraph of the similar office documents; The intelligent generation module analyzes the historical document layout of each paragraph main review personnel in different historical meetings, marks the coincidence contents of the meeting minutes final draft and the similar office documents, obtains marked contents, performs multi-modal office document intelligent generation on the meeting minutes final draft by using an AI intelligent generation system, and pushes the main review personnel.

Images