Method and device for generating summary

Abstract

The invention provides a summary generation method and device.The method comprises the steps that a first text and a first record text corresponding to the first text are obtained, and the first record text comprises concerned content in the first text; a second record text corresponding to the second text is obtained, the second record text is a summary text of the second text, and the first text and the second text are texts with the same theme; determining associated content of the first record text from the first text; and based on the associated content and the second record text, performing summary generation on the first text to obtain a summary text of the first text. According to the method and the device, the associated content of the first record text concerned by the user is determined, and the second record text related to the first text is combined, so that accurate matching of summary generation and user preferences is realized, the pertinence and individuation degree of information are improved, and the continuity, accuracy and reliability of the summary text content are ensured.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a minutes generation method and device. Background Art

[0002] Minutes generation refers to the extraction of content from long texts, thereby refining information that can represent the core content of the text. Minutes can help relevant personnel grasp the content of the text more directly and effectively.

[0003] Conventional minutes generation solutions typically use a general, large model to identify key information from the target text and generate a summary of the target text based on this key information. However, this approach is unable to generate meeting minutes with different content tailored to the needs of different personnel. In other words, it cannot meet the needs of different personnel for the same target text. Summary of the Invention

[0004] The present invention provides a method and device for generating minutes, which are used to solve the defects in the prior art.

[0005] The present invention provides a method for generating minutes, comprising the following steps: Acquire a first text and a first record text corresponding to the first text, where the first record text includes the focus content in the first text; Obtaining a second record text corresponding to the second text, where the second record text is a summary text of the second text, and the first text and the second text are texts on the same subject; determining, from the first text, associated content of the first record text; Based on the associated content and the second record text, a minutes text is generated for the first text to obtain a minutes text of the first text.

[0006] According to a minutes generation method provided by the present invention, generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text includes: Constructing a generation prompt text, wherein the generation prompt text is used to provide key information or instructions for guiding the generation of minutes; splicing the generated prompt text, the associated content, the second record text, and the first text to obtain a spliced text; Based on the minutes generation model, the concatenated text is applied to generate minutes for the first text to obtain a minutes text of the first text.

[0007] According to a method for generating minutes provided by the present invention, the minutes text of the first text includes a summary text and a key point text; Generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text includes: Based on the associated content, extracting a summary of the first text to obtain the summary text; Based on the second record text and the summary text, key points are generated for the first text to obtain the key point text.

[0008] According to a method for generating minutes provided by the present invention, generating key points of the first text based on the second record text and the summary text to obtain the key point text includes: Extracting core information from the summary text and extracting key information from the second record text; Based on the association between the core information and the key information, and the difference between the core information and the key information, key points are generated for the first text to obtain the key point text.

[0009] According to a method for generating minutes provided by the present invention, generating key points of the first text based on the association between the core information and the key information, and the difference between the core information and the key information, to obtain the key point text, includes: Determining a logical chain of the key text based on the association between the core information and the key information; enhancing the logic chain based on the differences between the core information and the key information; Based on the enhanced logical chain, key points are generated for the first text to obtain the key point text.

[0010] According to a method for generating minutes provided by the present invention, the second record text is determined based on the following steps: Extracting keywords from the document data associated with the first text and / or the recorded text; The second record text is obtained by searching the historical records based on the keyword.

[0011] According to a method for generating minutes provided by the present invention, the step of searching the historical minutes based on the keyword to obtain the second record text includes: Combining the keyword, synonyms of the keyword, and related words of the keyword to obtain a query phrase; Based on the keyword and the query phrase, the historical records are searched to obtain the second record text.

[0012] According to a minutes generation method provided by the present invention, the second record text is stored in a minutes database; generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text, and then further comprising: The minutes database is updated based on the minutes text of the first text.

[0013] The present invention also provides a method for generating minutes, comprising the following modules: A first acquiring unit is configured to acquire a first text and a first record text corresponding to the first text, wherein the first record text includes a focus content in the first text; A second acquiring unit is configured to acquire a second record text corresponding to the second text, where the second record text is a summary text of the second text, and the first text and the second text are texts on the same subject; a determining unit, configured to determine associated content of the record text from the first text; A generating unit is configured to generate a minutes text of the first text based on the associated content and the second record text to obtain a minutes text of the first text.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the minutes generation methods described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described minutes generation methods.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned minutes generation methods.

[0017] The minutes generation method and device provided by the present invention, by determining the associated content of a first recorded text of user interest and combining it with a second recorded text related to the first text, not only accurately matches the minutes generation with the user's preferences, improving the pertinence and personalization of the information, but also ensures the continuity and integrity of the minutes content. Furthermore, the second recorded text and the first recorded text can verify each other, correcting any omissions or errors, thereby ensuring the accuracy and completeness of the generated minutes text. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 It is a flow chart of the minutes generation method provided by the present invention.

[0020] Figure 2 It is a flow chart of another method for generating minutes provided by the present invention.

[0021] Figure 3 It is a structural diagram of the minutes generating device provided by the present invention.

[0022] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] The minutes generation method provided by the present invention is suitable for application scenarios of generating text minutes. By using the minutes generation method provided by the present invention, text that is consistent with the user's focus can be generated, thereby meeting the text minutes needs of different users.

[0025] The above-mentioned application scenarios for generating text minutes specifically refer to scenarios where minute content needs to be generated, including but not limited to specific application scenarios such as generating meeting minutes, generating document summaries, and extracting news highlights. To facilitate understanding of the technical solution of the present invention, the following embodiments are all described using the application to the scenario of generating meeting minutes as an example.

[0026] Currently, the generation of meeting minutes primarily relies on a common approach for transcribing content from meeting speech recognition. This approach first segments the text transcribed from speech recognition into smaller paragraphs to accommodate the length constraints of inputs processed by large, general-purpose models. Subsequently, each paragraph is stripped of colloquial expressions, irrelevant details, or content that deviates from the topic, and the core content is extracted to form a concise paragraph corresponding to each paragraph. The core information from the meeting discussion is then identified and listed within each concise paragraph, and the core information from each concise paragraph is summarized to generate a complete meeting minutes text.

[0027] However, with the increasing maturity and widespread application of speech recognition technology and large-scale models, although meeting records can be generated in real time and meeting minutes text containing key information can be quickly produced based on this, this solution still faces many technical problems. In particular, the participants in the meeting have diverse roles and their respective focus on the meeting varies significantly. Therefore, the current general meeting minutes generation solutions often cannot meet the personalized and differentiated needs of participants. In particular, with the popularization of personal office equipment such as voice recorders and office notebooks, participants have an increasingly urgent need to generate meeting minutes that better suit their needs, which further highlights the shortcomings of the above solutions in terms of flexibility and customization.

[0028] Based on the above-mentioned technical status, the present invention provides a method for generating minutes. In the subsequent embodiments of the present invention, the specific processing content of the technical solution of the present invention is introduced by taking the generation of meeting minutes as an example. When the technical solution of the present invention is applied to other scenarios, its specific execution process can refer to the introduction of the various embodiments of the present invention. It should be noted that the technical solution of the present invention is not only applicable to generating minutes from text, thereby obtaining minutes in text form, but also applicable to generating minutes from voice, obtaining minutes in text or voice form, or generating minutes from text to obtain minutes in voice form. When generating minutes for non-text data, or generating non-text minutes, it can be achieved by converting the non-text data into text data, or converting the generated text minutes into non-text form. The main processing of the minutes generation can still refer to the introduction of the embodiments of the present invention.

[0029] in, Figure 1 It is a flow chart of the method for generating minutes provided by the present invention, such as Figure 1 As shown, the method includes step 110 , step 120 and step 130 .

[0030] Step 110: Obtain a first text and a first record text corresponding to the first text, where the first record text includes the focus content in the first text.

[0031] Step 120: Obtain a second record text corresponding to the second text. The second record text is a summary text of the second text. The first text and the second text are texts on the same subject.

[0032] Here, the first text refers to the text for which minutes are to be generated. This first text can be a text of any content and in any language obtained through any means. Specifically, the first text can be directly obtained text, such as academic documents, news releases, books, etc., or text obtained by speech recognition, such as the recognition result text obtained by performing speech recognition on a conference recording, or the recognition result text obtained by recognizing the speech of a speaker. In theory, any form of data content can be converted into text form, thereby serving as the above-mentioned first text, and the minutes of this first text can be generated through subsequent processing.

[0033] In addition, the first record text corresponding to the first text includes the focus content in the first text. This focus content can be understood as information in the first text that the user considers important and requires special recording or emphasis. The first record text can reflect the focus content in the first text through methods such as abstracts, summaries, annotations of key information, or extraction of core points. Specifically, the first record text may include key information (such as core points, data, conclusions, etc. extracted from the first text), summary content (such as a brief summary of a part or the entire first text), annotations or comments (such as explanations, explanations, or supplements to certain content in the first text), etc.

[0034] That is to say, the first record text can represent the user's interest or focus on the first text, and at the same time represent the user's demand for the content of the generated text minutes. It is used to provide a reference for generating the minutes text of the above-mentioned first text, so as to generate a minutes text that meets the user's focus or contains the user's content of interest.

[0035] The first recorded text can be input by the user, or the user's handwritten record can be recognized by text recognition (such as using OCR technology to recognize handwritten records and convert them into recorded text), or it can be pre-set before executing the text minutes generation. The specific form of the recorded text can be a fixed sentence structure, a keyword or phrase, a short text sentence or text paragraph, or even a combination of multiple subject texts. For example, taking a meeting scenario as an example, the recorded text can be a combination of a meeting topic text and a discussion points text. The meeting topic text is used to represent the meeting topic that the user is concerned about, and the discussion points text is used to represent the discussion points that the user is concerned about.

[0036] For example, in a meeting minutes generation scenario, speech recognition is performed on the meeting recording to obtain text corresponding to the meeting recording, which serves as the first text described above. Simultaneously, a first recorded text input by the user representing the meeting content of interest or concern is obtained. This first recorded text can be a phrase or short sentence summarized by the user based on the meeting content, a simple meeting record taken by the user at the venue, or keywords, phrases, or search conditions determined by the user based on the desired meeting minutes content.

[0037] When a processing device processes text, it is actually processing text features. Therefore, the aforementioned "obtaining the first text and the first recorded text" can mean obtaining the original text of the first text and the first recorded text, then performing feature extraction on the obtained first text and the first recorded text to obtain the features of the first text and the first recorded text for subsequent text-memorandum generation processing; or it can mean directly obtaining the features of the first text and the first recorded text for subsequent text-memorandum generation processing.

[0038] In addition, the second text and the first text are texts on the same subject. For a meeting scenario, the same subject here can be the same project, the same topic, etc. The second text may contain background information, discussion progress, unresolved issues, etc. related to the first text. The second record text is the minutes text of the second text, and the second record text also contains background information, discussion progress, unresolved issues, etc. related to the first text. For example, the first text is the speech transcription text corresponding to the third meeting of Project A, and the second text may include the speech transcription text corresponding to the first meeting of Project A and the speech transcription text corresponding to the second meeting of Project A. The second record text includes the minutes text of the first meeting of Project A and the minutes text of the second meeting of Project A, that is, the second record text can be regarded as a historical minutes text under Project A. The minutes text of the first meeting and the minutes text of the second meeting record background information, discussion progress, unresolved issues, etc. related to Project A.

[0039] For example, the second record text can be retrieved based on the subject (e.g., project name) corresponding to the first text by searching a minutes database, where minutes on different topics are stored. The second record text can also be determined based on the semantic relevance between the first text and various minutes texts. For example, minutes with a relevance greater than a threshold can be used as the second record text.

[0040] Step 130: Determine the associated content of the first record text from the first text.

[0041] Specifically, the associated content of the first record text refers to the text content in the first text that is related to the first record text, such as text content whose similarity with the first record text is greater than the set similarity threshold, or text content that is semantically similar or related to the first record text, all of which can be used as the associated content of the first record text.

[0042] Exemplarily, an embodiment of the present invention compares the first record text with each text fragment of the first text in sequence to determine the text similarity or semantic similarity between the first record text and each text fragment of the first text, thereby determining the correlation between the first record text and the first text fragment, and identifying associated content related to the first record text from the first text.

[0043] It can be understood that since the associated content of the first record text contains text information related to the first record text, if the associated content of the first record text can be given priority consideration when generating a text minutes of the first text, the final generated minutes text can contain more associated information of the first record text, thereby making the generated minutes text match the first record text.

[0044] Step 140: Generate a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text.

[0045] Specifically, because the second transcript may contain summaries of previous discussions, decisions, and background information related to the first transcript, combining the second transcript during the minutes generation process ensures that not only the direct content of the first transcript is captured, but also that this direct content is connected to previous discussions, decisions, and background information. This not only helps to build a complete information chain, but also allows users to better understand the context and development of the current discussion. By ensuring the continuity of the minutes, the readability and comprehensibility of the information are enhanced, allowing users to more easily track the development of the discussion and better participate in subsequent discussions or decisions.

[0046] For example, a company is developing a new product and has held a series of meetings. During these meetings, the team discussed various aspects, including product design, feature planning, and market positioning. After each meeting, corresponding meeting minutes are generated. Minutes A from the first meeting detail the product's core features and target user groups. Minutes B from the second meeting outline adjustments and optimizations to certain designs in Minute A, as well as the technical resources required to implement them.

[0047] Let's assume the first text is a transcript of the third meeting, which began discussing market positioning and marketing strategy. By citing Minutes A and B in Minutes C, the resulting minutes from the third meeting, this provides a coherent link between the two. This allows users to clearly see the logical chain of the entire decision-making process, from product design to market positioning, ensuring information continuity and avoiding information gaps. Minutes A and B are considered the second transcript.

[0048] In addition, related content related to the recorded text is determined from the first text. These related contents not only reflect the user's direct interests, but also imply the user's preferences and concerns. Generating minutes based on these related contents means that the generated minutes will closely revolve around the core of the user's interests and provide highly customized minutes text. In other words, the minutes text generated based on related content can not only meet the user's needs for obtaining specific information, but also improve the user's reading experience and satisfaction by matching the user's preferences. As a result, the user can quickly obtain the content they really care about from the minutes, thereby saving time and improving efficiency.

[0049] It should be noted that the second record text and the first record text are complementary in content. The second record text focuses on recording the overall process and key decisions of the meeting, while the first record text focuses on recording the user's insights and concerns. Combining the second record text and the record text can form a more comprehensive minutes text of the first text. In addition, the second record text and the first record text can verify each other in terms of information. By comparing the two, possible omissions or errors in information can be discovered and corrected, thereby improving the accuracy and reliability of the minutes text of the first text. For example, there may be ambiguities and ambiguities in the second record text, while the first record text is used to represent the user's focus in the meeting and can often capture subtle differences and omissions. Based on these concerns, the ambiguities and ambiguities in the second record text can be further explored and clarified, thereby ensuring the accuracy and completeness of the minutes text.

[0050] It can be understood that the above steps 130 and 140 can also be implemented through a large model, that is, the large model can determine the associated content of the first record text from the first text, and generate a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text.

[0051] The minutes generation method provided by the embodiments of the present invention, by determining the relevant content of the first recorded text of the user's interest and combining it with the second recorded text, not only accurately matches the minutes generation with the user's preferences, improving the pertinence and personalization of the information, but also ensures the continuity and integrity of the minutes content. Furthermore, the second recorded text and the first recorded text can verify each other, correcting any omissions or errors, thereby ensuring the accuracy and completeness of the generated minutes text.

[0052] Exemplarily, key information of concern to the user can be determined based on the associated content, and relevant content matching the key information can be extracted from the second record text, and the associated content and the related content can be integrated to form the final minutes text of the first text.

[0053] Based on the above embodiment, the minutes text of the first text includes a summary text and a key point text; Based on the associated content and the second record text, a minutes text of the first text is generated to obtain a minutes text of the first text, including: Based on the related content, extract a summary of the first text to obtain a summary text; Based on the second record text and the summary text, key points are generated for the first text to obtain a key text.

[0054] Specifically, the minutes text of the first text is usually composed of two major components: summary text and key points text. As a refined overview of the first text, the core purpose of the summary text is to accurately capture and convey the core information and core issues of the text, while eliminating redundant information and details to ensure that users can quickly grasp the main idea of the text. Given that the associated content reflects the specific focus of the user, in order to achieve personalized customization of the minutes text, the embodiment of the present invention implements a summary extraction strategy for the first text based on the associated content to ensure that the obtained summary text can closely fit the user's core concerns and interest tendencies, thereby effectively improving the user's efficiency in reading and understanding the minutes, and enabling users to quickly focus on the information points they care about.

[0055] For example, by deeply analyzing the related content, key information elements can be extracted. These elements may include phrases, sentences, or even paragraphs, and are usually closely related to the theme, topic, or core discussion point of the user's interest in the first text. Subsequently, these key information elements are integrated, organized, and reconstructed to generate a summary text.

[0056] In addition, the key text refers to a detailed elaboration and summary of the key information and decision-making results in the first text, which is used to further deepen and concretize the summary text in order to provide users with a deeper understanding and action guide.

[0057] Therefore, when generating key points, it's important to integrate the core information extracted from the summary. Furthermore, to ensure the coherence and completeness of the generated minutes, in addition to the summary, a secondary record, representing the context of the previous discussions and the basis for the decision, is also necessary. By integrating these two components, the generated key points not only accurately reflect the core issues and decision outcomes of the user's concern, but also ensure the coherence of the minutes, providing users with a clear and complete record of the discussions and decisions.

[0058] For example, the key points text can be obtained through the following steps: First, an in-depth analysis of the second transcript is conducted to extract the previous discussion content, decision-making basis, and background information directly related to the key issues and decision results in the first transcript. Then, these key elements are combined with the core information and main discussion points extracted from the summary text to form the key points text.

[0059] It should be noted that when generating key points, consistency must be maintained between the key points and the summary text, allowing users to smoothly transition from the summary overview to the detailed key points, achieving seamless integration and in-depth understanding of the information. To this end, a consistency verification step can be implemented: the key points and each information unit in the summary text (the information unit here can be a text fragment, a sentence, or a word segment) are compared one by one. Text similarity algorithms or semantic analysis techniques are used to assess whether the key information in the summary text is accurately and fully reflected in the key points text. If the text similarity or semantic similarity of a certain information unit in the key points text exceeds a preset threshold, it can be considered that the information is effectively reflected in the key points text. Conversely, if the similarity is below the threshold, it indicates that the information unit in the summary text is not fully elaborated in the key points text, that is, there is information omission. For such omitted information, appropriate supplementation and explanation should be made in the key points text to ensure the completeness and coherence of the information.

[0060] Based on any of the above embodiments, key points are generated for the first text based on the second record text and the summary text to obtain a key point text, including: Extracting the core information of the summary text and extracting the key information of the second record text; Based on the association between the core information and the key information, and the difference between the core information and the key information, key points are generated for the first text to obtain a key point text.

[0061] Specifically, the second record text is directly related to the first text, recording discussions, decisions, or background information related to the first text. The content in the summary text is directly derived from the first text and is a simplification and summary of the first text's content. Therefore, there must be a certain degree of connection between the second record text and the summary text. This connection may be reflected in the themes, topics, or key points they share. Because of this connection between the second record text and the summary text, the core topics and main content of the key text can be determined based on their common points or intersections.

[0062] Furthermore, the second transcript may contain information not mentioned in the summary. This information may supplement, deepen, or elaborate on the content of the first transcript from a different perspective. In other words, the second transcript and the summary are not only related but may also differ. These differences can reveal important information omitted from the summary but emphasized in the second transcript. This information can then be used to identify content that requires additional emphasis or supplementation in the key points.

[0063] Furthermore, considering that there may be redundant, secondary or repetitive information in the summary text and the second record text, this information is not important for the generation of the key points text and may even interfere with the understanding of the core content. Therefore, before generating key points for the first text, it is necessary to extract the core information of the summary text and the key information of the second record text. Core information refers to the content in the summary text that directly reflects the main theme and key points of the first text, while key information refers to the content in the second record text that is closely related to the core issues and is crucial to understanding the key points. Since the core information eliminates the redundant and secondary information in the summary text, and the key information eliminates the non-key content in the second record text, based on the relationship between the core information and the key information (such as issues of common concern, mutually supporting views, etc.), and the differences between the core information and the key information (such as different focuses, supplementary information, etc.), a comprehensive, accurate and concise key points text can be formed.

[0064] Alternatively, natural language processing techniques (such as text summarization algorithms and keyword extraction) can be used to extract the core information of the summary text, while information extraction or text analysis techniques can be used to extract key information from the second transcript. Based on this information, semantic analysis can be used to determine the relationships between them (such as causal relationships, parallel relationships, etc.), and comparison or induction can be used to identify their differences (such as information emphasis and expression methods). Based on this foundation, combined with the overall content and background information of the first transcript, the logical framework and key points of the key text can be more accurately constructed, resulting in a concise and complete key text.

[0065] Based on any of the above embodiments, key points are generated for the first text based on the association between the core information and the key information, as well as the difference between the core information and the key information, to obtain a key point text, including: Determine the logical chain of key text based on the relationship between core information and key information; Strengthen the logical chain based on the difference between core information and key information; Based on the enhanced logical chain, key points are generated for the first text to obtain a key text.

[0066] Considering that the connection between core information and key information may be reflected in the themes, topics, or key points they share, we can extract the main points or arguments of the key text based on this connection. These main points or arguments are then sorted and organized to form the logical chain of the key text. This logical chain is used to represent the core content and structural framework of the key text, thereby ensuring that the final key text is logically coherent and consistent.

[0067] Furthermore, considering that the summary text may lack certain key details, background information, etc., thus causing the obtained logical chain to be incomplete, this missing information in the summary text can be enhanced and supplemented by the supplementary information in the second record text. Therefore, based on the differences between the core information and the key information, the embodiment of the present invention extracts relevant information from the second record text to supplement the missing parts in the logical chain, thereby enhancing the completeness and depth of the logical chain.

[0068] Since the enhanced logical chain is more comprehensive in content and clearer in structure, when key points are generated for the first text based on the enhanced logical chain, the key text obtained can more accurately reflect the main theme and key points of the first text, while avoiding omissions and redundancies of information, thereby improving the practicality and readability of the key text.

[0069] For example, we can first extract the main arguments or thematic viewpoints based on the relationship between core and key information, and then establish the logical chain of the key text. Then, by analyzing the differences between the core and key information, we can identify and obtain necessary supplementary information, and use this information to enhance the logical chain. Finally, based on the enhanced logical chain, we generate key points. By refining and summarizing the core elements of the first text and combining them with the enhanced logical chain, we ultimately produce a concise, clear, and well-structured key text.

[0070] Based on any of the above embodiments, the second record text is determined based on the following steps: Extracting keywords from the document data and / or recorded text associated with the first text; The historical records are searched based on the keywords to obtain the second record text.

[0071] Specifically, the material data associated with the first text refers to various data types related to the content or context of the first text. For example, the material data may include image data, document data (such as files in PDF, Word, Excel, etc.), audio data, video data, and other forms of unstructured or structured data. Keywords in the material data are used to represent the core information, theme, or key elements in the material data.

[0072] Furthermore, by searching historical records based on keywords in the data, keyword matching or semantic analysis techniques can be used to find records containing these keywords in the historical records database, thereby obtaining secondary records related to the first text. These secondary records may contain similar topics, discussion content, or background information as the first text, helping to more fully understand the background and context of the first text.

[0073] In addition, the keywords in the record text are also used to represent the core content or key information of the record text. Since the record text is related to the first text, searching the historical records based on the keywords in the record text can also obtain a second record text related to the first text.

[0074] For example, for image data, image recognition and analysis can be performed using a large image model to extract valid information and convert it into structured data. Named entity recognition can then be performed on the structured data to identify keywords within the image data. For document data, document analysis techniques can be used to parse and extract information, extract valid information, and convert it into structured data. Named entity recognition can then be performed on the structured data to identify keywords within the document data.

[0075] When searching historical records based on keywords, fuzzy search can be used, i.e., based on keyword similarity or semantic association. Because fuzzy search can handle incomplete matches between keywords, it can expand the search scope and improve the recall rate of search results. Specifically, fuzzy search can be implemented by considering spelling variants, synonyms, near-synonyms, or related concepts of keywords. This allows fuzzy search to find related or similar secondary records even if the input keyword does not exactly match the actual wording in the record.

[0076] For example, when processing document data, if the keyword is "artificial intelligence," then fuzzy search may find second record text containing related concepts such as "AI," "machine learning," and "deep learning," because these words are semantically related to "artificial intelligence." Similarly, when processing image data, if the keyword obtained through image recognition is "building," then fuzzy search may find second record text containing related words such as "architecture," "building," and "edifice," because these words have similar meanings when describing buildings.

[0077] It can be seen that the fuzzy search method can make keyword-based search more flexible and accurate, and help users find the second record text related to the first text more comprehensively.

[0078] Based on any of the above embodiments, searching the historical records based on keywords to obtain the second record text includes: Combine keywords, their synonyms, and related words to obtain a query phrase; Based on the keywords and the query phrase, the historical records are searched to obtain the second record text.

[0079] Specifically, searching historical minutes based solely on keywords may miss historical minutes that differ in their wording but are substantively related because the keywords are overly narrow or specific. This omission results in an incomplete secondary record that fails to accurately reflect the relevant information in the historical minutes. For example, searching solely based on the keyword "meeting agenda" may miss minutes that mention "discussion topics" or "meeting content," even though these minutes are closely related to the "meeting agenda."

[0080] Based on this, the embodiment of the present invention expands and modifies the keywords to obtain a richer query phrase, which can represent the keywords and their related multiple expressions and associated content, thereby avoiding the retrieval omission problem caused by overly narrow keyword expressions.

[0081] Specifically, when expanding and revising keywords, embodiments of the present invention fully consider synonyms and related words of keywords. The introduction of synonyms can include words that have similar meanings but different expressions as keywords, thereby increasing search flexibility. For example, the synonyms "discussion topic" and "meeting content" for "meeting agenda" can increase the probability of retrieving relevant minutes.

[0082] Furthermore, the introduction of related terms can introduce vocabulary that is thematically or contextually related to the keyword, further expanding the scope of the search. For example, the related terms for "meeting agenda" such as "meeting minutes," "meeting minutes," and "decision-making matters" can guide the search system to focus on other documents and content closely related to the meeting agenda.

[0083] Searching historical records based on keywords can retrieve secondary records directly related to the keywords; searching historical records based on query phrases can retrieve secondary records related to the keywords, their synonyms, and related words. This shows that searching historical records with keywords and query phrases can fully capture the content of the records in various expressions and contexts related to the keywords, and thus fully retrieve secondary records.

[0084] For example, keywords, synonyms of keywords, and related words of keywords can be combined based on natural language processing technology or manual judgment to obtain query phrases. For example, for the keyword "meeting topic", its synonyms may include "discussion topic", "meeting content", etc., and related words may include "meeting record", "meeting minutes", "decision-making matters", etc. These words are combined to obtain query phrases such as "meeting topic / discussion topic / meeting content / meeting record / meeting minutes / decision-making matters" (the combination of each word in the query phrase can be appropriately adjusted as needed during actual retrieval), which is used to search in historical minutes, so as to obtain the minutes content related to "meeting topic" more comprehensively.

[0085] Based on any of the above embodiments, generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text includes: Construct and generate prompt text, which is used to guide the key information or instructions of minutes generation; Splicing the generated prompt text, the associated content, the second record text, and the first text to obtain a spliced text; Based on the minutes generation model, the concatenated text is applied to generate minutes for the first text to obtain a minutes text of the first text.

[0086] Specifically, generation prompt text refers to the text prompts constructed during the minutes generation process to guide the minutes generation model to better understand and process input information (such as related content, second record text, and first text). Generation prompt text usually contains key information or instructions for guiding the model minutes generation, aiming to help the model clarify task objectives, understand context, or highlight processing priorities.

[0087] For example, suppose you want to generate minutes for a meeting on "The Application of AI Technology in Education." The prompt text might be, "Please generate a minutes text for the meeting on 'The Application of AI Technology in Education' based on the following related content and the second record text, focusing on the types of AI technologies discussed in the meeting, their application scenarios, and future development trends." After constructing the generated prompt text, the generated prompt text, associated content, the second record text, and the first record text are concatenated together to integrate all relevant information. The resulting concatenated text is used as input to the minutes generation model, which automatically generates a minutes text for the first record based on the information in the input text. The minutes generation model can be a natural language processing model based on deep learning technologies, such as Transformer and BERT.

[0088] In addition, in order to meet the input length requirements of the minutes generation model, before splicing the prompt text, associated content, second record text and first text, the first text can be divided into multiple smaller text blocks, and the corresponding associated content and second record text can be determined for each text block. The minutes corresponding to each text block can be generated with reference to the method of the above embodiment, and finally the minutes corresponding to each text block can be combined to obtain the final minutes text of the first text.

[0089] After generating the first-pass minutes, if the minutes do not fully meet the user's actual needs, the following measures can be taken to improve them: first, adjust the content of the generated prompt text to ensure that it more accurately guides the minutes generation process; second, fine-tune the parameters of the minutes generation model to improve the model's generation performance. These measures can be continued until the final first-pass minutes fully meet the user's expectations and needs.

[0090] When the minutes generation model uses the concatenated text to generate minutes for the first text, the minutes generation method can refer to any of the above-mentioned embodiments to generate minutes for the first text. For example, the first text can be abstracted based on the associated content to obtain a summary text; and the first text can be key-point generated based on the second record text and the summary text to obtain a key-point text.

[0091] Based on any of the above embodiments, the second record text is stored in the minutes database; Based on the associated content and the second record text, a minutes is generated for the first text to obtain a minutes text of the first text, which further includes: Based on the minutes text of the first text, the minutes database is updated.

[0092] Specifically, the minutes database can be a relational database. Relational databases provide structured storage, allowing minutes to be organized according to a predetermined format (such as meeting date, subject, and participants). Furthermore, by storing minutes in a structured manner, relational databases can quickly filter and sort minutes based on keywords (such as meeting subject and participant names) and time ranges.

[0093] After the minutes text of the first text is generated, the minutes text is stored as a new minutes text in the minutes database to maintain the timeliness and integrity of the minutes database. In addition, in order to improve the response speed and throughput of queries in the minutes database, frequently queried minutes texts or parts of them can be cached in the memory to reduce the number and time of database access.

[0094] Based on any of the above embodiments, Figure 2 It is a flow chart of another method for generating minutes provided by the present invention, such as Figure 2 As shown, taking the method applied to a conference scenario as an example, the method includes: Acquire conference audio data and handwritten notes of users at the conference site, perform speech recognition on the audio data to obtain a first text; perform OCR recognition on the handwritten notes to obtain a first recorded text corresponding to the first text, perform keyword extraction on the first recorded text, and obtain keywords of the first recorded text. At the same time, obtain image data and document data related to the meeting, perform image recognition and analysis on the image data through a large image model, extract valid information and convert it into structured data, and then perform named entity recognition on the structured data to obtain keywords in the image data. Perform text parsing and information extraction on the document data through document analysis technology, extract valid information and convert it into structured data, and similarly perform named entity recognition on the structured data to obtain keywords in the document data. The keywords of the recorded text, the keywords of the image data, and the keywords of the document data constitute the final keywords.

[0095] Based on natural language processing technology, keywords, synonyms of keywords and related words of keywords are combined to obtain a query phrase. Based on the keywords and the query phrase, a fuzzy search is performed in the minutes database to obtain a second record text.

[0096] The keyword, the second record text and the first text are input into the minutes generation model, the minutes generation model generates minutes for the first text to obtain the minutes text of the first text, and the minutes text of the first text is stored in the minutes database.

[0097] The minutes generating device provided by the present invention is described below. The minutes generating device described below and the minutes generating method described above can be referenced to each other.

[0098] Based on any of the above embodiments, Figure 3 It is a structural diagram of the minutes generating device provided by the present invention, such as Figure 3 As shown, the device includes: A first acquiring unit 310 is configured to acquire a first text and a first record text corresponding to the first text, where the first record text includes a focus content in the first text; A second acquiring unit 320 is configured to acquire a second record text corresponding to the second text, where the second record text is a summary text of the second text, and the first text and the second text are texts on the same subject; The determining unit 330 is configured to determine the associated content of the first record text from the first text; The generating unit 340 is configured to generate a minutes text of the first text based on the associated content and the second record text, to obtain a minutes text of the first text.

[0099] Based on any of the above embodiments, generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text includes: Construct and generate prompt text, which is used to guide the key information or instructions of minutes generation; Splicing the generated prompt text, the associated content, the second record text, and the first text to obtain a spliced text; Based on the minutes generation model, the concatenated text is applied to generate minutes for the first text to obtain a minutes text of the first text.

[0100] Based on any of the above embodiments, the minutes text of the first text includes a summary text and a key point text; Based on the associated content and the second record text, a minutes text of the first text is generated to obtain a minutes text of the first text, including: Based on the related content, extract a summary of the first text to obtain a summary text; Based on the second record text and the summary text, key points are generated for the first text to obtain a key text.

[0101] Based on any of the above embodiments, key points are generated for the first text based on the second record text and the summary text to obtain a key point text, including: Extracting the core information of the summary text and extracting the key information of the second record text; Based on the association between the core information and the key information, and the difference between the core information and the key information, key points are generated for the first text to obtain a key point text.

[0102] Based on any of the above embodiments, key points are generated for the first text based on the association between the core information and the key information, as well as the difference between the core information and the key information, to obtain a key point text, including: Determine the logical chain of key text based on the relationship between core information and key information; Strengthen the logical chain based on the difference between core information and key information; Based on the enhanced logical chain, key points are generated for the first text to obtain a key text.

[0103] Based on any of the above embodiments, the second record text is determined based on the following steps: Extracting keywords from the document data and / or recorded text associated with the first text; The historical records are searched based on the keywords to obtain the second record text.

[0104] Based on any of the above embodiments, searching the historical records based on keywords to obtain the second record text includes: Combine keywords, their synonyms, and related words to obtain a query phrase; Based on the keywords and the query phrase, the historical records are searched to obtain the second record text.

[0105] Based on any of the above embodiments, the second record text is stored in the minutes database; Based on the associated content and the second record text, a minutes is generated for the first text to obtain a minutes text corresponding to the first text, and then the following is further included: Based on the minutes text of the first text, the minutes database is updated.

[0106] Figure 4 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call logic instructions in the memory 430 to execute a minutes generation method, which includes: obtaining a first text and a first record text corresponding to the first text, the first record text including the focus content in the first text; obtaining a second record text corresponding to a second text, the second record text being a minutes text of the second text, the first text and the second text being texts on the same subject; determining associated content of the first record text from the first text; and generating a minutes text for the first text based on the associated content and the second record text to obtain a minutes text of the first text.

[0107] Furthermore, the logic instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0108] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the minutes generation method provided by the above methods, which includes: obtaining a first text, and a first record text corresponding to the first text, the first record text including the focus content in the first text; obtaining a second record text corresponding to the second text, the second record text is the minutes text of the second text, and the first text and the second text are texts on the same subject; determining the associated content of the first record text from the first text; and generating minutes for the first text based on the associated content and the second record text to obtain the minutes text of the first text.

[0109] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the minutes generation method provided by the above-mentioned methods, the method comprising: obtaining a first text, and a first record text corresponding to the first text, the first record text including the focus content in the first text; obtaining a second record text corresponding to a second text, the second record text being the minutes text of the second text, the first text and the second text being texts on the same subject; determining the associated content of the first record text from the first text; and generating minutes for the first text based on the associated content and the second record text to obtain the minutes text of the first text.

[0110] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0111] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0112] 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 various embodiments of the present invention.

Claims

1. A method for generating minutes, characterized in that: include: Acquire a first text and a first record text corresponding to the first text, where the first record text includes a focus content of the first text; Obtaining a second record text corresponding to the second text, where the second record text is a summary text of the second text, and the first text and the second text are texts on the same subject; determining, from the first text, associated content of the first record text; Based on the associated content and the second record text, a minutes text is generated for the first text to obtain a minutes text of the first text.

2. The method for generating minutes according to claim 1, wherein: Generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text includes: Constructing a generation prompt text, wherein the generation prompt text is used to provide key information or instructions for guiding the generation of minutes; splicing the generated prompt text, the associated content, the second record text, and the first text to obtain a spliced text; Based on the minutes generation model, the concatenated text is applied to generate minutes for the first text to obtain a minutes text of the first text.

3. The method for generating minutes according to claim 1 or 2, characterized in that: The minutes text of the first text includes a summary text and a key points text; Generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text includes: Based on the associated content, extracting a summary of the first text to obtain the summary text; Based on the second record text and the summary text, key points are generated for the first text to obtain the key point text.

4. The method for generating minutes according to claim 3, wherein: Generating key points of the first text based on the second record text and the summary text to obtain the key point text includes: Extracting core information from the summary text and extracting key information from the second record text; Based on the association between the core information and the key information, and the difference between the core information and the key information, key points are generated for the first text to obtain the key point text.

5. The method for generating minutes according to claim 4, characterized in that: Generating key points of the first text based on the association between the core information and the key information, and the difference between the core information and the key information to obtain the key point text, includes: Determining a logical chain of the key text based on the association between the core information and the key information; enhancing the logic chain based on the differences between the core information and the key information; Based on the enhanced logical chain, key points are generated for the first text to obtain the key point text.

6. The method for generating minutes according to claim 1 or 2, characterized in that: The second record text is determined based on the following steps: Extracting keywords from the document data associated with the first text and / or the recorded text; The second record text is obtained by searching the historical records based on the keyword.

7. The method for generating minutes according to claim 6, wherein: The step of searching the historical records based on the keyword to obtain the second record text includes: Combining the keyword, synonyms of the keyword, and related words of the keyword to obtain a query phrase; Based on the keyword and the query phrase, the historical records are searched to obtain the second record text.

8. The method for generating minutes according to claim 1 or 2, characterized in that: The second record text is stored in the minutes database; generating a minutes of the first text based on the associated content and the second record text to obtain a minutes text of the first text, and then further comprising: The minutes database is updated based on the minutes text of the first text.

9. A minutes generating device, characterized in that: include: A first acquiring unit is configured to acquire a first text and a first record text corresponding to the first text, wherein the first record text includes a focus content in the first text; A second acquiring unit is configured to acquire a second record text corresponding to the second text, where the second record text is a summary text of the second text, and the first text and the second text are texts on the same subject; a determining unit, configured to determine associated content of the first record text from the first text; A generating unit is configured to generate a minutes text of the first text based on the associated content and the second record text to obtain a minutes text of the first text.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the minutes generation method according to any one of claims 1 to 8 is implemented.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating minutes according to any one of claims 1 to 8 is implemented.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for generating minutes according to any one of claims 1 to 8 is implemented.

Images