A knowledge sharing method, system and computer device based on large language model

Abstract

This invention discloses a knowledge sharing method based on a large language model. The method includes using the large language model to support knowledge sharers in various forms of knowledge sharing; absorbing acquired knowledge based on the large language model and accumulating it into a knowledge database; answering questions from questioners based on the large language model; and supplementing existing knowledge gaps through question-and-answer sessions with knowledge sharers. The invention also provides a knowledge sharing system and a computer device. The method provided by the invention can effectively optimize the knowledge sharing process while also providing high-quality responses to questions.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and in particular relates to a knowledge sharing method, system and computer equipment based on a large language model. Background Art

[0002] Large Language Models (LLMs) are a type of natural language processing technology based on deep learning. They are primarily trained on massive amounts of data to understand and generate natural language text. These models rely on hundreds of millions of parameters to operate. By learning from large corpora, they are capable of handling complex language tasks such as text generation, question-answering systems, machine translation, and semantic analysis. In recent years, with the growth of model size and advancements in computing power, the performance of large language models has become increasingly superior, making them a crucial tool in the field of modern artificial intelligence.

[0003] On the other hand, with the development of internet technology, numerous question-and-answer (Q&A)-based knowledge sharing platforms have emerged. In China, there are Zhihu, Baidu Knows, and Weibo Q&A, while abroad, there are Quora, Reddit, and Stack Overflow. These platforms cover a wide range of topics, from daily life to professional knowledge, greatly facilitating knowledge acquisition for internet users. However, these traditional Q&A platforms also have numerous shortcomings: 1. A significant proportion of knowledge sharing occurs on a question-first, followed by an answer, resulting in delayed responses to the questioner's questions. 2. While domain experts can organize and publicly share knowledge through tools like Zhihu columns, this still requires the questioner to possess certain information retrieval skills to find the appropriate answer. This is time-consuming and often limited by their search skills, preventing them from retrieving the optimal answer. 3. The knowledge sharer's answers are publicly available, posing a potential risk of privacy leakage, which undoubtedly undermines the motivation for knowledge sharing. The impact of the knowledge shared by the knowledge sharer on the questioner can only be analyzed through user likes and read counts. This hinders proper incentives for the sharer and the development of a business model that charges the questioner for information. Consequently, a healthy closed loop of paid knowledge services is not formed, hindering the wider sharing of knowledge.

[0004] Patent document CN119442333A discloses a watershed knowledge sharing method, device, computer equipment and medium. It calculates the knowledge evaluation value of the corresponding watershed business process by using the cost information, first ranking coefficient, second ranking coefficient and business relevance index of each business process that is beneficial to the business process in different business domains, thereby realizing the knowledge value evaluation of the relevant information of each business process, and carrying out subsequent watershed business knowledge sharing based on the knowledge evaluation value. Watershed knowledge sharing is realized based on the value of watershed business process information, and watershed business knowledge of different values ​​is shared with users who meet the corresponding requirements.

[0005] Patent document CN119203090A discloses a method and system for sharing private domain knowledge for different users, wherein the method includes the following steps: S1, when receiving an access request from the current user, obtaining the identity information and historical access information of the current user; S2, confirming the access permission level of the current user based on the identity information and historical access information; S3, judging whether the access permission level matches the access request; S4, if the access permission level matches the access request, dividing K groups of private domain knowledge into m groups of accessible knowledge and n groups of forgotten knowledge according to the access permission level, wherein each group of private domain knowledge corresponds to a LoRA sub-model trained, K=m+n; S5, merging the m LoRA sub-models corresponding to the m groups of accessible knowledge to obtain a merged model, and opening access rights to the merged model to the current user, wherein the current user obtains the m groups of accessible knowledge through the merged model. Summary of the Invention

[0006] The purpose of the present invention is to provide a knowledge sharing method, system and computer device based on a large language model, which can effectively optimize the steps of knowledge sharing and also provide high-quality question responses.

[0007] To achieve the first objective of the present invention, the following technical solution is provided: a knowledge sharing method based on a large language model, comprising:

[0008] Knowledge upload: Users input knowledge content and preset knowledge domain labels through a pre-built large language model. The large language model performs similarity matching between the knowledge content and the prior knowledge base corresponding to the knowledge domain labels to output the top N prior knowledge contents with the highest similarity.

[0009] The user makes judgments based on the first N pieces of prior knowledge output:

[0010] If there is prior knowledge content similar to the input knowledge content, the prior knowledge in the prior knowledge base is merged according to the input knowledge content;

[0011] If there is no similar prior knowledge content, the keywords in the input knowledge content are extracted and interactive text is generated to interact with the user. The user's reply text is used as the explanation content of the corresponding keyword as new prior knowledge and added to the prior knowledge base of the corresponding knowledge domain label;

[0012] Knowledge query: Users select a knowledge domain and ask questions using the large language model, and obtain knowledge prompt words in the question content based on a preset prompt word module;

[0013] Similarity matching is performed based on the prior knowledge base corresponding to the knowledge prompt words in the knowledge field to output the prior knowledge content with the highest similarity and use it as a basis to output the corresponding reply text through the large language model.

[0014] Specifically, the knowledge domain labels include general academic classification, subject professional classification and application field classification.

[0015] Specifically, the prior knowledge base uses embedding technology to convert input content into encoding vectors for storage.

[0016] Specifically, when the user makes judgments based on the first N pieces of prior knowledge content output, similarity matching needs to be performed again based on the user's input knowledge content after each prior knowledge base update is completed until all keywords in the input knowledge content can be matched to corresponding prior knowledge content.

[0017] Specifically, the large language model also includes a user reward mechanism, which refers to the process of quantifying the user's contribution to solving the input problem and giving the user reward points after the knowledge content uploaded by the user is adopted for generating the reply text.

[0018] Specifically, the expression of the user reward mechanism is as follows: ; Among them, m represents the total number of users sharing knowledge this time, Indicates the sharer j For the problem i Contribution points, .

[0019] In order to achieve the second object of the present invention, the following technical solution is provided: a knowledge sharing system, through the above-mentioned knowledge sharing method based on a large language model, includes an interaction unit, a knowledge uploading unit, a knowledge questioning unit, and a reward evaluation unit;

[0020] The interactive unit is used for users to input questions and upload knowledge, and the uploaded knowledge includes knowledge content and knowledge domain labels;

[0021] The knowledge uploading unit is used to process the uploaded knowledge content to update the corresponding prior knowledge base;

[0022] The knowledge questioning unit analyzes the input question and the data in the prior knowledge base to output the corresponding prior knowledge and generates a corresponding reply text based on the prior knowledge;

[0023] The reward evaluation unit quantifies the contribution to solving the input problem after the knowledge content is adopted to generate the reply text, so as to output the reward points for each user.

[0024] In order to achieve the third purpose of the present invention, the following technical solution is provided: a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute the above-mentioned knowledge sharing method based on a large language model.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] Based on the question-and-answer data between the questioner and the questioner, combined with the prior knowledge base for analysis, the quantitative traceability of knowledge is completed, which is used to reasonably distribute rewards to knowledge sharers, and ultimately makes it possible to build a virtuous cycle of paid knowledge sharing. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of a knowledge sharing method based on a large language model provided in this embodiment. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] This embodiment provides a knowledge sharing method based on a large language model, which includes:

[0030] Knowledge upload: Users input knowledge content and preset knowledge domain labels through a pre-built large language model. The large language model performs similarity matching between the knowledge content and the prior knowledge base corresponding to the knowledge domain labels to output the top N prior knowledge contents with the highest similarity.

[0031] The knowledge domain labels in this embodiment include general academic classification, subject professional classification, and application domain classification.

[0032] The user inputs knowledge content and preset knowledge domain labels through a pre-built large language model. The large language model performs similarity matching between the knowledge content and the prior knowledge base corresponding to the knowledge domain labels to output the top N prior knowledge contents with the highest similarity.

[0033] The user makes judgments based on the first N pieces of prior knowledge output. That is, each time the prior knowledge base is updated, similarity matching needs to be performed again based on the user's input knowledge until all keywords in the input knowledge can be matched to the corresponding prior knowledge content:

[0034] If there is prior knowledge content similar to the input knowledge content, the prior knowledge in the prior knowledge base is merged according to the input knowledge content;

[0035] If there is no similar prior knowledge content, the keywords in the input knowledge content are extracted and interactive text is generated to interact with the user, and the user's reply text is used as the explanation content of the corresponding keywords as new prior knowledge and added to the prior knowledge base of the corresponding knowledge domain label.

[0036] Knowledge query: Users select a knowledge domain and ask questions using the large language model, and obtain knowledge prompt words in the question content based on a preset prompt word module;

[0037] Similarity matching is performed based on the prior knowledge base corresponding to the knowledge prompt words in the knowledge field to output the prior knowledge content with the highest similarity and use it as a basis to output the corresponding reply text through the large language model.

[0038] More specifically, if Figure 1 The schematic diagram of the method provided in this embodiment is shown as follows: Figure 1 It can be seen that after knowledge sharers obtain valuable knowledge in life and work, they can input it into the large language model A in the form of text, voice, picture, video, etc. at any time to complete the knowledge summary 1.

[0039] The large language model A judges the received knowledge. If it finds any parts that it does not understand, it will perform the interactive actions of asking questions 2 and answering 3 with the knowledge sharer, asking the knowledge sharer to further explain the corresponding knowledge points, so as to learn the parts that it does not understand one by one until the knowledge points are fully understood.

[0040] The large language model A uses embedding technology to convert the understood knowledge points into vector form that can be directly used by the model and stores them in the vector knowledge database, completing the accumulation of knowledge4.

[0041] When a questioner asks a question to the device of the present invention, the large language model A will retrieve 6 relevant knowledge from the knowledge vector database, and then the large language model A will analyze the questioner's question and the retrieved relevant knowledge, thereby completing the interactive action of asking questions 4 and answering 7 with the questioner.

[0042] If it is determined that the retrieved knowledge is not sufficient to answer the questioner's question, the large language model A will analyze the topic corresponding to the question, then find the three most relevant knowledge sharers based on the topic, and interact with the knowledge sharers to ask and answer questions, thereby obtaining further knowledge supplements and completing further knowledge accumulation11. Finally, the large language model A answers the questioner based on the supplemented knowledge10.

[0043] The large language model B reads 12 and analyzes the conversation between the large language model A and the questioner, and accesses 13 the knowledge database. For answers that provide practical help, it quantifies and analyzes the contribution of each knowledge sharer to these questions, thereby completing the work of knowledge tracing and rewarding 14.

[0044] In this embodiment, the knowledge content mentioned is uploaded in the form of text, voice, video, and image, and is converted into a vector form that can be directly used by the computer through embedding technology and stored in the prior knowledge base.

[0045] Except for text input, other input forms are processed as follows:

[0046] Voice input: Use an open source model to convert it into text. In this embodiment, the IE Whisper model is used to process voice data.

[0047] Image type input: Use a multimodal large model to understand it and output the corresponding text description. In this embodiment, the ie Qwen2.5-VL-72B or InternVL 1.5 model is used to process the image data.

[0048] Video input: First, frame sampling is performed to construct a set of images corresponding to the video. Each image in the set is then understood using a large multimodal model, and a corresponding text description is output. Finally, all text descriptions are concatenated and polished along the timeline of the video to serve as the final text description of the video data.

[0049] After converting the input knowledge content into a unified data form based on the above-listed methods, the final text description will be refined through a large language model to obtain the final refined text expression.

[0050] After completing the above data conversion, complete the vectorized storage process of the text description through the following steps:

[0051] Text preprocessing and segmentation, due to the token limitations of large language models, requires segmenting longer texts into smaller segments (for example, each 256 words is a segment), with each segment serving as a unit of knowledge.

[0052] Text vectorization uses text embedding technology to vectorize the segmented text and use semantically similar text segments to have close vector representations. In this embodiment, Sentence Transformers is used to complete the text embedding work.

[0053] Vector storage and indexing: vectorized text is stored in a priori knowledge base, thereby indexing each document fragment to support fast retrieval. In this embodiment, a public database such as Chroma, Milvus or Pinecone is used to build the corresponding priori knowledge base.

[0054] In this embodiment, the knowledge query part includes processing logic for two situations. One is when there is corresponding knowledge content in the prior knowledge base, the query question is responded to; the other is when the retrieved knowledge is not sufficient to answer the questioner's question, the knowledge content needs to be supplemented.

[0055] When the system is unable to adequately answer the questioner's question, the large language model will first analyze the topic of the question, and then find the three most relevant knowledge sharers based on the topic. For the knowledge points that the system does not understand, it will ask the knowledge sharers to further explain the corresponding knowledge points and obtain further knowledge supplements from the knowledge sharers.

[0056] The large language model in this embodiment also includes a user reward mechanism. The user reward mechanism is to quantify the contribution of the user's uploaded knowledge content to solving the input problem and give the user reward points after the content is adopted for reply text generation. The specific process is as follows:

[0057] First, we need to quantitatively analyze how much each answer helps the questioner. Assuming there are n answers in a question-and-answer session, we can first give each answer a score. This score can be directly input by the questioner, or it can be automatically generated by the large language model B through analysis of the questioner's answer.

[0058] That is, the scores are normalized to obtain the final quantitative score of answer i: Then, combined with the prior knowledge base, we quantitatively analyze the contribution of each knowledge sharer to the questions that are actually helpful. Assuming there are m knowledge sharers, the final contribution of knowledge sharer j to the questioner is: ; Among them, m represents the total number of users sharing knowledge this time, Indicates the sharer j For the problem i Contribution points, .

[0059] A knowledge sharing system provided in this embodiment is implemented by the knowledge sharing method based on the large language model provided in the above embodiment, and includes an interaction unit, a knowledge uploading unit, a knowledge questioning unit, and a reward evaluation unit.

[0060] The interactive unit is used for users to input questions and upload knowledge, and the uploaded knowledge includes knowledge content and knowledge domain labels.

[0061] The knowledge uploading unit is used to process the uploaded knowledge content to update the corresponding prior knowledge base.

[0062] The knowledge questioning unit analyzes the input question and the data in the prior knowledge base to output the corresponding prior knowledge and generates the corresponding reply text based on the prior knowledge.

[0063] The reward evaluation unit quantifies the contribution to solving the input problem after the knowledge content is adopted to generate the reply text, so as to output the reward points of each user.

[0064] The above system can provide each knowledge sharer with a quantitative contribution value for each problem, which can be used to assist in reasonable rewards for knowledge sharers and ultimately build a virtuous cycle of paid knowledge sharing.

[0065] In addition, this embodiment also provides a computer device, including: a memory and a processor, the memory and the processor are communicatively connected to each other, wherein computer instructions are stored in the memory, and the processor executes the steps of the knowledge sharing method based on the large language model provided in the above embodiment by executing the computer instructions.

[0066] In addition, the terms "upper", "lower", "inner", "outer", "front", and "back" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the present invention.

[0067] Of course, the above description is only a specific embodiment of the present invention and is not intended to limit the scope of implementation of the present invention. Any equivalent changes or modifications made based on the structure, features and principles described in the scope of the patent application of the present invention should be included in the scope of the patent application of the present invention.

[0068] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A knowledge sharing method based on a large language model, characterized in that: include: Knowledge upload: Users input knowledge content and preset knowledge domain labels through a pre-built large language model. The large language model performs similarity matching between the knowledge content and the prior knowledge base corresponding to the knowledge domain labels to output the top N prior knowledge contents with the highest similarity. The user makes judgments based on the first N pieces of prior knowledge output: If there is prior knowledge content similar to the input knowledge content, the prior knowledge in the prior knowledge base is merged according to the input knowledge content; If there is no similar prior knowledge content, the keywords in the input knowledge content are extracted and interactive text is generated to interact with the user. The user's reply text is used as the explanation content of the corresponding keyword as new prior knowledge and added to the prior knowledge base of the corresponding knowledge domain label; Knowledge query: Users select a knowledge domain and ask questions using the large language model, and obtain knowledge prompt words in the question content based on a preset prompt word module; Perform similarity matching based on the prior knowledge base corresponding to the knowledge prompt words in the knowledge field to output the prior knowledge content with the highest similarity and use it as a basis to output the corresponding reply text through the large language model; If it is determined that the retrieved knowledge is not sufficient to answer the user's question, the subject of the question is analyzed through the large language model, and an interactive action of asking and answering questions is carried out with the knowledge uploader, so as to obtain further prior knowledge supplements and complete the supplementation of the prior knowledge base; The large language model also includes a user reward mechanism, which is to quantify the user's contribution to solving the input problem and give the user reward points after the knowledge content uploaded by the user is adopted for the generation of the reply text; The expression of the user reward mechanism is as follows: ; Among them, m represents the total number of users sharing knowledge this time, represents the contribution of sharer j to question i, , ; ;in, represents the quality score of the i-th answer, , is the total number of responses.

2. The knowledge sharing method based on a large language model according to claim 1, characterized in that: The knowledge domain labels include general academic classification, subject professional classification and application domain classification.

3. The knowledge sharing method based on a large language model according to claim 1, characterized in that: The prior knowledge base uses embedding technology to convert input content into encoding vectors for storage.

4. The knowledge sharing method based on a large language model according to claim 1, characterized in that: When the user makes judgments based on the first N pieces of prior knowledge content output, each time the prior knowledge base is updated, similarity matching needs to be re-performed based on the user's input knowledge content until all keywords in the input knowledge content can be matched to the corresponding prior knowledge content.

5. A knowledge sharing system, characterized in that: The method is implemented by the large language model-based knowledge sharing method according to any one of claims 1 to 4, comprising an interaction unit, a knowledge uploading unit, a knowledge questioning unit, and a reward evaluation unit; The interactive unit is used for users to input questions and upload knowledge, and the uploaded knowledge includes knowledge content and knowledge domain labels; The knowledge uploading unit is used to process the uploaded knowledge content to update the corresponding prior knowledge base; The knowledge questioning unit analyzes the input question and the data in the prior knowledge base to output the corresponding prior knowledge and generates a corresponding reply text based on the prior knowledge; The reward evaluation unit quantifies the contribution to solving the input problem after the knowledge content is adopted to generate the reply text, so as to output the reward points for each user.

6. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the steps of the knowledge sharing method based on a large language model according to any one of claims 1 to 4 by executing the computer instructions.

Images