CN117668176A - Intelligent knowledge explanation system based on large language model - Google Patents
Intelligent knowledge explanation system based on large language model Download PDFInfo
- Publication number
- CN117668176A CN117668176A CN202311505141.4A CN202311505141A CN117668176A CN 117668176 A CN117668176 A CN 117668176A CN 202311505141 A CN202311505141 A CN 202311505141A CN 117668176 A CN117668176 A CN 117668176A
- Authority
- CN
- China
- Prior art keywords
- language model
- knowledge
- large language
- questions
- vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000013598 vector Substances 0.000 claims abstract description 55
- 239000012925 reference material Substances 0.000 claims abstract description 17
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims abstract description 6
- 238000000034 method Methods 0.000 claims description 13
- 239000012634 fragment Substances 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 239000004956 Amodel Substances 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 229920006104 Amodel® Polymers 0.000 description 2
- 230000001427 coherent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent knowledge interpretation system based on a large language model, which comprises: the input module is used for receiving the problem input by the user; the feature extraction module is used for extracting feature vectors of the problems by adopting an embedded model; the searching and matching module is used for searching the reference materials which are most matched with the problem in the vector database based on the feature vector of the problem; the knowledge generation module is used for integrating the questions, the historical chat records and the reference data through templates, inputting the integrated questions, the historical chat records and the reference data into the large language model, and generating explanation knowledge corresponding to the questions through calculation of the large language model; and the output module is used for outputting explanation knowledge. The system can realize accurate answer and knowledge explanation without limitation.
Description
Technical Field
The invention belongs to the technical field of knowledge question and answer, and particularly relates to an intelligent knowledge interpretation system based on a large language model.
Background
The online customer service is an indispensable ring for providing services to the outside by application programs, most merchants or public places currently adopt a robot customer service mode to provide 24-hour uninterrupted online services, and the all-weather service can meet various requirements of users at any time and any place and improve the satisfaction degree of the users on the requirements.
Compared with manual customer service, the robot customer service can reduce the operation cost of enterprises to a certain extent, one robot customer service can provide services for a plurality of users at the same time, and can realize quick response and solve the consultation problem, thereby meeting the user demands and enhancing the competitiveness of the enterprises.
Although the robot customer service can provide all-weather service and has the advantages of saving cost, fast response, solving problems and expanding service range and coverage, the current robot customer service also has a plurality of problems, including:
1. mechanical dialogue: the chat mode of traditional robot customer service is generally hard and mechanized, and cannot provide a natural conversation experience similar to that of real people. This results in a lack of humanization and emotional communication when the user often perceives a conversation with the robot.
2. Lack of context understanding: traditional robot customer service often cannot accurately understand the context information in a conversation and cannot respond appropriately according to the content of the previous conversation. Thus, the user may need to repeatedly provide the same information or interpretation, increasing the complexity and time cost of the interaction.
3. Lack of flexibility and creativity: the robotic customer service system is typically operated based on preset modes and logic, lacking creativity and flexibility. In a scenario facing complex problems or requiring creative thinking, they may not provide innovative and personalized solutions.
4. Traditional keyword matching: traditional robot customer service mainly relies on keyword matching technology, and this method is easily limited by limitations of input keyword selection and matching algorithms. It cannot handle complex problems or track dialog contexts and cannot provide a higher level of semantic understanding.
Disclosure of Invention
In view of the above, the present invention aims to provide an intelligent knowledge interpretation system based on a large language model, which can realize accurate answers and knowledge interpretation without limitation.
In order to achieve the above object, the present invention provides an intelligent knowledge interpretation system based on a large language model, comprising:
the input module is used for receiving the problem input by the user;
the feature extraction module is used for extracting feature vectors of the problems by adopting an embedded model;
the searching and matching module is used for searching the reference materials which are most matched with the problem in the vector database based on the feature vector of the problem;
the knowledge generation module is used for integrating the questions, the historical chat records and the reference data through templates, inputting the integrated questions, the historical chat records and the reference data into the large language model, and generating explanation knowledge corresponding to the questions through calculation of the large language model;
and the output module is used for outputting explanation knowledge.
In one embodiment, the questions entered by the user are in text form or in speech form, and when in speech form, the questions in speech form are converted into text form and then input to the feature extraction module.
In one embodiment, the vector database stores knowledge of vector representations, and the construction process includes:
after various types of reference material documents of the knowledge are cut into document fragments, feature vectors of the document fragments are extracted as the knowledge of vector representation and stored.
In one embodiment, the various types of reference documents include pdf documents, word documents, and txt documents.
In one embodiment, when searching the reference data which is most matched with the problem in the vector database, the vector with the highest similarity is selected as the reference data which is most matched with the problem by calculating the similarity between the feature vector of the problem and each vector in the vector database.
In one embodiment, the similarity is calculated using a cosine similarity algorithm, a Euclidean distance algorithm, or a Pearson correlation coefficient algorithm.
In one embodiment, the large language model is calculated by capturing information in a historical conversation based on historical chat records and reference material understanding context information to generate accurate and comprehensive interpretation knowledge.
In one embodiment, the output narrative knowledge is in text form or in speech form.
In one embodiment, the large language model is trained prior to being applied, comprising:
the large language model is pre-trained on the large data set, so that the large language model learns language knowledge and semantic understanding capability, and the pre-trained large language model is fine-tuned in a specific field.
inoneembodiment,thelargelanguagemodelincludesaGPTfamilyoflanguagemodules,aPaLM-Emodel,anERNIEBotmodel,aHflmodel,aMeenamodel,anM6model,aLaMDAmodel,aPaLM-Amodel.
Compared with the prior art, the invention has the beneficial effects that at least the following steps are included:
the embedded model is used for extracting the characteristics of the questions of the user, and converting the questions into vector representations, so that the semantics and the context information of the questions can be better captured, and the understanding and accurate answering capabilities of the questions are improved.
Based on the feature vector of the question, searching and matching the most similar answer knowledge in the vector database as reference material can accurately match the related knowledge and provide more accurate and proper answer.
Answers are obtained from a wider context using a large language model in combination with user questions and historical chat logs and references. By passing the history chat log, the background and context of the conversation can be better understood and grasped, making the answer more perceived and personalized for a real chat.
By comprehensively utilizing the embedded model, the similarity matching algorithm and the large language model, more accurate and personalized answer is realized. Compared with traditional robot customer service, the method is not limited to fixed template and keyword matching, but can adapt to various contexts and complex problems, and provides more accurate and complete solutions.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a large language model based intelligent knowledge interpretation system according to an embodiment;
FIG. 2 is a schematic diagram of the construction of a vector database according to an embodiment;
FIG. 3 is a flow chart of intelligent knowledge interpretation provided by an embodiment;
fig. 4 is a detailed flow chart of user intelligence questions and answers provided by an embodiment.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
As shown in fig. 1, an intelligent knowledge interpretation system 100 based on a large language model provided in an embodiment includes an input module 110, a feature extraction module 120, a search matching module 130, a knowledge generation module 140, and an output module 150.
The input module 120 is configured to receive a question input by a user, which may be a question input in text form or a question input in voice form. When a question in a voice form is input, the question in a voice form is converted into a text form and then input to the feature extraction module 120.
The feature extraction module 120 is configured to extract feature vectors of the problem using an embedded model. Specifically, the problems in the text form are input into an embedded model after initial coding, and the embedded model can adopt a convolutional neural network constructed by a convolutional layer, a pooling layer and an activation layer to realize feature extraction.
The search matching module 130 is configured to search the vector database for a reference material that best matches the problem based on the feature vector of the problem. As shown in fig. 2, the vector database construction process is as follows: and acquiring various types of reference data documents uploaded by user definition, cutting the various types of reference data documents of knowledge into document fragments, extracting feature vectors of the document fragments as knowledge represented by vectors, and storing the feature vectors. Various types of documents include pdf documents, word documents, and txt documents.
The knowledge generation module 140 is configured to integrate the questions, the history chat records, and the reference data through templates, input the integrated questions, history chat records, and reference data into a large language model, and generate explanation knowledge corresponding to the questions through calculation of the large language model. Specifically, when searching the reference materials which are most matched with the problem in the vector database, the vector with the highest similarity is screened as the reference materials which are most matched with the problem by calculating the similarity between the feature vector of the problem and each vector in the vector database. The similarity can be calculated by adopting a cosine similarity algorithm, a Euclidean distance algorithm or a Pearson correlation coefficient algorithm.
The large language model is a model constructed based on deep learning technology and capable of generating natural language text, which mainly builds a huge vocabulary by learning massive text data and predicts the occurrence probability of the next word according to the input context information.
inanembodiment,thelargelanguagemodulesforintelligentknowledgeinterpretationincludeGPTserieslanguagemodules,PaLM-Emodels,ERNIEBotmodels,Hflmodels,Meenamodels,M6models,LaMDAmodels,PaLM-Amodels. Among them, the GPT series of OpenAI is one of the earliest large language models, which has hundreds of billions of parameters, and can realize powerful natural language understanding and generating capability. The PaLM-E model has 5400 hundred million parameters and is unique in that the language model and the visual model can be combined to realize multi-mode understanding and generation. The ERNIE Bot model can generate high-quality text content, has 3.5 hundred million parameters based on a transducer architecture, and supports Chinese and English. The Hfl model is focused on understanding and generation in the financial field, has 2 hundred million parameters, and can generate professional financial analysis reports and investment suggestions. The Meena model can generate realistic text and images, which have 2.6 billions of parameters based on the transducer architecture. The M6 model has 3 hundred million parameters and can be applied to various scenes including searching, recommending, advertising, etc. The LaMDA model is a large language model based on a transducer architecture, can generate, understand and generate text contents, and is unique in that a conversational interaction mode is adopted, so that a user can communicate with a computer more naturally. thePaLM-amodelisbasedonthePaLMarchitecture,andcanbeappliedtovariousscenes,includingtextclassification,question-answeringsystems,andthelike.
The large language model is trained before being applied, and comprises: the large language model is pre-trained on the large data set, so that the large language model learns language knowledge and semantic understanding capability, and the pre-trained large language model is fine-tuned on the specific field, so that the method is more suitable for the specific application field.
The output module 150 is configured to output the interpretation knowledge, and the specific output interpretation knowledge is in text form or voice form.
Through the modules, the embedded model is used for extracting the feature vector of the problem of the user, the problem is converted into the vector representation, and the reference data which is most similar to the question of the user is searched and extracted in the vector database through the similarity algorithm. The user's questions, history chat log, and the above-matched references are passed to a large language model to enable it to understand the context and conversation history. Compared with traditional robot customer service, the system can better capture information in the previous dialogue and generate answers according to the context, so that a more coherent and natural dialogue experience is realized. By means of the capability of the large language model, the intelligent knowledge interpretation system can provide more accurate answers, and the limitation of the traditional keyword matching method is eliminated. The method can understand the meaning of the questions, conduct semantic analysis and give more accurate and comprehensive answers.
As shown in fig. 3, the process of intelligent knowledge interpretation by using the intelligent knowledge interpretation system includes:
receiving a user input question through an input module; converting the user problem into a feature vector through a feature extraction module; performing similarity search in a vector database based on the feature vector of the problem by a search matching module, and matching to obtain reference materials with highest similarity; integrating the questions, the history chat records and the reference data through templates by a knowledge generation module, inputting the integrated questions, the history chat records and the reference data into a large language model, and generating explanation knowledge corresponding to the questions through calculation of the large language model; and outputting and displaying the generated explanation knowledge through an output module.
As shown in fig. 4, the embodiment further provides an intelligent question-answering detailed process implemented by using the intelligent knowledge explanation system, including:
1. a user initiates a question at a client;
2. acquiring a text of a user question, and converting the user question into a feature vector through an embedded model;
3. in a vector database storing reference materials, matching the reference materials of a plurality of documents with highest similarity to the problem of the user by using a similarity algorithm;
4. the service back end acquires a history chat record before the user in the cache, encapsulates the reference materials matched in the previous step, the user problems and a preset questioning template, and sends the reference materials, the user problems and the preset questioning template to the large language model;
5. the large language model analyzes the historical chat record based on the input and obtains the context information, tries to select proper text fragments and language expression modes to construct explanation knowledge according to the semantics, the questions and the reference materials of the historical chat record, and records the answers of the user questions and the large language model into a historical record cache;
6. returning the explanation knowledge of the true person of the large language model;
7. when the question is continuously asked, the steps 1-6 are repeated to continuously obtain accurate answers, realize the conversation feeling with a real person and provide unlimited answering capability.
The intelligent knowledge explanation system based on the large language model provided by the embodiment can be used for question-answering scenes in various fields, can be used for knowledge question-answering scenes in public places such as cultural relic exposition, can also be used for consultation question-answering scenes in the field of electronic commerce, and has the advantages that:
1. feature vector extraction: and extracting the feature vector of the user question through the embedded model. Thus, the problem can be converted into vector representation, and subsequent similarity calculation and text matching are facilitated.
2. Cosine similarity matching: and searching and extracting the text most similar to the user question in the vector database by using a cosine similarity algorithm. This process enables quick finding of relevant knowledge resources for subsequent answers.
3. Context understanding: the user's questions and history chat records are passed to a large language model to enable it to understand the context and conversation history. The system is better able to capture information in previous conversations and generate answers based on context, compared to traditional robot customer service, making conversations more coherent, natural, and more similar to the feel of real-person chat.
4. Accurate answer: by means of the capability of the large language model, the intelligent knowledge interpretation system can provide more accurate answers, and the limitation of the traditional keyword matching method is avoided. The method can understand the meaning of the questions, conduct semantic analysis and give more accurate and comprehensive answers.
In a word, the intelligent knowledge interpretation system based on the large language model overcomes the defects of the traditional robot customer service, so that a user can obtain better interaction experience and more accurate knowledge solution.
The foregoing detailed description of the preferred embodiments and advantages of the invention will be appreciated that the foregoing description is merely illustrative of the presently preferred embodiments of the invention, and that no changes, additions, substitutions and equivalents of those embodiments are intended to be included within the scope of the invention.
Claims (10)
1. An intelligent knowledge interpretation system based on a large language model, comprising:
the input module is used for receiving the problem input by the user;
the feature extraction module is used for extracting feature vectors of the problems by adopting an embedded model;
the searching and matching module is used for searching the reference materials which are most matched with the problem in the vector database based on the feature vector of the problem;
the knowledge generation module is used for integrating the questions, the historical chat records and the reference data through templates, inputting the integrated questions, the historical chat records and the reference data into the large language model, and generating explanation knowledge corresponding to the questions through calculation of the large language model;
and the output module is used for outputting explanation knowledge.
2. The intelligent knowledge interpretation system based on large language model as claimed in claim 1, wherein the questions inputted by the user are text form or voice form, and when the questions are voice form, the questions are converted into text form and then inputted into the feature extraction module.
3. The intelligent knowledge interpretation system based on large language model of claim 1, wherein the vector database stores knowledge of vector representation, and the construction process includes:
after various types of reference material documents of the knowledge are cut into document fragments, feature vectors of the document fragments are extracted as the knowledge of vector representation and stored.
4. The large language model based intelligent knowledge interpretation system of claim 1, wherein the various types of reference documents include pdf documents, word documents, and txt documents.
5. The intelligent knowledge interpretation system based on large language model as claimed in claim 1, wherein when searching the reference data which is the best match with the problem in the vector database, the vector with the highest similarity is selected as the reference data which is the best match with the problem by calculating the similarity between the feature vector of the problem and each vector in the vector database.
6. The intelligent knowledge interpretation system based on large language model of claim 5, wherein the similarity is calculated by using a cosine similarity algorithm, a Euclidean distance algorithm or a Pearson correlation coefficient algorithm.
7. The large language model based intelligent knowledge interpretation system of claim 1, wherein the large language model, when calculated, captures information in the historical conversations based on the historical chat log and the reference material understanding context information to generate accurate and comprehensive interpretation knowledge.
8. The large language model based intelligent knowledge interpretation system of claim 1, wherein the output interpretation knowledge is in text form or in speech form.
9. The large language model based intelligent knowledge interpretation system of claim 1, wherein the large language model is trained prior to being applied, comprising:
the large language model is pre-trained on the large data set, so that the large language model learns language knowledge and semantic understanding capability, and the pre-trained large language model is fine-tuned in a specific field.
10. thelargelanguagemodelbasedintelligentknowledgeinterpretationsystemofclaim1,whereinthelargelanguagemodelincludesaGPTserieslanguagemodule,aPaLM-Emodel,anERNIEBotmodel,aHflmodel,aMeenamodel,anM6model,alamamodel,aPaLM-amodel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311505141.4A CN117668176A (en) | 2023-11-13 | 2023-11-13 | Intelligent knowledge explanation system based on large language model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311505141.4A CN117668176A (en) | 2023-11-13 | 2023-11-13 | Intelligent knowledge explanation system based on large language model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117668176A true CN117668176A (en) | 2024-03-08 |
Family
ID=90083587
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311505141.4A Pending CN117668176A (en) | 2023-11-13 | 2023-11-13 | Intelligent knowledge explanation system based on large language model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117668176A (en) |
-
2023
- 2023-11-13 CN CN202311505141.4A patent/CN117668176A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008322B (en) | Method and device for recommending dialogues in multi-turn conversation scene | |
| CN111641514B (en) | Conference intelligence system, method for conference intelligence, and storage medium | |
| CN105843381B (en) | Data processing method for realizing multi-modal interaction and multi-modal interaction system | |
| CN106685916B (en) | Intelligent device and method for electronic conference | |
| CN112818107B (en) | Conversation robot for daily life and chat method thereof | |
| KR102100214B1 (en) | Method and appratus for analysing sales conversation based on voice recognition | |
| US20230394247A1 (en) | Human-machine collaborative conversation interaction system and method | |
| CN111666400B (en) | Message acquisition method, device, computer equipment and storage medium | |
| CN111813909A (en) | Intelligent question answering method and device | |
| CN112632244A (en) | Man-machine conversation optimization method and device, computer equipment and storage medium | |
| CN111429157A (en) | Method, device and equipment for evaluating and processing complaint work order and storage medium | |
| CN112364148B (en) | Deep learning method-based generative chat robot | |
| CN117521675A (en) | Information processing method, device, equipment and storage medium based on large language model | |
| KR20190046062A (en) | Method and apparatus of dialog scenario database constructing for dialog system | |
| CN113076770A (en) | Intelligent figure portrait terminal based on dialect recognition | |
| CN113392265A (en) | Multimedia processing method, device and equipment | |
| CN113901189A (en) | Digital human interaction method and device, electronic equipment and storage medium | |
| CN109684444A (en) | A kind of intelligent customer service method and system | |
| CN111858875A (en) | Intelligent interaction method, device, equipment and storage medium | |
| KR20180116104A (en) | Method and apparatus of dialog scenario database constructing for dialog system | |
| CN110931002B (en) | Man-machine interaction method, device, computer equipment and storage medium | |
| Kaviya et al. | Artificial intelligence based farmer assistant chatbot | |
| KR20210009266A (en) | Method and appratus for analysing sales conversation based on voice recognition | |
| CN117668176A (en) | Intelligent knowledge explanation system based on large language model | |
| CN116009692A (en) | Virtual character interaction strategy determination method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |