CN116842152A - Science and technology policy question-answering method and device for fine-tuning language big model - Google Patents
Science and technology policy question-answering method and device for fine-tuning language big model Download PDFInfo
- Publication number
- CN116842152A CN116842152A CN202310681071.1A CN202310681071A CN116842152A CN 116842152 A CN116842152 A CN 116842152A CN 202310681071 A CN202310681071 A CN 202310681071A CN 116842152 A CN116842152 A CN 116842152A
- Authority
- CN
- China
- Prior art keywords
- policy
- document
- model
- language
- technical
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000005516 engineering process Methods 0.000 title claims description 8
- 239000013598 vector Substances 0.000 claims abstract description 93
- 239000011159 matrix material Substances 0.000 claims description 36
- 238000012549 training Methods 0.000 claims description 26
- 238000010276 construction Methods 0.000 claims description 9
- 238000005520 cutting process Methods 0.000 claims description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses a technological policy question-answering method and device for fine-tuning a language model, and relates to the technical field of language models. The method comprises the following steps: collecting questions and answers from the technical policy field data, and constructing a question-answer data set of the technical policy field; saving the technical policy rules as a text data file; fine tuning the pre-trained language big model by using a question-answer data set in the technical policy field; analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval; when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, then the acquired document vector is used as a Prompt, and a trimmed language large model is input to obtain a model output result. The application realizes a more accurate, efficient and highly-adaptive technological policy question-answering system and provides more reliable support for technological policy decision and consultation.
Description
Technical Field
The application relates to the technical field of language models, in particular to a technological policy question-answering method and device for fine-tuning a language model.
Background
Currently, large language models have made important breakthroughs in the field of natural language processing, such as GPT-3, GLM, and the like. The models can generate high-quality texts, and have wide application prospects. However, these models have some drawbacks in dealing with problems in certain areas. In particular, when applied to the technical policy field, existing language big models often lack understanding of the expertise and complexity, resulting in poor effectiveness in solving the technical policy problem, easy misleading answers or lack of confidence.
Disclosure of Invention
The technical problem to be solved by the application is to provide a technological policy question-answering method and a technological policy question-answering device for fine-tuning a language big model, which uses a question-answering data set in the technological policy field to fine-tune a pre-trained language big model, and uses vectorization engineering to enable the language big model to have higher accuracy and reliability in question-answering tasks in the technological policy field.
In a first aspect, the present application provides a technological policy question-answering method for fine-tuning a language big model, including:
technology policy data set construction: collecting questions and answers from the technical policy field data, and constructing a question-answer data set of the technical policy field; saving the technical policy rules as a text data file;
language big model fine tuning: fine tuning a pre-trained language large model GLM by using a question-answer data set in the technical policy field, and simulating the updating quantity of parameters by adding a bypass beside an original model and low-rank decomposition;
vectorization engineering: analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval;
answer output: when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, the acquired document vector is used as a Prompt, a trimmed language big model is input, and a model output result is output as an answer.
Further, the language big model fine tuning specifically includes:
let the pre-trained matrix be W 0 ∈R d*k Its update is expressed as:
W 0 +ΔW＝W 0 +BA,B∈R d*r ,A∈R r*k ，
where d, k represents the dimension of the matrix, rank r<<min (d, k), R represents a real set, W 0 Representing a pre-training matrix, the training method is as follows:
during training, the original model is fixed, and only the dimension-reducing matrix A and the dimension-increasing matrix B are trained;
during reasoning, BA is added to the original parameters, and no extra reasoning delay is introduced;
initializing, wherein A adopts Gaussian distribution initialization, B initializes to all 0, and a bypass is ensured to be a 0 matrix at the beginning of training;
rank selection: for general tasks, r=any of 1,2,4,8 values, and for some widely divergent tasks, a larger r is chosen.
Further, the vectorization process specifically includes the following steps:
loading a document: analyzing text data files of technical policy regulations, and converting all the grabbed policy files into a TXT text;
splitting a document: cutting TXT text into document vectors, wherein the length of the document vectors does not exceed a specified value, and the last character is a specified separator;
storing and retrieving document vectors: the document vectors are stored in a database supporting vector retrieval for matching the most similar document vectors from among the document vectors stored in the database after calculating the text vectors from the input text.
Further, all policy files captured are converted into one TXT text using LangChain library.
Further, the database supporting vector retrieval is a vector store.
In a second aspect, the present application provides a technological policy question-answering apparatus for fine-tuning a language big model, comprising:
the science and technology policy data set construction module: collecting questions and answers from the technical policy field data, and constructing a question-answer data set of the technical policy field; saving the technical policy rules as a text data file;
language big model fine tuning module: fine tuning a pre-trained language large model GLM by using a question-answer data set in the technical policy field, and simulating the updating quantity of parameters by adding a bypass beside an original model and low-rank decomposition;
the vectorization engineering module: analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval;
answer output module: when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, the acquired document vector is used as a Prompt, a trimmed language big model is input, and a model output result is output as an answer.
Further, the language big model fine tuning module is specifically configured to perform:
let the pre-trained matrix be W 0 ∈R d*k Its update is expressed as:
W 0 +ΔW＝W 0 +BA,B∈R d*r ,A∈R r*k ，
where d, k represents the dimension of the matrix, rank r<<min (d, k), R represents a real set, W 0 Representing a pre-training matrix, the training method is as follows:
during training, the original model is fixed, and only the dimension-reducing matrix A and the dimension-increasing matrix B are trained;
during reasoning, BA is added to the original parameters, and no extra reasoning delay is introduced;
initializing, wherein A adopts Gaussian distribution initialization, B initializes to all 0, and a bypass is ensured to be a 0 matrix at the beginning of training;
rank selection: for general tasks, r=any of 1,2,4,8 values, and for some widely divergent tasks, a larger r is chosen.
Further, the vectorization module is specifically configured to perform the following steps:
loading a document: analyzing text data files of technical policy regulations, and converting all the grabbed policy files into a TXT text;
splitting a document: cutting TXT text into document vectors, wherein the length of the document vectors does not exceed a specified value, and the last character is a specified separator;
storing and retrieving document vectors: the document vectors are stored in a database supporting vector retrieval for matching the most similar document vectors from among the document vectors stored in the database after calculating the text vectors from the input text.
Further, all policy files captured are converted into one TXT text using LangChain library.
Further, the database supporting vector retrieval is a vector store.
One or more technical solutions provided in the embodiments of the present application at least have the following technical effects or advantages:
the application effect of the language big model in the technical policy field can be improved through improvement points such as the construction of the technical policy data set, the fine adjustment method of the model, the injection of field knowledge and the like, the input limitation of the model is solved through vectorization and segmentation, and finally the answer accuracy is greatly improved through the Prompt engineering. By the method, a more accurate, efficient and highly-adaptive technological policy question-answering system can be realized, and more reliable support is provided for technological policy decision and consultation.
The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.
Drawings
The application will be further described with reference to examples of embodiments with reference to the accompanying drawings.
FIG. 1 is a flow chart of a method according to a first embodiment of the application;
fig. 2 is a schematic structural diagram of a device in a second embodiment of the present application.
Detailed Description
The embodiment of the application provides a technological policy question-answering method and device for fine-tuning a language big model, which are based on the trained language big model, and combines the structure and content characteristics of documents to realize automatic interpretation and knowledge extraction of scientific research documents, thereby improving the accuracy of interpretation and understanding.
The technical scheme in the embodiment of the application has the following overall thought:
data set construction: questions and answers are collected from relevant documents, policy documents and expert knowledge in the technical policy field, a question-answer data set specific to the technical policy field is constructed, and technical policy regulations are stored as text documents.
Fine tuning of the pre-training model: by introducing domain knowledge and data related to the technical policy in the pre-training process and fine-tuning specific task data of the domain, the language model has better technical policy understanding capability.
Vectorization engineering: aiming at scientific and technological data in a certain field, the input limit of a model is solved through vectorization and segmentation by collecting files, and finally, the answer accuracy is greatly improved through a Prompt project. .
Example 1
The embodiment provides a technological policy question-answering method for fine-tuning a language big model, as shown in fig. 1, which may include the following procedures:
technology policy data set construction: questions and answers are collected from technical policy domain data (e.g., domain knowledge, policy rules, related terms, and practice experience related to a technical policy, etc.), a question-answer data set of the technical policy domain is constructed, the data format may be { q, a }, q represents the collected technical policy questions, and a represents answers to the technical policy questions. The technological policy rules are stored as text data files, all policy rules can be collected into one folder, and the storage formats are pdf and txt.
Language big model fine tuning: the pre-trained language big model GLM is fine-tuned by using a question-answer data set in the technical policy field, and the update amount of parameters is simulated by adding a bypass beside the original model and low-rank decomposition. The process is as follows:
let the pre-trained matrix be W 0 ∈R d*k Its update is expressed as:
W 0 +ΔW＝W 0 +BA,B∈R d*r ,A∈R r*k ，
where d, k represents the dimension of the matrix, rank r<<min (d, k), R represents a real set, W 0 Representing a pre-training matrix, the training method is as follows:
during training, the original model is fixed, and only the dimension-reducing matrix A and the dimension-increasing matrix B are trained;
during reasoning, BA is added to the original parameters, and no extra reasoning delay is introduced;
initializing, wherein A adopts Gaussian distribution initialization, B initializes to all 0, and a bypass is ensured to be a 0 matrix at the beginning of training;
rank selection: for general tasks, r=any of 1,2,4,8 values, and for some widely divergent tasks, a larger r is chosen. R=8 in this embodiment.
Vectorization engineering: analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval; the process is as follows:
loading a document: analyzing text data files of technical policy regulations, and converting all the grabbed policy files into a TXT text by using a Langchain library;
splitting a document: cutting TXT text into document vectors, wherein the length of the document vectors does not exceed a specified value (in order to meet the limitation of the input length of a model, such as 2048token is not exceeded), and the last characters are specified separators (such as line feed, period, and the like), so that the integrity of semantics is ensured;
storing and retrieving document vectors: the document vectors are stored in a database (e.g., vector store) supporting vector retrieval for matching the most similar document vectors from among the document vectors stored in the database after calculating the text vectors from the input text, i.e., for finding several most relevant documents based on user questions.
Answer output: when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, the acquired document vector is used as a Prompt, a language big model after fine adjustment is input, and a model output result is used as an answer to be output, so that a more accurate result is obtained.
Based on the same inventive concept, the application also provides a device corresponding to the method in the first embodiment, and the details of the second embodiment are shown.
Example two
In this embodiment, a technological policy question-answering device for fine-tuning a language big model is provided, as shown in fig. 2, including:
the science and technology policy data set construction module: collecting questions and answers from the technical policy field data, and constructing a question-answer data set of the technical policy field; saving the technical policy rules as a text data file;
language big model fine tuning module: fine tuning a pre-trained language large model GLM by using a question-answer data set in the technical policy field, and simulating the updating quantity of parameters by adding a bypass beside an original model and low-rank decomposition;
the language big model fine tuning module is specifically configured to execute:
let the pre-trained matrix be W 0 ∈R d*k Its update is expressed as:
W 0 +ΔW＝W 0 +BA,B∈R d*r ,A∈R r*k ，
where d, k represents the dimension of the matrix, rank r<<min (d, k), R represents a real set, W 0 Representing a pre-training matrix, the training method is as follows:
during training, the original model is fixed, and only the dimension-reducing matrix A and the dimension-increasing matrix B are trained;
during reasoning, BA is added to the original parameters, and no extra reasoning delay is introduced;
initializing, wherein A adopts Gaussian distribution initialization, B initializes to all 0, and a bypass is ensured to be a 0 matrix at the beginning of training;
rank selection: for general tasks, r=any of 1,2,4,8 values, and for some widely divergent tasks, a larger r is chosen.
The vectorization engineering module: analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval;
the vectorization module is specifically configured to perform the following steps:
loading a document: analyzing text data files of technical policy regulations, and converting all the grabbed policy files into a TXT text by using a Langchain library;
splitting a document: cutting TXT text into document vectors, wherein the length of the document vectors does not exceed a specified value, and the last character is a specified separator;
storing and retrieving document vectors: the document vectors are stored in a database (e.g., vector store) supporting vector retrieval for matching the most similar document vector from among the document vectors stored in the database after computing the text vectors from the input text.
Answer output module: when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, the acquired document vector is used as a Prompt, a trimmed language big model is input, and a model output result is output as an answer.
The application can improve the application effect of the language large model in the technical policy field through data set construction, fine tuning method, field knowledge injection and other improvement points, solves the model input limitation through vectorization and segmentation modes, and finally greatly improves the answer accuracy through the Prompt engineering. By the method, a more accurate, efficient and highly-adaptive technological policy question-answering system can be realized, and more reliable support is provided for technological policy decision and consultation. The method has the following advantages of 1 and high accuracy: by fine tuning the data set in the technical policy field, the language big model has higher accuracy and reliability in question-answering tasks in the field. 2. High efficiency: through injecting domain-specific knowledge, the language big model can better understand and answer questions related to technical policies, and the question and answer efficiency is improved. 3. High adaptability: the method can be suitable for different types of technological policy question-answering tasks, and has certain universality and flexibility. 4. Providing domain-specific support: practitioners in the technical policy field can rapidly acquire accurate technical policy related information by using the method of the application, thereby improving the working efficiency and the decision quality.
It will be apparent to those skilled in the art that embodiments of the present application may be provided as a method, apparatus, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While specific embodiments of the application have been described above, it will be appreciated by those skilled in the art that the specific embodiments described are illustrative only and not intended to limit the scope of the application, and that equivalent modifications and variations of the application in light of the spirit of the application will be covered by the claims of the present application.
Claims (10)
1. A technological policy question-answering method for fine-tuning a language big model, comprising:
technology policy data set construction: collecting questions and answers from the technical policy field data, and constructing a question-answer data set of the technical policy field; saving the technical policy rules as a text data file;
language big model fine tuning: fine tuning a pre-trained language large model GLM by using a question-answer data set in the technical policy field, and simulating the updating quantity of parameters by adding a bypass beside an original model and low-rank decomposition;
vectorization engineering: analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval;
answer output: when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, the acquired document vector is used as a Prompt, a trimmed language big model is input, and a model output result is output as an answer.
2. The method according to claim 1, wherein the language big model fine-tuning specifically comprises:
let the pre-trained matrix be W 0 ∈R d*k Its update is expressed as:
W 0 +ΔW＝W 0 +BA,B∈R d*r ,A∈R r*k ，
where d, k represents the dimension of the matrix, rank r<<min (d, k), R represents a real set, W 0 Representing a pre-training matrix, the training method is as follows:
during training, the original model is fixed, and only the dimension-reducing matrix A and the dimension-increasing matrix B are trained;
during reasoning, BA is added to the original parameters, and no extra reasoning delay is introduced;
initializing, wherein A adopts Gaussian distribution initialization, B initializes to all 0, and a bypass is ensured to be a 0 matrix at the beginning of training;
rank selection: for general tasks, r=any of 1,2,4,8 values, and for some widely divergent tasks, a larger r is chosen.
3. The method according to claim 1 or 2, characterized in that the vectorization step comprises in particular the following steps:
loading a document: analyzing text data files of technical policy regulations, and converting all the grabbed policy files into a TXT text;
splitting a document: cutting TXT text into document vectors, wherein the length of the document vectors does not exceed a specified value, and the last character is a specified separator;
storing and retrieving document vectors: the document vectors are stored in a database supporting vector retrieval for matching the most similar document vectors from among the document vectors stored in the database after calculating the text vectors from the input text.
4. A method according to claim 3, characterized in that: all policy files fetched are converted into one TXT text using LangChain library.
5. The method according to claim 1, characterized in that: the database supporting vector retrieval is a vector store.
6. A technological policy question-answering apparatus for fine-tuning a language model, comprising:
the science and technology policy data set construction module: collecting questions and answers from the technical policy field data, and constructing a question-answer data set of the technical policy field; saving the technical policy rules as a text data file;
language big model fine tuning module: fine tuning a pre-trained language large model GLM by using a question-answer data set in the technical policy field, and simulating the updating quantity of parameters by adding a bypass beside an original model and low-rank decomposition;
the vectorization engineering module: analyzing and segmenting a text data file of the technical policy regulation, and storing segmented document vectors into a database for supporting vector retrieval;
answer output module: when the user problem is acquired, at least one section of document vector matched with the problem is acquired through a database for supporting vector retrieval, the acquired document vector is used as a Prompt, a trimmed language big model is input, and a model output result is output as an answer.
7. The apparatus according to claim 6, wherein the language big model fine tuning module is specifically configured to perform:
let the pre-trained matrix be W 0 ∈R d*k Its update is expressed as:
W 0 +ΔW＝W 0 +BA,B∈R d*r ,A∈R r*k ，
where d, k represents the dimension of the matrix, rank r<<min (d, k), R represents a real set, W 0 Representing a pre-training matrix, the training method is as follows:
during training, the original model is fixed, and only the dimension-reducing matrix A and the dimension-increasing matrix B are trained;
during reasoning, BA is added to the original parameters, and no extra reasoning delay is introduced;
initializing, wherein A adopts Gaussian distribution initialization, B initializes to all 0, and a bypass is ensured to be a 0 matrix at the beginning of training;
rank selection: for general tasks, r=any of 1,2,4,8 values, and for some widely divergent tasks, a larger r is chosen.
8. The apparatus according to claim 6 or 7, wherein the vectorization module is specifically configured to perform the following steps:
loading a document: analyzing text data files of technical policy regulations, and converting all the grabbed policy files into a TXT text;
splitting a document: cutting TXT text into document vectors, wherein the length of the document vectors does not exceed a specified value, and the last character is a specified separator;
storing and retrieving document vectors: the document vectors are stored in a database supporting vector retrieval for matching the most similar document vectors from among the document vectors stored in the database after calculating the text vectors from the input text.
9. The apparatus according to claim 8, wherein: all policy files fetched are converted into one TXT text using LangChain library.
10. The apparatus according to claim 6, wherein: the database supporting vector retrieval is a vector store.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310681071.1A CN116842152A (en) | 2023-06-09 | 2023-06-09 | Science and technology policy question-answering method and device for fine-tuning language big model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310681071.1A CN116842152A (en) | 2023-06-09 | 2023-06-09 | Science and technology policy question-answering method and device for fine-tuning language big model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116842152A true CN116842152A (en) | 2023-10-03 |
Family
ID=88171607
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310681071.1A Pending CN116842152A (en) | 2023-06-09 | 2023-06-09 | Science and technology policy question-answering method and device for fine-tuning language big model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116842152A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235243A (en) * | 2023-11-16 | 2023-12-15 | 青岛民航凯亚系统集成有限公司 | Training optimization method for large language model of civil airport and comprehensive service platform |
| CN117633179A (en) * | 2023-11-30 | 2024-03-01 | 武汉铁路职业技术学院 | Intelligent question-answering assistant system for railway industry |
| CN117744633A (en) * | 2024-02-21 | 2024-03-22 | 中国铁道科学研究院集团有限公司电子计算技术研究所 | Text error correction method and device and electronic equipment |
| CN117807962A (en) * | 2024-03-01 | 2024-04-02 | 之江实验室 | Method and device for writing consultation administrative texts, storage medium and electronic equipment |
-
2023
- 2023-06-09 CN CN202310681071.1A patent/CN116842152A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235243A (en) * | 2023-11-16 | 2023-12-15 | 青岛民航凯亚系统集成有限公司 | Training optimization method for large language model of civil airport and comprehensive service platform |
| CN117633179A (en) * | 2023-11-30 | 2024-03-01 | 武汉铁路职业技术学院 | Intelligent question-answering assistant system for railway industry |
| CN117633179B (en) * | 2023-11-30 | 2024-06-21 | 武汉铁路职业技术学院 | Intelligent question-answering assistant system for railway industry |
| CN117744633A (en) * | 2024-02-21 | 2024-03-22 | 中国铁道科学研究院集团有限公司电子计算技术研究所 | Text error correction method and device and electronic equipment |
| CN117807962A (en) * | 2024-03-01 | 2024-04-02 | 之江实验室 | Method and device for writing consultation administrative texts, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116842152A (en) | Science and technology policy question-answering method and device for fine-tuning language big model | |
| CN116701431A (en) | Data retrieval method and system based on large language model | |
| CN109800879B (en) | Knowledge base construction method and device | |
| CN117076693A (en) | Method for constructing digital human teacher multi-mode large language model pre-training discipline corpus | |
| US20240281659A1 (en) | Augmenting machine learning language models using search engine results | |
| CN117609444B (en) | Searching question-answering method based on large model | |
| CN111881264B (en) | Method and electronic equipment for searching long text in question-answering task in open field | |
| CN111552773A (en) | Method and system for searching key sentence of question or not in reading and understanding task | |
| CN117668181A (en) | Information processing method, device, terminal equipment and storage medium | |
| CN113590811A (en) | Text abstract generation method and device, electronic equipment and storage medium | |
| CN106776782B (en) | Semantic similarity obtaining method and device based on artificial intelligence | |
| CN117708282A (en) | Knowledge question-answering method and system based on large language model | |
| CN110674283B (en) | Intelligent extraction method and device for text abstracts, computer equipment and storage medium | |
| CN116680387A (en) | Dialogue reply method, device, equipment and storage medium based on retrieval enhancement | |
| CN111126066B (en) | Method and device for determining Chinese congratulation technique based on neural network | |
| CN114756617A (en) | Method, system, equipment and storage medium for extracting structured data of engineering archives | |
| CN114595684A (en) | Abstract generation method and device, electronic equipment and storage medium | |
| CN109388786B (en) | Document similarity calculation method, device, equipment and medium | |
| CN112036183A (en) | Word segmentation method and device based on BilSTM network model and CRF model, computer device and computer storage medium | |
| CN118396123B (en) | Text generation method, apparatus, computer program product, electronic device, and medium | |
| CN117033733B (en) | Intelligent automatic classification and label generation system and method for library resources | |
| CN114896362A (en) | Method and device for analyzing similarity of patent texts, storage medium and equipment | |
| CN118427329A (en) | Method and system for processing public accumulation policy question-answer data | |
| CN118503376A (en) | Data processing method and device for multi-hop problem | |
| CN117851574A (en) | Text dialogue method and device and electronic equipment |
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 |