Cultural data large model construction method and cultural data question and answer platform

By constructing a multimodal fusion architecture and a method of fine-tuning training of sample sets for Q&A, the cultural data model is optimized, and the problem of inaccurate processing of multicultural background content in the existing technology is solved, and more accurate understanding and generation of cultural information is achieved. .

CN119179760APending Publication Date: 2024-12-24BEIJING JIUYANG BOWEN CULTURE DEV CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411054505.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-12-24

AI Technical Summary

Technical Problem

Existing large language models are difficult to achieve accuracy when dealing with content from multicultural contexts and specific contexts, resulting in bias in understanding and generating culturally relevant content, affecting the efficiency of cross-cultural communication and may cause misunderstanding and communication barriers.

Method used

By obtaining multimodal cultural data from different fields, building a multimodal fusion architecture, training NLP and CV models, and fine-tuning the sample set using Q&A to optimize the parameters of the cultural data big model to improve the understanding of multicultural background and generation ability.

Benefits of technology

It has achieved a deep integration of cultural knowledge, and the content output is more accurate when facing multicultural backgrounds, which has improved the efficiency and accuracy of cross-cultural communication.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119179760A_ABST
    Figure CN119179760A_ABST
Patent Text Reader

Abstract

The invention discloses a culture data large model construction method and a culture data question and answer platform, and the method comprises the steps: obtaining a large amount of multi-modal culture data in different fields, and carrying out the preprocessing of the culture data; the multi-modal culture data comprises text data, images and video data; constructing a multi-modal fusion architecture, training an NLP model by using the text data, and training a CV large model by using image and video data; associating output features of the NLP model and the CV large model by using a multi-modal large model to obtain a culture data large model; and obtaining question and answer pair sample sets in different fields, and performing fine tuning training on the culture data large model by using the question and answer pair sample sets in different fields so as to perform parameter optimization on the culture data large model. The large culture model has deeper fusion on culture knowledge, and the output content is more accurate in the face of multiple culture backgrounds.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of cultural data big models, and in particular to a method for constructing a cultural data big model. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models (LLMs) have become a key force in promoting scientific and technological progress and innovation. As a breakthrough technology in the field of artificial intelligence, its importance cannot be ignored. Large language models have powerful data processing and pattern recognition capabilities, enabling machines to handle unprecedented complex tasks, greatly expanding the scope of application of artificial intelligence.

[0003] In applications in the cultural field, although existing large language models can be used, existing large models often have difficulty achieving satisfactory accuracy when processing content with multicultural backgrounds and specific contexts. This is because the complexity and diversity of cultural information far exceeds the scope of the model's training data, causing the model to be prone to deviations when understanding and generating culturally relevant content. For example, the model may not be able to accurately identify and use expressions in a specific culture. These problems not only affect the efficiency of cross-cultural communication, but may also cause misunderstandings and communication barriers, highlighting the urgency and importance of further optimizing large models in the cultural field. Summary of the invention

[0004] Based on this, in response to the above technical problems, a method for building a cultural data big model and a cultural data question and answer platform are provided to solve the problem of inaccurate output when the existing technology uses big models to face content with multiple cultural backgrounds and specific contexts.

[0005] In a first aspect, a method for constructing a large cultural data model is provided, the method comprising:

[0006] Acquire a large amount of multimodal cultural data in different fields and preprocess the cultural data; the multimodal cultural data includes text data, image and video data;

[0007] Construct a multimodal fusion architecture, use the text data to train the NLP model, and use the image and video data to train the CV large model; use the multimodal large model to associate the output features of the NLP model and the CV large model to obtain a cultural data large model;

[0008] Sample sets of question-answer pairs in different fields are obtained, and the sample sets of question-answer pairs in different fields are used to fine-tune the large cultural data model, so as to optimize the parameters of the large cultural data model.

[0009] In the above scheme, optionally, the cultural data includes text data in multiple cultural fields, including classical literature, historical documents, folklore, art, and philosophy.

[0010] In the above scheme, optionally, the text data is in various forms, including industry reports, academic papers, books, and web page content.

[0011] In the above scheme, optionally, the preprocessing of the cultural data includes: cleaning the collected text to remove irrelevant information; and then performing word segmentation and removing stop words.

[0012] In the above solution, optionally, the fine-tuning training of the cultural data large model using question-answer pair sample sets from different fields includes:

[0013] The question in the question-answer pair is input into the cultural data model, and the answer is output. By comparing the similarity between the expected answer corresponding to the question in the question-answer pair and the output answer, the parameters of the cultural data model are fine-tuned.

[0014] In the above scheme, further optionally, the fine-tuning training of the cultural data big model using a sample set of question-answer pairs from different fields also includes: inputting questions in the question-answer pairs into the cultural big model, outputting answers, and fine-tuning the parameters of the cultural data big model based on the scores of the output answers by experts.

[0015] In the above scheme, optionally, after preprocessing the cultural data, the method further includes: marking the cultural data, and the marked content includes: text, appendix, notes, index, and footnotes.

[0016] In a second aspect, a cultural data question and answer platform is provided, wherein the question and answer platform is constructed based on the cultural data big model described in the first aspect.

[0017] In a third aspect, a computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a method for constructing a large cultural data model described in the first aspect are implemented.

[0018] In a fourth aspect, a computer program product includes a computer program / instructions, which, when executed by a processor, implement the steps of a method for constructing a large cultural data model as described in the first aspect.

[0019] This application has at least the following beneficial effects:

[0020] This application acquires a large amount of multimodal cultural data from different fields, constructs a multimodal fusion architecture, uses text data to train the NLP model, uses image and video data to train the CV model, uses the multimodal model to associate the output features of the NLP model and the CV model, and obtains the cultural data model. At the same time, it obtains question-answer pair sample sets from different fields, trains the cultural data model, and optimizes the parameters of the cultural model. In this way, the cultural model of this application has a deeper integration of cultural knowledge, and the output content is more accurate when facing a multicultural background. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a method for constructing a large cultural data model is provided for an embodiment of the present application;

[0022] Figure 2 This is an architectural diagram of a large cultural data model for an embodiment of the present application;

[0023] Figure 3 This is a specific flow chart of a method for constructing a large cultural data model in one embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0025] In one embodiment, Figure 1 As shown, a method for constructing a large cultural data model is provided, characterized in that the method comprises:

[0026] Step S1: Acquire a large amount of multimodal cultural data in different fields and preprocess the cultural data; the multimodal cultural data includes text data, image and video data.

[0027] In step S1, a large amount of texts of different cultural fields, styles, genres and periods are collected as data sources, and the collected cultural data of multiple cultural categories are preprocessed to obtain a high-quality multi-corpus, wherein the cultural data includes multiple fields, including but not limited to classical literature, historical documents, folklore, art, and philosophy. The cultural data can be obtained from industry reports, academic papers, books, and web page content.

[0028] Cultural data includes the following:

[0029] (1) Chinese cultural heritage specimen data, which is based on the data from previous national cultural resource censuses, including the data already obtained from censuses such as ancient books, cultural relics, art museum collections, local operas, and intangible cultural heritage.

[0030] (2) Chinese national cultural gene data mainly comes from the large database of revolutionary cultural relics and the big data system of museums. These data are collected with high precision, imported into the underlying associated integration system of the national cultural big data system, and data annotation, association are completed, and Chinese national cultural elements, symbols and identifiers are extracted.

[0031] (3) Chinese cultural material data is a material library that can be combined and used, which is formed by taking the digitalization achievements of cultural resources as raw materials, integrating and applying various new technologies, extracting the elements of Chinese culture, and classifying and tagging them. According to the cultural presentation elements, the Chinese cultural material library can be divided into the Chinese Character Library, the Chinese Sound Library, the Chinese Image Library, the Chinese Music Library, the Chinese Dance Library, the Chinese Drama Library, etc.

[0032] Clean the collected cultural data to remove the noise in the data, such as irrelevant characters and special symbols; then perform word segmentation to split the text into processable units, such as words, phrases or sentences; and remove stop words: delete common words that are not very helpful for model training, such as "de", "he", etc.; and perform standardization: unify the text format, such as converting to lowercase, normalizing synonyms, etc.; word embedding: convert words into numerical vectors, and pre-trained word vector models such as Word2Vec, GloVe, etc. can be used; so that the large LLM can better learn the content of cultural data.

[0033] Step S2: Construct a multi-modal fusion architecture, train the NLP model with the said text data, and train the CV large model with image and video data; use the multi-modal large model to associate the output features of the NLP model and the CV large model to obtain a cultural data large model.

[0034] In step S2, according to the characteristics of cultural data, adjust the model architecture, such as increasing the attention to cultural-specific vocabulary and expressions. Adopt specific training strategies, such as Supervised Fine-Tuning, DPO, RHLF, etc., to improve the adaptability of the model in the cultural field.

[0035] During the training process, when constructing a large model for literary creation, a multi-modal fusion architecture of LLM based on Transformer can also be adopted to integrate various information sources such as text and images, so that the model can understand and combine different forms of artistic elements to inspire innovative inspiration.

[0036] Step S3: Obtain sample sets of question-answer pairs in different fields, and use the sample sets of question-answer pairs in different fields to fine-tune the large cultural data model, so as to optimize the parameters of the large cultural data model.

[0037] In step S3, a large number of question-answer pairs in different fields are obtained through experts, and question-answer pairs with different requirements can be obtained, such as cultural content creation, literary content interpretation and appreciation, cultural market analysis, etc.; the expected answers to the questions given by experts are evaluated separately on the multiple question answers output by the model to adjust the weights and parameters.

[0038] At the same time, during the use of the model, the model can also be continuously trained based on user feedback to establish a comprehensive user evaluation system that includes language accuracy, plot coherence, character creation, innovation, reader appeal, etc., so that users can comprehensively evaluate the output content and ensure that the model is comprehensively evaluated in multiple aspects of literary creation.

[0039] In the above-mentioned method for constructing a large cultural data model, a large amount of cultural data from different fields is obtained and input into the existing LLM large model for training. At the same time, a sample set of question and answer pairs from different fields is obtained to train the cultural data large model and optimize the parameters of the cultural large model. In this way, the cultural large model of this application has a deeper integration of cultural knowledge, and the output content is more accurate when facing a multi-cultural background.

[0040] In one embodiment, fine-tuning the cultural data big model using question-answer pair sample sets from different fields includes:

[0041] The question in the question-answer pair is input into the cultural data model, and the answer is output. By comparing the similarity between the expected answer corresponding to the question in the question-answer pair and the output answer, the parameters of the cultural data model are fine-tuned.

[0042] In one embodiment, the use of question-answer pair sample sets from different fields to fine-tune the cultural data big model also includes: inputting questions in the question-answer pairs into the cultural big model, outputting answers, and fine-tuning the parameters of the cultural data big model based on the scores of the output answers by experts.

[0043] In one embodiment, after preprocessing the cultural data, the method further includes: marking the cultural data, and the marking content includes: text, appendix, notes, index, and footnotes.

[0044] In this embodiment, the cultural data is annotated so that the cultural data can be better understood in context according to the location of the cultural data in the article.

[0045] In one embodiment, a cultural data question and answer platform is built based on the above-mentioned cultural data big model.

[0046] In this example, the API interface of the model service is designed, including input and output formats, request restrictions, etc. The back-end logic is developed, including API services, database interactions, model calls, etc. The user interface is designed to ensure ease of use and accessibility. The front-end logic is implemented, including user input, display results, and interactive feedback.

[0047] The cultural data platform constructed in this application can be used in the following aspects:

[0048] Cultural content creation: Provide creative inspiration and writing assistance to writers, screenwriters, etc., and automatically generate storylines or character descriptions based on specific cultural backgrounds.

[0049] It can generate various forms of educational content, including historical explanations, literary appreciation, art reviews, etc. By learning and understanding a large amount of cultural data, the model can generate educational materials with rich content and diverse perspectives, thus improving the quality of education.

[0050] Help cultural institutions manage and analyze historical documents and automatically generate descriptions and background information for artifacts.

[0051] Cultural market analysis: Provide market trend analysis for cultural enterprises and predict the popularity of cultural products.

[0052] The innovation of this application lies in:

[0053] Cultural Specificity: Models are custom trained for the specific needs of a cultural domain, improving understanding of culturally specific vocabulary and concepts.

[0054] Intelligent analysis depth: The model is able to conduct in-depth cultural content analysis, such as cultural element identification and metaphor understanding, to provide users with richer information.

[0055] Interaction-friendliness: Users can interact with the model through natural language, which improves the model's ease of use and user satisfaction.

[0056] Wide application range: The model is not only applicable to cultural research, but can also play an important role in education, entertainment, tourism and other industries.

[0057] By implementing the cultural big model, it is expected to improve the efficiency of cultural data analysis. Automated analysis tools will significantly reduce the time and cost of manual processing. Enhance the creativity of cultural content. Creators can use the model to quickly generate drafts or get inspiration, improving the efficiency of creation. Promote cultural inheritance and innovation. The model provides a new means for cultural research and education, and helps to protect and inherit cultural heritage.

[0058] In one embodiment, a computer device is provided, which may be a server. The computer device includes a processor, a memory and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for constructing a large model of cultural data is implemented.

[0059] In one embodiment, a computer program product is also provided, including a computer program / instruction, which involves all or part of the process in the above embodiment method.

[0060] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0061] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for constructing a large cultural data model, characterized in that: The method comprises: Acquire a large amount of multimodal cultural data in different fields and preprocess the cultural data; the multimodal cultural data includes text data, image and video data; Construct a multimodal fusion architecture, use the text data to train the NLP model, and use the image and video data to train the CV large model; use the multimodal large model to associate the output features of the NLP model and the CV large model to obtain a cultural data large model; Sample sets of question-answer pairs in different fields are obtained, and the sample sets of question-answer pairs in different fields are used to fine-tune the large cultural data model, so as to optimize the parameters of the large cultural data model.

2. The method for constructing a large cultural data model according to claim 1, characterized in that: The cultural data includes text data in multiple cultural fields, including classical literature, historical documents, folklore, art, and philosophy.

3. The method for constructing a large cultural data model according to claim 1, characterized in that: The text data are in various forms, including industry reports, academic papers, books, and web page content.

4. The method for constructing a large cultural data model according to claim 1, characterized in that: The preprocessing of the cultural data includes: cleaning the collected text to remove irrelevant information; and then performing word segmentation and removing stop words.

5. The method for constructing a large cultural data model according to claim 1, characterized in that: The fine-tuning training of the cultural data large model using question-answer pair sample sets from different fields includes: The question in the question-answer pair is input into the cultural data model, and the answer is output. By comparing the similarity between the expected answer corresponding to the question in the question-answer pair and the output answer, the parameters of the cultural data model are fine-tuned.

6. The method for constructing a large cultural data model according to claim 5, characterized in that: The method of fine-tuning the cultural data big model using a sample set of question-answer pairs from different fields also includes: inputting questions in the question-answer pairs into the cultural big model, outputting answers, and fine-tuning the parameters of the cultural data big model based on the scores of the output answers by experts.

7. The method for constructing a large cultural data model according to claim 1, characterized in that: After the cultural data is preprocessed, the method further includes: marking the cultural data, and the marking content includes: main text, appendix, notes, index, and footnotes.

8. A cultural data question and answer platform, characterized in that: The question-and-answer platform is built based on the cultural data big model described in any one of claims 1-7.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Multi-modal question and answer implementation method and device for heterogeneous historical and cultural heritage data

    CN121658595A