Scholarlink

<p>Digital human, as a novel form of human–computer interaction, achieves real-time interaction with users through intelligent technologies such as computer vision, natural language processing, and network communication. In the context of knowledge question answering for the vertical fields of cultural heritage and tourism, traditional digital human question-answering algorithms lack the ability to effectively mine latent semantic information from background data, leading to imprecise responses to user queries. To address this, the study proposes a method called NGR, which enhances the question-answering capability of digital humans using automatically constructed knowledge graphs. Specifically, it first extracts entity-relation pairs using a pre-trained large language model to construct and preprocess the knowledge graph. Then, based on the characteristics of different entity-relationship pairs, it groups the graph nodes into communities and generates community summaries. Finally, the community summaries and background information are used as important reference points for generating answers, thereby improving the mining of latent semantic information. At the same time, the study combines NGR, the lip-sync synthesis video model MuseTalk, TTS, and streaming service technologies to develop a cultural heritage and tourism digital human, Graph-Human, capable of real-time question-answering interaction and video rendering. The study evaluates the performance of this digital human on two aspects: graph node extraction and question-response layers, using the CLUENER2020 and AttractionDetailsQA datasets. The evaluation results show that it outperforms other models. Meanwhile, the digital human was evaluated on the HDTF dataset for the perceptual quality of lip images. Experimental results demonstrate that the lip image quality is satisfactory and the video rendering performance is strong, enabling real-time rendering. This suggests its potential to provide knowledge-based question-answering services to the public in vertical domains such as cultural heritage and tourism.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Graph-human: an intelligent question–answering digital human model enhanced by automatically constructed knowledge graph for cultural and tourism applications

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Penglin Wang,
  • Donghui Shi,
  • Mengya Li,
  • Jose Aguilar

摘要

Digital human, as a novel form of human–computer interaction, achieves real-time interaction with users through intelligent technologies such as computer vision, natural language processing, and network communication. In the context of knowledge question answering for the vertical fields of cultural heritage and tourism, traditional digital human question-answering algorithms lack the ability to effectively mine latent semantic information from background data, leading to imprecise responses to user queries. To address this, the study proposes a method called NGR, which enhances the question-answering capability of digital humans using automatically constructed knowledge graphs. Specifically, it first extracts entity-relation pairs using a pre-trained large language model to construct and preprocess the knowledge graph. Then, based on the characteristics of different entity-relationship pairs, it groups the graph nodes into communities and generates community summaries. Finally, the community summaries and background information are used as important reference points for generating answers, thereby improving the mining of latent semantic information. At the same time, the study combines NGR, the lip-sync synthesis video model MuseTalk, TTS, and streaming service technologies to develop a cultural heritage and tourism digital human, Graph-Human, capable of real-time question-answering interaction and video rendering. The study evaluates the performance of this digital human on two aspects: graph node extraction and question-response layers, using the CLUENER2020 and AttractionDetailsQA datasets. The evaluation results show that it outperforms other models. Meanwhile, the digital human was evaluated on the HDTF dataset for the perceptual quality of lip images. Experimental results demonstrate that the lip image quality is satisfactory and the video rendering performance is strong, enabling real-time rendering. This suggests its potential to provide knowledge-based question-answering services to the public in vertical domains such as cultural heritage and tourism.