Knowledge Graph Systems in the Era of Large Models

Background of the Special Issue

In 2012, Google first proposed the knowledge graph. Over the past decade, knowledge graph technology has made significant progress, widely applied in internet scenarios such as search, Q&A, and recommendation, as well as in various vertical fields like healthcare, e-commerce, and finance, thus being regarded as one of the core driving forces for the development of artificial intelligence. Meanwhile, since the emergence of ChatGPT in 2022, which has demonstrated excellent performance in language understanding and knowledge Q&A, large models have attracted widespread attention in academia and industry, being considered capable of memorizing and applying world knowledge. Knowledge graphs and large models are complementary means of representing and processing knowledge. On one hand, knowledge graphs enhance the language understanding capabilities of large models; on the other hand, they enrich the knowledge representation methods of knowledge graphs. The integrated development of both can effectively promote the intelligentization process of information systems across various industries and fields.

Against this backdrop, this special issue invites research teams from academia and industry to share knowledge graph and large model collaborative design patterns, large model enhanced reasoning and open-domain Q&A technologies, new paradigms for knowledge graph applications, and intelligent information systems based on large models. It aims to introduce the latest research progress and technical solutions regarding the design patterns, architecture, technology, and applications of knowledge graph systems in the era of large models, providing references for researchers and developers of intelligent information system applications.

Article Overview

This special issue includes six articles centered on the theme of knowledge graph systems in the era of large models, sharing typical cases, discussing the latest technological advancements, and exploring future development directions.

The article titled “Exploration of Collaborative Models between Knowledge Graphs and Large Language Models” written by Professor Wang Xin from Tianjin University discusses the collaborative model between knowledge graphs and large language models, highlighting their important positions in the field of artificial intelligence. It introduces knowledge graph-enhanced large language models, large language model-enhanced knowledge graphs, and their interactive integration with specific cases, aiming to achieve more powerful and flexible intelligent system applications. The article discusses future development directions, indicating that the collaborative model between knowledge graphs and large language models is expected to become a breakthrough point for “neural + symbolic” artificial intelligence, paving new methods towards general artificial intelligence research and becoming a new model for the implementation of artificial intelligence in various fields.

The article titled “Language and Knowledge Reasoning Enhanced by Large Models” written by Professor Chen Huajun from Zhejiang University introduces that knowledge graphs and large language models are two important means of representing and processing knowledge, capable of completing different types of reasoning tasks such as common sense reasoning, multimodal reasoning, concept reasoning, and logical rule reasoning. The authors discuss language and knowledge reasoning enhanced by large models from the perspective of the integration of language models and knowledge graphs, covering the reasoning capabilities inherent to large language models and the reasoning enhanced by large language models for knowledge graphs, and finally discuss the challenges faced by large model language reasoning and large model knowledge reasoning.

The article titled “Research Progress on Open-Domain Knowledge Graph Q&A” written by Professor Qi Guilin from Southeast University first summarizes the research progress of open-domain knowledge graph Q&A models, including semantic parsing-based, retrieval ranking-based, and pre-trained model-based knowledge graph Q&A methods, and then explores the application of large models in knowledge graph Q&A. Finally, it discusses the development and challenges in the field of knowledge graph Q&A under the era of large models, including addressing more complex and diverse questions, answering questions based on multi-source heterogeneous knowledge, and improving the efficiency of large model knowledge-enhanced Q&A.

The article titled “Intelligent Systems Based on Large Models: Architecture and Applications” written by Wang Wenguang, Vice President of Daguan Data, introduces the cutting-edge developments in intelligent systems and large models, abstracting the basic components of intelligent systems. The article provides a detailed introduction to the technologies and functions of each component in the architecture of intelligent systems based on large models, focusing on their feasibility, controllability, and universality. It also discusses how large models and knowledge graphs can collaborate to achieve controllable and universal intelligent systems, outlines the applications of intelligent systems based on large models, and summarizes the challenges faced by intelligent systems and future research directions.

The article titled “A New Paradigm for Knowledge Graph Applications in the Era of Large Models” written by senior technical expert Chen Jiao from Hang Seng Electronics Co., Ltd. compares the interrelationship between large models and knowledge graphs in industrial application scenarios, analyzing their advantages and disadvantages and proposing strategies for their complementary integration. It introduces typical application cases that combine large models and knowledge graphs, focusing on the high degree of overlap in knowledge Q&A, contextual understanding learning capabilities, and logical reasoning capabilities. Based on a thorough analysis of their complementary capabilities and shortcomings, the author proposes a new collaborative application paradigm for the synergistic application of large model and knowledge graph technologies, along with application landing cases based on the SPG framework, emphasizing the importance and necessity of jointly building an ecosystem for large models and knowledge graphs.

The article titled “Enterprise-Level Intelligent Systems Based on Enhanced Large Language Models” written by Hu Fanghuai, Chief Technology Officer of Haiyi Knowledge Information Technology Co., Ltd. (PlantData), introduces technologies related to enhancing large language models to mitigate or reduce defects such as hallucinations, biases, and outdated knowledge. Key technologies for enhancing large language models include advanced prompt engineering, retrieval knowledge enhancement, tool learning, and central control collaboration. Based on this, the article proposes an overall architecture for intelligent applications that combine large language models and knowledge graphs, describing the mutual enhancement and collaborative realization of applications in centralized control. Finally, the article looks forward to new development directions for large language models, including multimodal and embodied extension as well as intelligent agent systems.

Outlook

The six articles in this special issue conduct a comprehensive analysis and reflection on the design patterns, architecture, technology, and applications of knowledge graph systems in the era of large models, introducing the latest dynamics from academia and industry regarding the deep integration of knowledge graphs and large models for intelligent systems, while also highlighting the challenges and development directions ahead. The technical research and practical applications of knowledge graphs and large models are currently thriving, and it is hoped that this special issue can inspire researchers in related fields in theory, technology, and practice, attracting more experts, scholars, and engineers to pay attention and jointly promote the development of knowledge graph and large model integration research and system applications.

Wang Xin

Distinguished Member of CCF, Secretary-General of CCF Information Systems Committee, Executive Member of CCF Database Committee, Executive Member of CCF Big Data Expert Committee. Professor at Tianjin University. Main research directions include knowledge graphs and graph databases. [email protected]

Wang Haofen

Senior Member of CCF, Secretary-General of CCF Shanghai Branch, Chairman of CCF SIGKG, Deputy Director of Terminology Working Committee. Distinguished Researcher at Tongji University. Main research directions include knowledge graphs and natural language processing. [email protected]