Awesome AI Agents › Programming Languages

HKUDS/GraphAgent

⭐ 371 Jupyter Notebook added to this list on 2025-03-09 repository created 2024-12-18

GraphAgent is an advanced agentic graph language assistant designed to handle complex real-world data that is represented in both structured and unstructured formats. It addresses the challenges of understanding and utilizing explicit graph dependencies, such as social connections and user behaviors, alongside implicit semantic interdependencies often depicted through knowledge graphs. The system is composed of three main components: a Graph Generator Agent that constructs knowledge graphs to capture intricate semantic relationships; a Task Planning Agent that interprets diverse user queries and autonomously plans tasks; and a Task Execution Agent that efficiently carries out these tasks by automating tool selection and invocation. These components work in harmony, integrating language models with graph language models to reveal deep relational information and semantic dependencies within data. GraphAgent supports a variety of predictive tasks like node classification and generative tasks such as text generation, demonstrating versatility across different data scenarios. The project provides pre-trained models, including a graph action model and a multimodal graph-text tokenizer, available on Hugging Face, facilitating easy deployment and experimentation. It also offers inference code, training and evaluation datasets, and plans to release training code for users to train GraphAgent on their own data. The repository includes detailed instructions for installation, model usage, and inference examples, making it accessible for researchers and developers interested in graph-based AI applications. Extensive benchmarking results show that GraphAgent outperforms many existing models on various graph-related tasks, highlighting its effectiveness and robustness. The project also features a comprehensive dataset collection covering multiple domains such as IMDB, ACM, Arxiv papers, and more, supporting both predictive and generative tasks. Overall, GraphAgent represents a significant advancement in integrating graph structures with language models to enhance understanding and generation capabilities in complex data environments, making it a valuable tool for AI research and applications involving graph data.

https://github.com/HKUDS/GraphAgent

agentai-researchbenchmarkingdatasetsgenerative-tasksgraph-datagraph-generatorgraph-language-assistantgraph-language-modelsgraphagenthugging-faceinference-codeknowledge-graphslanguage-assistantlanguage-modelslarge-language-modelsnode-classificationpre-trained-modelspredictive-taskstask-executiontask-planningtext-generation