**GraphTranslator: Aligning Graph Model to Large Language Model for Open-ended Tasks** **Authors:** Mengmei Zhang, Mingwei Sun, Peng Wang, Shen Fan, Yanhu Mo, Xiaoxiao Xu, Hong Liu, Cheng Yang, and Chuan Shi **Abstract:** Large language models (LLMs) like ChatGPT have revolutionized various fields, especially for open-ended tasks. However, graph models (GMs) are limited to predefined tasks and struggle with open-ended tasks. To address this, the authors propose GraphTranslator, a framework that aligns GMs with LLMs to handle both predefined and open-ended tasks. GraphTranslator introduces a Translator module to bridge the modality gap between node embeddings and textual tokens, and a Producer module to generate alignment data. The framework is trained on real-world datasets, demonstrating effectiveness in zero-shot node classification and graph question answering. The code is available at: https://github.com/alibaba/GraphTranslator. **Key Contributions:** - A novel model, GraphTranslator, that aligns GMs with LLMs to handle both predefined and open-ended tasks. - A Translator module to convert node embeddings into token representations, bridging the modality gap. - A Producer module to generate alignment data through LLMs, seamlessly textualizing node embeddings. **Experiments:** - **Zero-shot Node Classification:** GraphTranslator outperforms baselines, achieving better performance on Taobao and ArXiv datasets. - **Graph Question Answering:** GraphTranslator shows strong performance in extracting, explaining, and reasoning graph information, with high legality rates and recall scores. **Conclusion:** GraphTranslator effectively aligns GMs with LLMs, leveraging LLMs' extended interface to handle open-ended tasks. The framework's effectiveness is demonstrated through zero-shot node classification and graph question answering experiments, highlighting its potential and commercial value.**GraphTranslator: Aligning Graph Model to Large Language Model for Open-ended Tasks** **Authors:** Mengmei Zhang, Mingwei Sun, Peng Wang, Shen Fan, Yanhu Mo, Xiaoxiao Xu, Hong Liu, Cheng Yang, and Chuan Shi **Abstract:** Large language models (LLMs) like ChatGPT have revolutionized various fields, especially for open-ended tasks. However, graph models (GMs) are limited to predefined tasks and struggle with open-ended tasks. To address this, the authors propose GraphTranslator, a framework that aligns GMs with LLMs to handle both predefined and open-ended tasks. GraphTranslator introduces a Translator module to bridge the modality gap between node embeddings and textual tokens, and a Producer module to generate alignment data. The framework is trained on real-world datasets, demonstrating effectiveness in zero-shot node classification and graph question answering. The code is available at: https://github.com/alibaba/GraphTranslator. **Key Contributions:** - A novel model, GraphTranslator, that aligns GMs with LLMs to handle both predefined and open-ended tasks. - A Translator module to convert node embeddings into token representations, bridging the modality gap. - A Producer module to generate alignment data through LLMs, seamlessly textualizing node embeddings. **Experiments:** - **Zero-shot Node Classification:** GraphTranslator outperforms baselines, achieving better performance on Taobao and ArXiv datasets. - **Graph Question Answering:** GraphTranslator shows strong performance in extracting, explaining, and reasoning graph information, with high legality rates and recall scores. **Conclusion:** GraphTranslator effectively aligns GMs with LLMs, leveraging LLMs' extended interface to handle open-ended tasks. The framework's effectiveness is demonstrated through zero-shot node classification and graph question answering experiments, highlighting its potential and commercial value.