Research on Domain Agent Based on Event Element Extraction

The rapid development of artificial intelligence and large language models has sparked a research boom in the construction of Agents across various fields. It can assist in accomplishing various complex tasks and offer greater convenience for human beings. However, it is limited in handling high-dimensional or implicit relationships and has difficulty capturing subtle patterns and dynamic changes in complex relationships. Therefore, this paper proposes SmartExtract, a domain-specific Agent based on event element extraction, which combines vector retrieval, prompt learning, and large language model fine-tuning techniques to perform in-depth mining of key elements in domain event information, improving the performance of the Agent in domain event element extraction tasks. In addition, SmartExtract incorporates efficient feature extraction and similarity search to provide rich background knowledge and contextual information for constructing intelligent model prompts, guiding the model to generate results that meet task requirements. In this paper, comparative experiments are performed on DSN-DEMAMD, and the results show that SmartExtract outperforms traditional methods, demonstrating the effectiveness of the proposed approach.