A Knowledge-Enhanced Multi-Task Learning Model for Domain-Specific Question Answering

This paper presents GLM-6B-QA, a question-answering system designed to improve the performance of tasks in specialized domains. GLM-6B-QA uses the ChatGLM-6B model and integrates dynamic document attention (DDAM), query refinement layer (QRL), and knowledge injection layer (KIL) within a multi-task learning (MTL) framework. This setup enhances the model's ability to understand and process complex documents, queries, and terminology. By adjusting attention dynamically, refining query representations, and incorporating external knowledge, the system improves question answering, document comprehension, and term extraction. This method addresses challenges in natural language processing tasks and provides a better approach for developing adaptable systems in specialized applications.