Against the backdrop of the continuous evolution of research paradigms driven by artificial intelligence for Science(AI4S), research practice increasingly relies on multi-source data and interpretable evidence chains. In mixed-methods research, qualitative materials and quantitative data can hardly be coherently integrated within the same logical chain, which tends to results in disconnection between statistical results and situational experiences, leaving evidence integration at a formal level. To address this issue, this paper proposed a qualitative-quantitative integration framework consisting of an upward pathway, a downward pathway, and a quality-control mechanism by leveraging the capabilities of large language models (LLMs) in semantic representation and situational understanding. The framework established interpretable connections between quantitative structures and qualitative materials through upward and downward analysis pathways, supplemented by a quality control mechanism to ensure the robustness of the reasoning process. To verify the feasibility and explanatory power of the framework, this paper tested it with real data from teacher collaboration processes. The results showed that the framework can facilitate mutual validation across data types within a unified representational space, providing a systematic technical path to break through the limitation of evidence juxtaposition in mixed-methods research and laying a foundation for the integrative innovation of social science research methods in the AI4S era..

At present, the innovative potential of large language models (LLM) in language writing instruction is becoming increasingly evident. Among them, creative writing, as an important means to cultivate students’ emotional expression, critical thinking, and innovative abilities, is facing practical challenges such as rigid teaching models, scarce teaching resources and single evaluation models, and the development of LLM has brought new opportunities to it. Therefore, this paper analyzed the applicability of LLM in creative writing instruction from the perspectives of world knowledge, emergent capabilities, theory of mind, chain of thought, and silicon-based samples, and proposed an LLM-based creative writing application framework aimed at integrating dynamic analysis and personalized feedback into the writing process through a combination of multimediainspired and interactive creativity stimulation. Based on this framework, a corresponding LLM4Writing platform was developed and implemented in a 5-week case study at S Elementary School in Shanghai. The case results showed that students have made significant improvements in creative conception and expression abilities, while their writing interest and creative potential have been effectively stimulated. Through research, this article was aimed to provide a theoretical foundation and practical guidance for the application of LLM in creative writing instruction, thereby promoting innovation of writing teaching models and the improvement of writing teaching quality.