The efficient matching of academic journals with research funds is a key issue in improving the efficiency of research resource allocation, shifting to personalized recommendation methods. This paper proposes a personalized fund recommendation method that integrates multi-source decision analysis and reinforcement learning within a trilateral feature fusion model of textile engineering journals, fiber science scholars, and research funds. By optimizing recommendation strategies through deep reinforcement learning and implementing an explainability framework, the system enhances transparency in identifying high-impact funding opportunities specifically tailored to the evolving technological landscape of textile manufacturing and material innovation. The study establishes a matching degree evaluation model based on multi-source data fusion, uses the proximal policy optimization algorithm for dynamic recommendation strategy learning, and constructs a multi-level explainability system including attention visualization, decision path tracing, and comparison cases. Experimental results show that this method significantly outperforms traditional recommendation algorithms in terms of recommendation accuracy, novelty, and user satisfaction. In the Top-5 recommendation scenario, the hit rate reaches 0.427, which is 12.3% higher than the best benchmark method, and the user score for explainability exceeds 4.0. This study provides strong support for intelligent matching of academic resources, an innovative approach that cannot be implemented by traditional methods.

fund recommendations, multi-source data fusion, reinforcement learning, explainable artificial intelligence, textile engineering