In scientific literature retrieval with formulas as the core content, it is crucial to integrate the overall characteristics of formulas and textual information to improve retrieval effectiveness. However, existing methods often fail to fully capture the deep interconnections between formulas and textual information, leading to suboptimal retrieval results. To address this problem, we propose the MiniLM-Formula model, which facilitates scientific and technical literature retrieval by leveraging joint formula-text features. Firstly, we design and optimize joint formula-text features to generate high-quality training samples. Then, MiniLM is fine-tuned on the training dataset to effectively learn the deeply fused representations of formulas and text, thereby improving retrieval accuracy. In addition, the Analytic Hierarchy Process (AHP) is introduced to achieve personalized literature reordering and enhance the relevance of retrieval results and user satisfaction. Experiments on the ArXiv dataset demonstrate that our method achieves MAP@10 of 81.0% and NDCG@10 of 85.6%.