In recent years, the exponential growth of space weather data has necessitated the development of more efficient retrieval technologies to enhance data accessibility. The Solar-Terrestrial system, the focal point of space weather researches, is a paradigm of complexity, often requiring a comprehensive analysis of interdisciplinary and multi-modal data. Traditional keyword-based retrieval methods frequently fail to capture the nuanced semantic relationships within space weather data, hindering scientists’ ability to obtain precise information and integrate data from various sources. To address these challenges, this study constructs an optimized retrieval and recommendation system for space science data grounded in metadata semantics implementing three key techniques. Based on Metadata for Space Science Data, this paper constructs a conceptual model of space weather data semantics network, implements a hybrid retrieval system that combines keywords matching with word embedding vectors similarity searches and designs an algorithm to compute the association similarity between datasets, constructing a dataset interlinking matrix for the recommendation of related datasets. Extensive experiments with multi-disciplinary datasets demonstrate that the retrieval optimization strategy proposed in this study substantially improves the accuracy and comprehensiveness of space weather data retrieval. This advancement provides robust data support for the in-depth analysis of space weather event mechanisms, holding significant academic value and practical application potential.