The recommendation of system capability indicators can provide a basis for combat effectiveness evaluation and improve the efficiency of indicator data collection, but the existing traditional methods are too subjective and inefficient. The article proposes an intelligent recommendation method for system capability indicators based on semantic understanding technology: firstly, crawling open-source weakly related semantic matching training sets, publicly available military articles and other relevant textual data, applying large language models to construct model training datasets suitable for the military domain; secondly, establishing a Chinese semantic matching model based on Sentence-BERT to achieve similarity scoring and ranking of input indicators and other texts; finally, designing simulation experiments to verify the feasibility and accuracy of this method, which can provide reliable support and reference for relevant decision-making.