Abstract: Schema mapping is one of the primary methods for achieving standardized data management and ensuring data quality and compliance. Existing field matching methods for schema mapping struggle to address the challenges in professional contexts, such as a large number of technical terms, short text lengths and difficulties in semantic understanding. This paper proposes a short text similarity computation method(SSCM) based on pre-trained language models, which can effectively improve the accuracy of corresponding field identification and matching in schema mapping and avoid field matching errors caused by character-level deviations in professional scenarios. Experimental results on the BankFieldSim dataset demonstrate that this method outperforms baseline models and can meet the high accuracy requirements for schema mapping in scenarios such as data governance and data integration.