A Text Matching Model Based on Keyword Attention Mechanism for Electronic Information Domain

In the domain of electronic information, the process of digital construction often necessitates the classification of data from diverse source tables across different factories into standardized tables. However, due to potential variations in naming conventions for identical types of data among various factories, automating accurate classification of data from source tables into standard tables has become an urgent problem that needs to be addressed. Given that both the source table and standard table fields should possess concise descriptions, this issue can be framed as a text matching problem between the descriptions of source table fields and standard table fields. To address this challenge effectively, we propose a text semantic matching model based on a keyword attention mechanism. Our model initially extracts keywords from the texts to be matched using CKPE and then incorporates these keywords along with the source texts into a text matching module for comparison purposes. The text matching module integrates a keyword attention mechanism layer onto BERT, which aims to emphasize the significance of keywords within the texts being matched. We conduct experimental analysis on two publicly available datasets and demonstrate that our model outperforms three commonly used text matching models. Furthermore, an ablation experiment confirms the effectiveness of incorporating an attention mechanism layer specifically designed for keywords within our proposed model.