Textual and Visual Features Alignment and Attention Interaction Based Multimodal Entity Alignment Method for Electric Power Operation and Inspection

Due to the accumulation of multi-source information such as information ledgers and two invoices in the power industry, there may be situations where the same power business entities come from multiple sources of information during the construction of a multimodal knowledge graph. Therefore, multimodal entity alignment is needed to achieve the identification and linking of equivalent entities. For multimodal entity alignment, it is crucial to study how to integrate multiple modal information of entities to obtain accurate semantic representations of multimodal entities. For multimodal entity alignment, it is very important to study how to integrate multiple modal information of entities to obtain accurate semantic representation of multimodal entities. Therefore, in order to reduce cross-modal semantic differences and extract key alignment information, this paper considers the semantic consistency and fine-grained feature interaction fusion of textual and visual modalities, and proposes a multimodal entity alignment method for electric power operation and inspection based on textual and visual features alignment and attention interaction. This paper designs the textual and visual features extraction module, the cross-modal feature fusion module, and the multimodal entity feature matching module. At the same time, by introducing contrastive learning of textual and visual modalities, attention mechanisms, and correlation calculation of textual and visual features, more accurate semantic representations of entities are obtained, thereby achieving efficient and accurate multimodal entity alignment for electric power operation and inspection. The improvement in alignment performance of the proposed method was verified through experimental comparison with the baseline control group on the power dataset.