A Multi-Head Collaborative Matrix Joint Extraction Method for the Chinese E-Government Domain

In Chinese government texts, the efficient extraction of entities and relationships is crucial for tasks such as information retrieval and intelligent question answering in e-government systems. Most current entity-relationship extraction methods often overlook the importance of entity type information in relation extraction, making it difficult to effectively capture complex multi-type entities and relationships, particularly in the government domain where the semantics are complex and diverse. To address this issue, this paper proposes a multi-head collaborative matrix joint extraction method, tailored to the characteristics of Chinese government texts. The approach leverages the Mengzi pre-trained model, which is more suited for Chinese entity-relationship extraction tasks, to dynamically encode word vectors from government texts, effectively solving the problem of insufficient information mining. Additionally, a multi-head collaborative matrix labeling scheme is introduced, allowing the model to incorporate entity type information from the text as an influencing factor in relation extraction. This enhances the model's ability to capture entity relationships and addresses issues of complex overlapping relationships and word ambiguity in Chinese texts. Experimental results show that this method significantly outperforms other benchmark models in terms of accuracy, recall, and F1 score on Chinese government datasets, providing more comprehensive and enriched information, and demonstrating strong advantages.