DSEN: A Distance-Based Semantic Enhancement Neural Network for Pronoun Anaphora Resolution

Anaphora resolution is one of the fundamental tasks in natural language processing, aiming to identify the specific entities referred to by pronouns or noun phrases. In recent years, with the advancement of deep learning, the performance of anaphora resolution has significantly improved. To accurately identify the entities referred to by pronouns based on context, various factors need to be considered, such as the semantic dependencies in the context, the positions of entity words and pronouns, among others. However, recent research has lacked consideration of the impact of word distances on the correlation between word pairs. In this paper, we propose a anaphora resolution model called DSEN, which stands for Distance-based Semantic Enhancement Neural Network for Pronoun Anaphora Resolution. Our model integrates explicit positional relationships between entity words and pronouns, key meanings, and semantic information. By considering the influence of word distances, our model effectively incorporates the dependencies between word pairs. The fusion of these factors empowers our model to achieve outstanding results on the CLUEWSC2020 anaphora resolution dataset.