Abstract: Guided by theories and technologies related to knowledge reorganization, this study conducts semantic mining and organization of allusion cultural resources to promote the inheritance and utilization of allusion culture. A model is proposed that integrates fine-grained co-reference relations and semantic features to link allusion terms. First, a co-reference network is constructed based on the citation relationships between ancient poems and allusion terms, and fine-grained co-reference relations, including positional co-reference and emotional co-reference, are added to build a fine-grained co-reference network. Then, Doc2vec is employed to extract the semantic features of each allusion term, and these features are integrated to reconstruct the co-reference network. Finally, a link prediction algorithm is applied to traverse the fine-grained co-reference network, achieving semantic association and organization of allusion terms. The association results are further analyzed from a path-based perspective, uncovering some regular patterns in domain knowledge. The constructed co-reference network consists of 5,869 nodes and 27,032 edges. The proposed method, which incorporates positional and emotional co-references as well as semantic features, achieves an accuracy of 0.963 in the task of linking allusion terms. Moreover, the analysis reveals that the shortest path order is negatively correlated with both the number of allusion term pairs and their similarity.

Key words: Knowledge reorganization, Allusion words, Co-citation network, Semantic features, Link prediction