Research on Joint Entity-Relation Extraction for the Tibetan “Epic of King Gesar” Based on Multi-Semantic Information Fusion and Global Pointer Network

The “Epic of King Gesar” is a poetic and historical Tibetan literary work and a treasure of Tibetan culture. Guided by the cultural preservation and inheritance of the Tibetan people in China, the paper deeply studies the joint entity-relation extraction method for the “Epic of King Gesar.” Addressing issues such as redundant entities, error accumulation from traditional pipeline methods, and insufficient semantic feature capture in Tibetan, the paper proposes an entity-relation extraction method based on multi-semantic information fusion and a global pointer network. (1) CINO was used to obtain dynamic word vectors for the entire text; (2) Global and local features of the Tibetan text were extracted using BiLSTM and CNN; (3) A multi-head attention mechanism was employed to fuse the text's multiple features; (4) The OneRel decoder was used to extract entity relations. Experiments show that this method improves the F1 score by 6.80% compared to the baseline OneRel model, providing a reference for research in specific domains of minority languages in low-resource settings.