Enhancing Knowledge graph with Selectional Preferences

Abstract

Abstract Knowledge graphs facilitate the extraction of knowledge from data and provide a comprehensive overview of all knowledge within departments, across departments, and global organizations. To enrich the extracted knowledge, several solutions have been proposed to complete the knowledge graph. This study investigates the effectiveness of using the selectional preferences technique to establish the dependency between triple entities in a knowledge graph. To this end, this paper proposes a three-phase approach, Selectional Preferences Knowledge Graph (SP-KG) to determine the relevance degree of all triple entities in the knowledge graph based on selectional preferences. The results demonstrate that the three-phase approach accurately identifies entity dependencies, which can be applied for knowledge extraction. Furthermore, this approach uses a Persian knowledge graph, which enhances the completeness of Persian language knowledge. Finally, the SP-KG model is evaluated on the SP-10K dataset proposed in state-of-the-art research to prove its accuracy. Similar results for both datasets indicate good performance. Glove and Word2Vec are also used to compare the proposed model with other methods. The accuracy results of the 2-way and 3-way pseudo-disambiguation demonstrate the high quality of the results produced by the proposed model.