Context-Aware Entity Summarization

Entity summarization aims at selecting a small subset of attribute-value pairs of an entity from a knowledge graph, which provides users with concrete information given an entity-related query. However, previous approaches focus on the "goodness" of the attribute-value pairs, paying little attention to user preference towards them. In this paper, we formalize the task of context-aware entity summarization, and propose an algorithm to solve this problem. We model user interest by mining the latent topics in a query log dataset. A modified Personalized PageRank algorithm is utilized to rank attribute-value pairs by leveraging three elements: relevance, informativeness and topic coherence. We evaluate our approach on real-world datasets and show that it outperforms the state-of-the-art approaches.