An Effective Framework for Question Answering over Freebase via Reconstructing Natural Sequences

Question answering over knowledge bases has rapidly developed with the continuous expansion of resources. However, how to match the natural language question to the structured answer entities in the knowledge bases remains a major challenge. In this paper, we propose an effective framework that bridges the gap between the given question and the answer entities, by reconstructing the intermediate natural sequences on the basis of the entities and relations in knowledge bases. The intuitive idea is that these intermediate sequences may encode rich semantic information that can identify the candidate answer entities. Experimental evaluation is conducted on a benchmark dataset WebQuestions. Results demonstrate the effectiveness of our proposed framework, i.e., it outperforms state-of-the-art models by up to 6.8% in terms of F1 score.