A Framework for Knowledge Graph Based Question Answering
Oral Defence Date:
Tuesday, July 10, 2018 - 10:00
Asst. Prof. Anagha Kulkarni and Prof. James Wong
Rich and comprehensive knowledge graphs (KG) of the Web, such as, Google KG, NELL, and Diffbot KG, are becoming increasingly prevalent and powerful as the underlying AI technology is rapidly progressing. In this work, we leverage this ongoing advancement for the task of answering questions posed from any domain and any type (factoid and non-factoid). We present a framework for knowledge graph based question answering systems, KGQA, and experiment with two instances of this framework that employ Diffbot and Google KG. One of the unique advantages of using a KG (versus searching the Web) is its rapid response time to queries. We leverage this feature to design the first component of our system where various representations of the user question are used to craft multiple queries. Second, many KGs infer connections between different objects in the KG. We use this linked-data information to enrich the base data that is retrieved from the KG. The multi-query setup is also used to improve the selection of answer-bearing objects through data fusion techniques. We exploit the ability of KGs to provide a structure to the data, and to identify non-traditional sources of answer, such as, product details, and video captions. For the empirical analysis of our system, we experiment with multiple datasets containing factoid and non-factoid questions. The results demonstrate that the proposed approach outperforms several other approaches across multiple metrics.
Automatic Question Answering, Diffbot Knowledge Graph, Google Knowledge Graph, Information Retrieval, Natural Language Processing