Today’s search engine provides a means to efficiently search huge and heterogeneous document collections. However, it is not easy to find useful information from a large number of candidates obtained as a result of the search, and this problem cannot be solved by merely devising the search engine document ranking. In this research, we focus on natural language processing methods useful for efficient interactive search. The research topics include NLP technologies such as context/topic-aware sentence compression, automatic question generation with a knowledge base, and argumentative text detection for decision making support.
Text Compression and Summarization
Text compression and summarization systems aim to produce a shorter version of a source text by preserving the key contents of the original. However, yielding an informative and grammatical compression (summary) is still a challenge. In this project, we tackle this issue by considering two aspects – the word (local) features such as part-of-speech tag of word and sentence (global) features such as readability of a whole sentence. Our experimental results demonstrate that these features coupled with techniques like deep learning and reinforcement learning can lead to compressions (summaries) with better quality (Yang et al: NLDB-2017 ; ACL-2018 short, accepted)