Effective reuse of existing crowdsourced intelligence present in Community Question Answering (CQA) forums requires efficient approaches for the problem of Duplicate Question Detection (DQD). Approaches which use standalone encoded representations for each of the questions in the question pair fail to use the cross question interactions between the two questions which impacts their performance adversely. In this paper, we propose two new schemes for DQD task. Our first approach leverages semantic relations and our second approach utilizes fine grained word level interactions across the two question sentences. We achieve test accuracy of 75.7% and 77.8% with our first and second approaches respectively on a publicly available DQD data set, demonstrating that cross question analysis information can help aid DQD task performance.
CITATION STYLE
Mannarswamy, S., & Chidambaram, S. (2018). GEMINIO: Finding duplicates in a question haystack. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10938 LNAI, pp. 104–114). Springer Verlag. https://doi.org/10.1007/978-3-319-93037-4_9
Mendeley helps you to discover research relevant for your work.