PROJECT TITLE :

Modeling and Learning Distributed Word Representation with Metadata for Question Retrieval - 2017

ABSTRACT:

Community question answering (cQA) has become an important issue because of the recognition of cQA archives on the Web. This paper focuses on addressing the lexical gap drawback in question retrieval. Question retrieval in cQA archives aims to seek out the present queries that are semantically equivalent or relevant to the queried queries. However, the lexical gap downside brings a new challenge for question retrieval in cQA. During this paper, we propose to model and learn distributed word representations with metadata of class data within cQA pages for question retrieval using two novel category powered models. One could be a basic category powered model known as MB-NET and the other one is an enhanced class powered model called ME-NET that will higher learn the distributed word representations and alleviate the lexical gap downside. To accommodate the variable size of word representation vectors, we have a tendency to use the framework of fisher kernel to rework them into the fastened-length vectors. Experimental results on massive-scale English and Chinese cQA data sets show that our proposed approaches will considerably outperform state-of-the-art retrieval models for question retrieval in cQA. Moreover, we tend to any conduct our approaches on massive-scale automatic analysis experiments. The analysis results show that promising and significant performance improvements can be achieved.