Why was this asked? Automatically recognizing multiple motivations behind community question-answering questions

Alexander Espina, Alejandro Figueroa

Abstract: Highlights • Automatic detection of multiple motivations behind cQA questions. • Five different multi-label algorithms were combined with 14 base classifiers. • Each combination was built on assorted linguistically-motivated features. • We discover some linguistic traits signalling the confluence of some motivations. Abstract Community Question Answering (cQA) services encourage their members to ask any kind of question, which later on can get multiple answers from other community fellows. The research objective of this paper is understanding, and particularly automatically detecting, what motivates community members to ask questions to their unknown peers. In so doing, we first crawled a set of cQA questions from Yahoo! Answers, each of which was manually labelled according to their multiple motivations afterwards. Thus, one of the innovative aspects of our work is exploring a wide variety of multi-label classification strategies for the automatic recognition of concurrent motivations behind cQA questions. In order to build effective models, high-dimensional feature spaces were constructed on top of assorted linguistic features, this way discovering some linguistic traits that characterize some of these combinations. Overall, our experiments reveal that multi-label classification frameworks hold a real promise for this task. More precisely, our best configuration finished with a Hamming Score of 0.71. In terms of features, our outcomes unveil that the concurrence of motivations is likely to be signalled by the complexity in writing and the distribution of entity mentions across the entire question, ergo both question titles and bodies are required to be able to recognize their confluence.