Abstract— Community questionanswering system (cQA), one of the fastest-growing user-generated content (UGC)portals, has raised as an enormous market, so to speak, for the fulfilment ofcomplex information needs. cQA enables users to ask/answer questions and searchthrough the archived historical question-answer (QA) pairs. Propose systempresent a novel scheme for answer selection in cQA settings. It comprises of anoffline learning and an online search component. In the offline learningcomponent, instead of time-consuming and labor-intensive annotation, weautomatically construct the positive, neutral, and negative training samples inthe forms of preference pairs guided by our data-driven observations. We thenpropose a robust pairwise learning to rank model to incorporate these threetypes of training samples. In the online search component, for a givenquestion, we first collect a pool of answer candidates via finding its similarquestions. We then employ the offline learned model to rank the answercandidates via pair wise comparison. We have conducted extensive experiments tojustify the effectiveness of our model on one general cQA dataset and onevertical cQA dataset.
Keywords: AnswerSelection, Community Question Answering(cQA), Pairwise Comparison