Question-Answering Aspect Classification with Multi-attention Representation

In e-commerce platforms, the question-answering style reviews are emerging, which usually contains much aspect-related information about products. In this paper, Question-answering (QA) aspect classification is a new task that aims to identify the aspect category of a given QA text pair. According to characteristics of QA-style reviews, we draw up annotation guidelines and build a high-consistency annotated corpus for QA aspect classification. Then, we propose a recurrent neural network based on multi-attention representation to tackle this new task. Specifically, we firstly segment the answer text into clauses, and then leverage the multi-attention representation layer to match the question text with clauses inside answer text and generate multiple attention representations of the question text, which extends feature information of the question text. The experimental results demonstrate that our method for QA aspect classification, which is based on multi-attention representation, can make the most of useful information in answer texts and perform better than some strong baselines in QA aspect classification.