Semantic Textual Similarity and Factorization Machine Model for Retrieval of Question-Answering

Abstract

Question and Answering (QA) in many collaborative social networks such as Yahoo!-answers, Stack Overflow have attracted copious users to post and transfer knowledge between users. This paper proposes an Adaptive global T-max Long Short-Term Memory-Convolutional Neural Network (ALSTM-CNN) method to retrieve semantically matching questions from historical questions and forecast the best answers by saving their effort and time. Moreover, a novel Field-aware Factorization Machine (FFM) classifier is adapted to rank the high-quality answers from large sparse data. This method has certain advantages include: (a) effectively learns the similarity based on simple pertained models with various multiple dimensions, (b) does not uses handcrafted features. This algorithm shows robust performance for various tasks (i.e., measuring textual similarity and paraphrase identification), when it employs on datasets such as Semantic Textual Similarity (STS) benchmark, Sentence Involving Compositional Knowledge (SICK), Microsoft Research Paraphrase Corpus (MRPC) and Wikipedia Question Answer dataset. The performance of our proposed method is compared with different classifiers and the result shows a better accuracy measure than other state-of-the-art methods.