A Deep Learning Study of Aspect Similarity Recognition

Aspect Similarity Recognition (ASR) is to identify whether two sentences express one or some aspects in common. This task is useful in review summarization where a summarized review needs to cover all aspects as well as avoid redundancy. To facilitate the application of supervised learning models for this task, we construct a dataset ASRCorpus containing two domains (i.e., LAPTOP and RESTAURANT). The four models i.e., Word Average, CNN, LSTM and BiLSTM are employed to evaluate the performances of machine learning methods. According to the experimental results, the recurrent neural networks are the most efficient methods. We also analysis some typical samples to identify the task's challenges.