Enhanced Aspect Level Sentiment Classification with Auxiliary Memory

Proceedings of the 27th International Conference on Computational Linguistics, pages 1077–1087 Santa Fe, New Mexico, USA, August 20-26, 2018. 1077 Enhanced Aspect Level Sentiment Classification with Auxiliary Memory Peisong Zhu, Tieyun Qian ∗ School of Computer Science, Wuhan University, China {zhups24, qty}@whu.edu.cn ∗Contact author Abstract In aspect level sentiment classification, there are two common tasks: to identify the sentiment of an aspect (category) or a term. As specific instances of aspects, terms explicitly occur in sen- tences. It is beneficial for models to focus on nearby context words. In contrast, as high level semantic concepts of terms, aspects usually have more generalizable representations. However, conventional methods cannot utilize the information of aspects and terms at the same time, be- cause few datasets are annotated with both aspects and terms. In this paper, we propose a novel deep memory network with auxiliary memory to address this problem. In our model, a main memory is used to capture the important context words for sentiment classification. In addition, we build an auxiliary memory to implicitly convert aspects and terms to each other, and feed both of them to the main memory. With the interaction between two memories, the features of aspects and terms can be learnt simultaneously. We compare our model with the state-of-the-art methods on four datasets from different domains. The experimental results demonstrate the effectiveness of our model. 1 Introduction Sentiment analysis (Pang and Lee, 2008; Liu, 2012) is a fundamental task in the field of natural language processing. As one of the subtasks of sentiment analysis, the goal of aspect level sentiment classification is to infer the sentiment polarity (e.g. positive, negative, neutral) of the aspect (e.g. food, service). Aspect level sentiment classification is arousing more and more researchers’ attention, as it can provide more all-round and deeper analysis than document or sentence level sentiment classification. There are two common scenes in aspect level sentiment classification. The first one aims to identify the sentiment of a predefined aspect, we call it ASC (aspect sentiment classification) for short. The other aims to identify the sentiment of an explicit term which occurs in the sentence, we call it TSC (term sentiment classification). We take the review “A mix of students and area residents crowd into this narrow, barely there space for its quick, tasty treats at dirt-cheap prices.” as an example. For ASC task, we need to infer the sentiment polarities for a set of predefined aspects including “food”, “price”, and “ambience”. Here, the sentiment polarities for “food” and “price” are positive while that for “ambience” is negative. For TSC task, we are given three terms “space”, “treats”, “prices” in the review and need to infer the sentiment polarities for these terms. Here, the sentiment polarities for terms “treats” and “prices” are positive while that for “space” is negative. In most cases, aspect level sentiment classification works in a supervised manner. Researchers usually extract features and use machine learning algorithms to train the sentiment classifier. Typical meth- ods are usually based on …

Content truncated to 500 words