Abstract
Aspect-level sentiment classification (ASC) has received much attention these years. With the successful application of attention networks in many fields, attention-based ASC has aroused great interest. However, most of the previous methods did not analyze the contribution of words well and the context-aspect term interaction was not well implemented, which largely limit the efficacy of models. In this paper, we exploit a novel method that is efficient and mainly adopts Multi-head Attention (MHA) networks. First, the word embedding and aspect term embedding are pre-trained by Bidirectional Encoder Representations from Transformers (BERT). Second, we make full use of MHA and convolutional operation to obtain hidden states, which is superior to traditional neural networks. Then, the interaction between context and aspect term is further implemented through averaging pooling and MHA. We conduct extensive experiments on three benchmark datasets and the final results show that the Interactive Multi-head Attention Networks (IMAN) model consistently outperforms the state-of-the-art methods on ASC task.
Highlights
Aspect-level sentiment classification (ASC) is a fundamental task in the field of sentiment analysis which aims to identify the sentiment polarity (e.g. Positive, Negative, Neutral) in a specific aspect term explicitly occurring in the context [1], [2]
We propose a new model named Interactive Multi-head Attention Networks (IMAN) for ASC task, which mainly adopts multiple Multi-head Attention (MHA) mechanisms
In addition to the Long Short-Term Memory (LSTM), many neural networks similar to LSTM structures are applied to ASC task, such as Recurrent Neural Network (RNN), bidirectional RNN (Bi-RNN) and gated RNN (GRNN) et al For example, Zhang et al [23] proposed a sentence-level neural model, which uses bidirectional GRNN to connect the words in a twitter and model the interaction between the target and its surrounding contexts
Summary
Aspect-level sentiment classification (ASC) is a fundamental task in the field of sentiment analysis which aims to identify the sentiment polarity (e.g. Positive, Negative, Neutral) in a specific aspect term explicitly occurring in the context [1], [2]. The human being asked to complete this task will selectively focus on certain parts of the context and obtain information where needed to establish an internal representation of an aspect term in mind [12] Inspired by this observation, attention mechanism is widely applied and introduced into the neural network architecture for a variety of ASC applications. We believe that only the coordination of targets and corresponding context can really improve the effectiveness of sentiment classification Motivated by these observations, we propose a new model named Interactive Multi-head Attention Networks (IMAN) for ASC task, which mainly adopts multiple Multi-head Attention (MHA) mechanisms.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
