SenHint: A Joint Framework for Aspect-level Sentiment Analysis by Deep Neural Networks and Linguistic Hints

Introduction

 The state-of-the-art techniques for aspect-level sentiment analysis focus on feature modeling using a variety of deep neural networks (DNN). Unfortunately, their practical performance may fall short of expectations due to semantic complexity of natural languages. Motivated by the observation that linguistic hints (e.g. explicit sentiment words and shift words) can be strong indicators of sentiment, we present a joint framework, SenHint, which integrates the output of deep neural networks and the implication of linguistic hints into a coherent reasoning model based on Markov Logic Network (MLN). In SenHint, linguistic hints are used in two ways: (1) to identify easy instances, whose sentiment can be automatically determined by machine with high accuracy; (2) to capture implicit relations between aspect polarities. We also empirically evaluate the performance of SenHint on both English and Chinese benchmark datasets. Our experimental results show that SenHint can effectively improve accuracy compared with the state-of-the-art alternatives.

Framework

An example of SenHint, including first-order logic rules and ground factor graph, has been shown in Figure 1. In the graph, aspect polarities are represented by variables (round nodes in the figure), and the influences of DNN output and linguistic implication are represented by factors (box nodes in the figure). The value of a variable indicates its corresponding aspect sentiment. There are two types of variables: evidence variables and infer variables. Evidence variables correspond to the easy aspect polarities that have been determined by explicit linguistic hints. They participate in the MLN inference, but their values are specified beforehand and remain unchanged in the inference process. The values of infer variables should instead be inferred based on the constructed MLN. Additionally, there are three types of factors: DNN factor, similar factor and opposite factor. The DNN factor simulates the effect of DNN output on aspect polarity. The similar factor and opposite factor represent implicit relations between aspect polarities.

Alt text

Document

SenHint: A Joint Framework for Aspect-level Sentiment Analysis by Deep Neural Networks and Linguistic Hints(Technical Report)

Download

Video on YouTube

Download poster

Source code

Domain sentiment words