Determing Aspect Ratings and Aspect Weights from Textual Reviews by Using Neural Network with Paragraph Vector Model
Aspect-based analysis currently becomes a hot topic in opinion mining and sentiment analysis. The major task here is how to detect rating and weighting for each aspect based on an input of a collection of users’ reviews in which only the overall ratings a
- PDF / 573,530 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 88 Downloads / 163 Views
University of Engineering and Technology, VNU, Hanoi, Vietnam [email protected] 2 Faculty of Information Technology, Ton Duc Thang University, Ho Chi Minh City, Vietnam [email protected] 3 Electric Power University, Hanoi, Vietnam [email protected]
Abstract. Aspect-based analysis currently becomes a hot topic in opinion mining and sentiment analysis. The major task here is how to detect rating and weighting for each aspect based on an input of a collection of users’ reviews in which only the overall ratings are given. Previous studies usually use a bag-of-word model for representing aspects thus may fail to capture semantic relations between words and cause an inaccuracy of aspect ratings prediction. To overcome this drawback, in this paper we will propose a model for aspect analysis, in which we first use a new deep learning technique from [8] for representing paragraphs and then integrate these representations into a neural network model to infer aspect ratings and aspect weights. The experiments are carried out on the data collected from hotel services with the aspects including “cleanliness”, “location”, “service”, “room”, and “value”. Experimental results show that our proposed method outperforms the well known studies for the same problem.
1
Introduction
In recent years, opinion mining and sentiment analysis has been one of the attracting topics of knowledge mining and natural language processing. It is the task of detecting, extracting and classifying opinions and sentiments concerning different topics, as mentioned in textual input. Some works have been done to this task such as rating the overall sentiment of a sentence/paragraph, or a textual review regardless of the entities (e.g., movies) from reviews, [13,15], opinion extraction and sentiment classification [3,4], detect comparative sentences from reviews [6,7], extracting information and summarization from reviews [9,12,23]. However these works fail to capture the sentiments over the aspects on which an entity can be reviewed. For example, the entity is a hotel which can contains some aspects as “cleanliness”, “location” and “service”. c Springer International Publishing Switzerland 2016 H.T. Nguyen and V. Snasel (Eds.): CSoNet 2016, LNCS 9795, pp. 309–320, 2016. DOI: 10.1007/978-3-319-42345-6 27
310
D.-H. Pham et al.
Sentiment for each aspect is the important information, therefore there are now more studies working on aspect based sentiment analysis. Hu and Liu [5] focused on the taks of determining aspects in given textual reviews. They assumed that product aspects are expressed by nouns and noun phrases and their frequencies are used for identifying aspects. Wu et al. [20] used a language model and a phrase dependency parser to detect product aspects, expression of opinion and relations between them. Several other works focused on rating aspects such as Snyder and Barzilay [16] proposed the good grief algorithm for modeling the dependencies among aspects and Titov and McDonald [17] used a topic based model and a regression model for ext
Data Loading...
