In the field of business intelligence, the context in which customers see, hear and think about a product plays an important role in their call of buying the product. Context-sensitive sentiment classification methods determine the polarity of the sentiment terms by considering the contexts of the target word. Most of the present techniques consider only product-level, user-level contexts for sentiment classification. These contexts are more general and depend on additional features to achieve good performance. Feature-level contexts e.g., car’s features include mileage and its context comprises city, highway, short-trips, long-trips and hill station providing fine-grained information needed for sentiment classification. This paper presents a neural network model referred to as Contextual Sentiment LSTM to automatically learn feature-level contexts based on an attention mechanism. Contextual Sentiment LSTM integrates background knowledge about feature, sentiment and context words from knowledge bases with the currently processed review text. A sentence vector generated based on the correlation between feature, opinions and context words in a sentence is classified using a softmax classifier. The proposed model is tested on the benchmark Car dataset and compared its contribution with progressive models like the CMLA, Sentic LSTM, CNN+LP, and RNCRF. Results demonstrate that our model achieves good sensitivity and accuracy when compared to others.
Digital Object Identifier (DOI)
K .Lavanya, S. and Parvathavarthini, B.
"Co-Extraction of Feature Sentiment and Context Terms for Context-Sensitive Feature-Based Sentiment Classification using Attentive-LSTM,"
Applied Mathematics & Information Sciences: Vol. 13:
5, Article 7.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol13/iss5/7