A Hybrid Deep Learning Architecture for Sentiment Analysis

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 482–493, Osaka, Japan, December 11-17 2016. A Hybrid Deep Learning Architecture for Sentiment Analysis Md Shad Akhtar, Ayush Kumar, Asif Ekbal, Pushpak Bhattacharyya Department of Computer Science & Engineering Indian Institute of Technology Patna, India {shad.pcs15,ayush.cs12,asif,pb}@iitp.ac.in Abstract In this paper, we propose a novel hybrid deep learning architecture which is highly efficient for sentiment analysis in resource-poor languages. We learn sentiment embedded vectors from the Convolutional Neural Network (CNN). These are augmented to a set of optimized features se- lected through a multi-objective optimization (MOO) framework. The sentiment augmented op- timized vector obtained at the end is used for the training of SVM for sentiment classification.We evaluate our proposed approach for coarse-grained (i.e. sentence level) as well as fine-grained (i.e. aspect level) sentiment analysis on four Hindi datasets covering varying domains. In order to show that our proposed method is generic in nature, we also evaluate it on two benchmark En- glish datasets. Evaluation shows that performance of the proposed method are consistent across all the datasets and often outperform the state-of-art systems. To the best of our knowledge, this is the very first attempt where such a deep learning model is used for sentiment analysis in less-resourced languages such as Hindi. 1 Introduction Sentiment Analysis (Pang and Lee, 2008) in natural language processing (NLP) deals with the problem of identifying the polarity in a user generated content. With growing social media platforms such as Twitter, Facebook etc., copious amount of data is being generated continuously. According to Domo’s Data Never Sleep 2.01, the global internet population is about 2.4 billion users. Online platforms such as Twitter alone generate over 300,000 tweets per minute2. At the same time more than 26K user reviews are posted on Yelp, an online user review portal. This tremendous amount of semi-structured data poses a great challenge in its efficient processing for any specific purpose. Sentiment analysis for web generated content e.g. tweets and online reviews, is a cumbersome problem mainly due to its unstructured and noisy nature (e.g. gr8, g8 etc. for great) and spelling and grammatical mistakes. Considering the challenges as mentioned above, authors have proposed their sentiment analyzers for Twitter data and/or online reviews (Kim and Hovy, 2004; Mohammad et al., 2013a; Gupta et al., 2015). However, most of the works have been done on the resource-rich languages such as English. India is a multi-lingual country with great linguistic and cultural diversities. There are 22 officially spoken languages. However, there have not been enough research works that address sentiment analysis involving Indian languages, except few such as (Balamurali et al., 2012; Bakliwal et al., 2012; Kumar et al., 2015). However, these existing works do not address the fine-grained sentiment analysis at the aspect level. The prime reason behind this is the scarcity of benchmark datasets and other resources/tools in Indian languages. In our work, we focus on sentiment analysis in Hindi, the official language of India and the …

Content truncated to 500 words