Contents lists available at ScienceDirect Information Processing and Management journal homepage: www.elsevier.com/locate/infoproman Deep learning-based sentiment classifcation of evaluative text based on Multi-feature fusion Asad Abdi a, ⁎ , Siti Mariyam Shamsuddin a , Shafaatunnur Hasan a , Jalil Piran b a UTM Big Data Centre (BDC), Universiti Teknologi Malaysia, Johor, Malaysia b Department of Computer Science and Engineering, Sejong University, Seoul, South Korea ARTICLEINFO Keywords: Deep learning Sentiment analysis Natural language processing Neural network ABSTRACT Sentimentanalysisconcernsthestudyofopinionsexpressedinatext.Duetothehugeamountof reviews, sentiment analysis plays a basic role to extract signifcant information and overall sentiment orientation of reviews. In this paper, we present a deep-learning-based method to classify a user's opinion expressed in reviews (called RNSA). To the best of our knowledge, a deep learning-based method in which a unifed feature set which is representative of word embedding, sentiment knowledge, sentiment shifter rules, sta- tistical and linguistic knowledge, has not been thoroughly studied for a sentiment analysis. The RNSA employs the Recurrent Neural Network (RNN) which is composed by Long Short-Term Memory (LSTM) to take advantage of sequential processing and overcome several faws in tra- ditionalmethods,whereorderandinformationaboutthewordarevanished.Furthermore,ituses sentiment knowledge, sentiment shifter rules and multiple strategies to overcome the following drawbacks: words with similar semantic context but opposite sentiment polarity; contextual polarity; sentence types; word coverage limit of an individual lexicon; word sense variations. To verify the efectiveness of our work, we conduct sentence-level sentiment classifcation on large-scale re- viewdatasets.Weobtainedencouragingresult.Experimentalresultsshowthat(1)featurevectors in terms of (a) statistical, linguistic and sentiment knowledge, (b) sentiment shifter rules and (c) word-embeddingcanimprovetheclassifcationaccuracyofsentence-levelsentimentanalysis;(2) our method that learns from this unifed feature set can obtain signifcant performance than one that learns from a feature subset; (3) our neural model yields superior performance improve- ments in comparison with other well-known approaches in the literature. 1. Introduction In recent years, with the rapid development of social media, a vast amount of reviews, opinions and feedbacks are constantly produced from all over the world every day. Many organizations and peoples follow user opinions and comments to decide the quality and performance of a product or service. Therefore, there is a need to analyze these data in order to extract and detect relevant information and the polarized opinion, respectively. On the other hands, it is important for companies and peoples to automaticallyidentifyauser'sopinionwhetheritispositiveornegative.Opinion/sentimentanalysisisatechniquethatcanbeused to determine and classify people's opinion according to their polarity. Sentiment analysis is a task to fnd automatically subjective https://doi.org/10.1016/j.ipm.2019.02.018 Received 6 September 2018; Received in revised form 24 January 2019; Accepted 28 February 2019 ⁎ Corresponding author. E-mail addresses: seyedasadollah-pd@utm.my (A. Abdi), mariyam@utm.my (S.M. Shamsuddin), shafaatunnur@utm.my (S. Hasan), piran@sejong.ac.kr (J. Piran). Information Processing and Management 56 (2019) 1245–1259 0306-4573/ © 2019 Published by Elsevier Ltd. T