Hybrid CNN-LSTM Model with GloVe Word Vector for Sentiment Analysis on Football Specific Tweets

Abstract

The football fans feelings get unfold during the different phases of the football match, they express their opinions, views, thoughts, judgments of the match, attitude towards player, and emotions, stance on social media like Twitter. The change in fans opinions are reflected in a series of tweets written by fans. This research work focuses on identifying and analyzing sentiment in tweets expressed by football fans on Twitter. To perform sentiment analysis, we analyzed different word embedding techniques to represent word vectors that capture semantic and syntactic information. In this paper, we use a global vector(GloVe) word embedding technique which produces word vector with substructure and leverages statistics. In addition to GloVe, we also construct a sentiment lexicon as additional information. The word vector produced by GloVe and sentiment lexicon is the inputs for the proposed hybrid CNN-LSTM deep learning model. The proposed CNN-LSTM model blend benefits of CNN and LSTM, CNN used to excerpt features from word embedding that reflect short-term sentiment dependency while LSTM used to build long-term sentiment relationships among words. This the paper also used machine learning algorithms such as Random Forest, Support Vector Machine, Multinomial Nave Bayes, KNearest Neighbours(KNN) and XG Boost for sentiment analysis and sentiment classification. We evaluated the performance of proposed hybrid CNN-LSTM with GloVe word embedding approach with 2018 FIFA world cup tweets dataset, our experiment results show 85.46% and 92.56% validation and testing accuracy respectively. Further, our experiment results also demonstrate that the Random Forest algorithm perform consistent and robust performance compared to other machine learning classifiers, it perceive fan’s sentiment during football events.