Aspect-Based Sentiment Classification of Online Product Reviews Using Hybrid Lexicon-Machine Learning Approach

Nowadays, a good number of customers express their experience with online products. These reviews have an important role in customers’ purchase decision process. There may be hundreds or thousands of unstructured and heterogeneous reviews for a popular product. Traditional text processing techniques have limited capability in extracting opinions on customers’ product reviews from huge data over the Internet. Although lexical approaches aim to map words to sentiments by building a lexicon, the process of developing a lexicon with sentiment scores for phrases and sentences becomes tedious and time consuming as data volume increases. Currently, text sentiment analysis requires fast and accurate techniques to decode and quantify the emotion in tweets. This paper presents a hybrid framework based on lexicon and machine learning (ML) algorithms to train previously seen tweets in order to predict the sentiments of some new input tweets into positive, negative, and neutral polarities. Tweepy library was used to extract tweets on Laptop reviews to identify some aspects and classify sentiments towards them into specific polarity. After data pre-processing, the implementation in Python used Natural Language Processing package called TextBlob to assign subjectivity and polarity scores to text. The scores were used by the ML algorithms to analyze and classify sentiments. A dataset of 2226 tweets was used for training and testing Support Vector Machine, Random Forest, and Naïve Bayes classifiers. Results indicate that Random Forest classifier outperforms others in the task of classifying sentiments on Laptop reviews with the highest accuracy (96%), precision (97%), and F1-score (96%).