| Summary: | Nearly six billion people globally use smartphones, and reviews about smartphones provide useful feedback concerning important functions, unique characteristics, etc. Social media platforms like Twitter contain a large number of such reviews containing feedback from customers. Conventional methods of analyzing consumer feedback such as business surveys or questionnaires and focus groups demand a tremendous amount of time and resources, however, Twitter’s reviews are unstructured and manual analysis is laborious and time-consuming. Machine learning and deep learning approaches have been applied for sentiment analysis, but classification accuracy is low. This study utilizes a transformer-based BERT model with the appropriate preprocessing pipeline to obtain higher classification accuracy. Tweets extracted using Tweepy SNS scrapper are used for experiments, while fine-tuned machine and deep learning models are also employed. Experimental results demonstrate that the proposed approach can obtain a 99% classification accuracy for three sentiments.
|
|---|