Applied Natural Language Processing: User Review Sentiment Classification Optimized by Modified Metaheuristics

E-commerce has become increasingly popular in recent years with increasingly many retailers even conducting business fully online. With the anonymity of the internet, an inability to inspect goods directly, the experiences of other users and seller reputation plays an increasingly important role in purchasing decisions. However, the sentiment expressed in reviews may vary and can be difficult to interpret without manual inspection. This work explores the potential of coupling text processing techniques with machine learning classifiers to handle automatic review sentiment classification. Since ML algorithms are not directly well suited to working with textual data, term frequency inverse document frequency is used to encode data in to a more appropriate format. The AdaBoost classifier is used to determine sentiment expressed in reviews, however, as the AdaBoost classifier is sensitive to parameter selection a modified metaheuristic algorithm is introduced to handle tuning. A comparative analysis of selected relevant optimizers reveals that the proposed optimizer achieves superior results, reaching an objective function score with an overall accuracy of 86.35% for the most effective classification model.