CUSTOMER PRODUCT REVIEWS CLASSIFICATION WITH LIFELONG SENTIMENT ANALYSIS

In this study, sentiment classification of customer comments collected from different areas was performed within the scope of lifelong learning approach. Customer comments collected from different products randomly selected from an online shopping site were used for this classification. Each product was considered as a different area for lifelong learning approach. Customer comments were preprocessed and Countvectorizer and TF-IDF methods were used to form word vectors. Lifelong naive Bayes and lifelong logistic regression classification methods were used within the scope of lifelong sentiment classification, and coded with Python programming language. In addition to these two methods, an aggregation-based sentiment classification model was presented inspired by the literature. Aggregation-based sentiment classification is based on a lifelong learning process based on the sum of word frequencies. The performances of all methods used were compared, and it was observed that the method that gave the best result was the aggregation-based sentiment classification model. When all the obtained results were evaluated, the problem addressed is important and up-to-date both in terms of testing and comparing the performances of lifelong sentiment classification algorithms and for e-commerce transactions. In addition, it can be said that this study, which was conducted for the first time using Turkish dataset within the scope of lifelong learning, is a guide for future studies.