Analysis of Turkish customer reviews with text mining techniques

Today, rapidly developing technology has caused shopping activities to be moved to e-commerce platforms. This has increased the value of customer reviews in purchasing. As these guideline customer reviews represent satisfaction and complaints about products or services, they provide a report for both individuals and companies. In particular, companies use various text mining techniques to detect negative comments as quickly as possible and thus improve business. However, studies on the analysis of Turkish texts are scarce. From this point of view, text mining and natural language processing techniques were used to classify Turkish customer reviews as positive, negative and neutral. In this direction, The Scoring System Dictionary-Based (SSDB), which is a dictionary-based approach in which new original dictionaries and language rules are produced for Turkish, is proposed. The SSDB algorithm is a new approach that ranks the words in the text by analyzing them in terms of meaning and includes rules that help to determine the meaning of the text more accurately. With the proposed approach, customer reviews scraped from an e-commerce platform were analyzed. The predictions of the algorithm were compared with five different machine learning methods trough various performance metrics. The SSDB algorithm is successful in Turkish text classification.