Abstract:
|
In this presentation, we report and visualize the results of evaluating customer reviews and purchasing behavior obtained from the supervised learning sentiment analysis of the consumer feedback data. The verbatim collected through phone, email, social media, chat and e-commerce was classified into categories such as complaint, praise and suggestion by our service associates. The text mining done with the Neural Networks models demonstrated low misclassification rates in training (6.5%) and validation (9%). Interestingly, comparing the modeling results and original coding of the verbatim, we found some data entry issues, in which the actual documents were misclassified by humans. The evaluation was enhanced by discriminant analysis, which tested the quality of fitting and linked classification categories (complaint, praise and suggestion) with the products mentioned in the customer feedback. The results were visualized via the ternary plots.
|