This project aims to conduct sentiment analysis to mine the sentiments of customers regarding their purchase from Daraz. To conduct sentiment analysis, I thoroughly preprocessed the data following a number of steps and experimented with multiple classification models (Naive Bayes Multinomial Model, J48, and others) using a cross-validation test to choose the best model. Finally, I developed the most precise and accurate model using Weka Knowledge Flow: using the Naive Bayes Multinomial Model the accuracy achieved was above 95%.
Data Set and Source: For the data set, I relied on the customer purchase dataset of Daraz that I collected from the Alibaba cloud platform. The dataset consists of 4 sentiment metrics: purchase rate (1-8), delivery satisfaction rate (0-5), repurchase rate (1-8), and bounce rate (0-5). These core 4 factors were critically converted to usable numeric data by the BI analysts at Daraz. Each customer's data was then divided into positive, neutral and negative sentiment for training and testing purposes.
ML algorithm used and justification: To develop the ML algorithm, I first used the experimenter option in Weka and tried to compare between models mostly used for sentiment analysis. After setting the parameters and checking the model using cross validation, I used the Naive Bayes Multinomial model for building our sentiment classifier. The program guesses the tag of the sentiments (positive, negative and neutral) using the Bayes theorem. It calculates each sentiment's likelihood for a given sample and outputs the sentiment with the greatest chance. For this, I used the Weka Knowledge Flow and tried to build the model with the flow diagram.
Findings: Findings show that the model's accuracy is 95.33%, precision is 95% and kappa statistics value of 0.93 (>0.7 is a very good model). This model helps to classify the sentiment of customers who shop at Daraz and also provide predicted sentiment. By classifying the sentiment, companies like Daraz can dig into what is causing more negative sentiment among customers and try to enhance customer satisfaction.