Sentiment Analysis on Turkish Customer Social Media Reviews

This project shows how Machine Learning and NLP can classify Turkish text reviews into positive or negative emotions. The models can help:

1. Companies to understand customer satisfaction.
2. Researchers to study emotional trends in Turkish language data.
3. Businesses to improve products and services using customer feedback.

The study followed a step-by-step approach:

1. Data Preprocessing

   • Cleaned and organized text.

   • Removed stop words (common words like ve, ama).

   • Handled missing values.

2. Text Representation

   • Used TF-IDF vectorization to turn text into numerical values.

   • This made the reviews ready for machine learning models.

3. Classification Models

   • Logistic Regression

   • Decision Trees

   • Random Forest

   • Support Vector Machines (SVM)

   • k-Nearest Neighbors (k-NN)

4. Model Evaluation

   • Measured with Accuracy, Precision, Recall, F1 Score, AUC.

   • Applied 5-Fold Cross Validation for reliability.

5. Visualization

   • WordCloud for most common words.

   • Histograms to analyze review lengths.

   •  

• Best Models: SVM and Logistic Regression gave the highest accuracy and AUC.

• Weakest Model: k-NN showed the lowest performance.

• Key Findings:

  • NLP and ML methods are effective in Turkish sentiment analysis.

  • Visualization helps to understand word patterns and review length trends.