Sentimental Analysis of Twitter Users from Turkish Content with Natural Language Processing

Area of Science:

• Artificial Intelligence
• Machine Learning
• Natural Language Processing
• Sentiment Analysis

Background:

• Social media use has increased, making posts valuable data for sentiment analysis.
• Turkish language presents unique challenges for sentiment analysis due to its agglutinative nature.

Purpose of the Study:

• To perform sentiment analysis on Turkish Twitter data using machine learning algorithms.
• To assess the impact of the pandemic on public opinion through sentiment analysis.
• To create a benchmark dataset for Turkish sentiment analysis.

Main Methods:

• Utilized Natural Language Processing techniques within a machine learning framework.
• Applied several machine learning algorithms to analyze Turkish tweets.
• Created a custom dataset, SentimentSet, by manually marking tweets related to the pandemic.

Main Results:

• Achieved classification accuracy up to approximately 87% on test data from both public and custom datasets.
• Demonstrated classification accuracy up to approximately 84% on a smaller, custom-generated test dataset.
• The results highlight language-specific sentiment analysis for Turkish.

Conclusions:

• Machine learning algorithms can effectively perform sentiment analysis on Turkish social media data.
• The developed SentimentSet dataset can serve as a benchmark for future research.
• This study contributes to understanding public opinion and language-specific sentiment analysis in Turkish.