A hybrid machine learning model for sentiment analysis and satisfaction assessment with Turkish universities using Twitter data

Social media platforms, such as X (Twitter), contain enormous amounts of data, and collecting valuable information from that data assists in making more informed decisions. In recent years, governments and institutions have begun to explore the possibilities of utilizing social networks as a platform to supply and enhance the quality of their services. Consequently, there is an increased demand to estimate people’s satisfaction with Turkish Universities written in Turkish wherever it is challenging and requires more effort to cope with its rich morphological structure to perform a Sentiment Analysis task. This study proposes a Turkish text sentiment analysis methodology for estimating people’s satisfaction with Turkish Universities by employing nine conventional machine learning models, deep learning techniques, and BERT-based transformers upon an original manually annotated dataset consisting of 17,793 tweets in Turkish. An innovative hybrid architecture, named BERT-BiLSTM-CNN, integrates Bidirectional Encoder Representations from Transformers (BERT), Bidirectional Long Short-Term Memory (BiLSTM), and a triple parallel Convolutional Neural Network (CNN) branch, achieving exceptional accuracy in sentiment analysis of Turkish tweets. During testing, the proposed architecture revealed an impressive accuracy rate of over 0.9101, an F1 Score of 0.8801, and a Receiver Operating Characteristic (ROC) of 0.9632 for analyzing sentiment, demonstrating that the model outperformed the state-of-the-art models, and it is capable of coping with the linguistic complexities of Turkish sentiment analysis. This work provides new insights into sentiment analysis by proposing a hybrid model that combines several computational methodologies to improve the understanding of attitudes.