TrBot: A Turkish Deep Learning Chatbot Utilizing Seq2Seq Model

In Natural Language Processing (NLP) and Artificial Intelligence (AI), chatbots, which are software programs designed to facilitate human-computer interaction through natural language, are becoming increasingly important. However, creating an effective chatbot remains a complex task, as it must accurately interpret user input and generate appropriate responses. This study presents TrBot, a general-purpose Turkish chatbot that utilizes deep learning techniques, specifically a seq2seq model with Long Short-Term Memory (LSTM) layers. This architecture allows TrBot to manage sequential dependencies and effectively generate coherent responses, offering advantages in handling the complex morphological structure of Turkish. In contrast to earlier Turkish chatbots that were application-specific, TrBot is designed for broad conversational use across various topics. In this study, we also proposed and created two comprehensive datasets: a QA dataset with 40,702 entries and a conversation dataset with 304,446 entries, both specifically designed to enhance TrBot’s performance. Trained on these datasets, TrBot achieved an accuracy of 80% on the QA dataset and 70% on the dialog dataset, with BLEU scores of 0.90 and 0.77 respectively, indicating substantial enhancements in response quality. In comparison, a Transformer-based model exhibited reduced training times but achieved lower accuracies of 60% on the QA dataset and 50% on the dialog dataset, with BLEU scores of 0.76 and 0.61 respectively, with the limited size of the datasets and available computational resources. The development of TrBot has significant implications, offering potential benefits in areas such as customer support, language learning, and other fields that require robust Turkish conversational capabilities. This study demonstrates that with adequate data and appropriate modeling techniques, it is possible to create effective conversational agents for complex languages like Turkish, paving the way for further advancements in this domain.