Empathy and Persona Based Arabic Conversational Agent

Abstract

Arabic NLP has seen significant advances in Natural Language Understanding (NLU) with language models such as AraBERT and ArBert. Despite that, the Arabic Natural Language Generation (NLG) remains a challenge. Due to the lack of Arabic datasets suitable to train NLG models and the lack of pre-trained Arabic models for NLG. Therefore, we plan to pave the way for future Arabic chatbot research and Arabic NLG. By fine-tuning an empathetic and personified, Arabic conversational agent. By using “persona” we make the conversation more engaging. And with “empathy” we make the conversation more human-like. Currently, the state of Arabic chatbot research is led by the paper “Empathetic BERT2BERT Conversational Model: Learning Arabic Language Generation with Little Data.” which has the empathy part but it’s still lacking on the chit-chat model part. And so, we have managed to integrated persona into our final model by integrating and fine-tuning the model, which is a BERT2BERT architecture, on the ArabicPersonaChat dataset. Scoring 0.79 on the BLEU benchmark on the “ArabicEmpatheticDialogues” dataset compared to the previous SOTA which scored 0.675.