روبوت محادثة لتحسين أنشطة التسويق بإستخدام التعلم العميق

Chatbot for Improving Marketing Activities Using Deep Learning

With rapid technological advancement, more organizations move from manual customer support to automated processes. In addition, the demand for conversational agents has increased dramatically. Chatbots are conversational agents that simulate conversations with humans through natural language, including voice and text.
Chatbots are typically implemented through either generative or retrieval-based methods. Retrieval-based chatbots rely on predefined answers and lack of flexibility, while generative-based model generates new responses. In addition, customers require assistance when purchasing products which providing this service take a significant amount of time and effort. The main aim of this thesis is to improve a smart marketing chatbot (SMC) and user request understanding. To achieve these aims, two models was proposed; the first is a generative-based chatbot using the seq2seq LSTM model. The second model understands user requests using intent classification and named entity recognition. Models are based on question and answer Amazon office products.
The intent classification implemented using deep learning with the BiLSTM model. The traditional approach of Name Entity Recognition (NER) is insufficient to extract relevant information based on the particular domain of this thesis. To overcome this, custom NER based on BiLSTM employed to provide more accurate and contextualized extraction for the proposed chatbot conversation.
The results show that the chatbot handled user questions fittingly with BLEU score of 57.38. The intent classifier model achieved 94.75% accuracy. Finally, custom NER based on BiLSTM achieved 96.94% accuracy. Moreover, in a comparison of proposed models and relevant researches, the proposed models outperform others. The performance of all models has been improved compared to those in previous studies especially after preprocessing and annotating the dataset as well as customizing the named entities.