A Comparison of Natural Language Understanding Platforms to Build Arabic Chatbots

Chatbots are computer programs that serve as virtual assistants by conversing with users in their natural languages. Basically, every chatbot contains a Natural Language Understanding (NLU) component at its core . A lot of platforms for building chatbots are created recently with a variety of functionalities, capacities, and price schemes. Chatbot services are based on NLU engines which deal with sentences to recognize information such as entities and intents. Numerous NLU platforms are also recently made available to act like an NLU component for chatbots, however choosing the right NLU for Arabic chatbots is still an issue. In order to choose a suitable platform to develop an Arabic chatbot, the categorization, and core features of the most common conversational platforms that support Arabic language (Watson, LUIS, Pandora, RASA and Wit.ai), are highlighted and compared in this paper. Among the studied platforms, the Wit.ai NLU platform is chosen as an equivelant, popular, and free available platform to be evaluated in terms of generating the structured language from the unstructured Arabic language. In the proposed model, the standard Arabic dialect is used for training and validation. A virtual bank customer service department from Al-Rasheed Bank in Mosul is presented as a knowledge base. In the phase of intent classification and entity extraction, Wit.ai proved to possess high values of precision and F1 score with 0.96 and 0.948, respectively.