Kurzbeschreibung
This paper explores the potential of using Skip-Thought sentence encoding as part of an intent based chatbot system. Skip-Thought vectors constitute a state of the art approach of sentence embedding that aims to allow machines to extract the meaning of a sentence.
For that purpose a support vector machine classifier is trained on a dataset of 64 intents using Skip-Thought encoding as a feature generator. The implementation is tested against industry leading platforms such as IBM Watson, Google Dialogflow or Cognigy AI.
Despite being built with a much lower complexity and using sig- nificantly less resources, the Skip-Thought vector based chatbot ranks 3 out of 5 with respect to accuracy and F1-Score, showing the high potential Skip-Thought embedding provides.