Joint Slot Filling And Intent Prediction for Natural Language Understanding in Frames Dataset

Spoken Dialogue System, Chatbots has emerged as an important research topic in artificial intelligence and natural language processing domain. The tasks in Spoken Dialogue System, Chatbots are mainly classified into three viz. domain classification, slot filling and intent prediction. In this paper, we present a novel method for slot filling and intent prediction by appending the intent information with each slot, which can be used in handling complex tasks such as travel planning. Inspired by Multi-Domain Joint Semantic Frame Parsing using Bidirectional RNN-LSTM, we trained the model using bidirectional RNN-LSTM to jointly predict the slot values and intent for a single text having multiple intents. This method proved to be successful in predicting slot values and intents with an accuracy of 90%. The system uses IOB tagged dataset generated from the Microsoft's Human-human goal oriented dataset(Frames Dataset) for training and testing.