An Implemented review for Intent Creation using different Clustering Techniques

One of the major challenges for any interactive system is understanding the user needs or technically called intent detection. A human interacts with a machine with a same expectation as with another person. The key factor for a machine to make a human like interaction is to know the intent of the user before answering the query. Thus intent detection is one of the crucial and important tasks of any interactive system. Moreover detection of intent prior requires labeled data for training under supervision which can be expensive. Though very few papers have been published showing the methods for intent creation, the main focus of this research is to create the intents before its detection. This paper proposes and implements various techniques like K-means, Agglomerative and LDA for creating the labels. After the creation of the clusters as labels, transfer learning technique called zero shot learning is applied for detecting the intents. Experiment analysis shows that Agglomerative clustering generates better clusters as compared to K-means and LDA.