Classifying travel-related intents in textual data using recurrent neural networks

Abstract

Intent classification refers to the process of identifying a set of intents of interest that appear in a given document. This thesis considers the task of annotating travel-related reviews with the travel intents that best represent the reviewer's reason for visiting the place of interest (POI). Fundamentally, this study falls within the field of text mining in that it extracts useful information from the textual data. However, unlike many other related text mining tasks, studies on intent classification have just begun and have yet to gain much prominence. Therefore this work investigates the feasibility of the task using various classifiers including recurrent neural networks with a domain-tailored word embedding model. The utility of the learned model is tested on a location recommendation task. In addition, by applying the model to a large unlabeled dataset, the paper presents some interesting findings regarding travel and tourism in the USA.