Dynamic Automated Labeling System for Real-Time User Intention Analysis

Abstract

Real-time online user responses can be useful for identifying people's reactions to business strategies or social phenomena. Many approaches have been proposed to classify user responses, yet most of them exploit either manual or automatic labeling that fails to reflect real-time information, preventing use as a generalized model. In this study, we propose a dynamic automated labeling technique that considers changes in user response over time by combining LDA (Latent Dirichlet Allocation) topic modeling and word2vec. Topic modeling is used to extract topics that will be used as true labels, and word2vec is used to map the dominant topic into label to annotate the users' textual data by calculating the vector similarity. Experimental results on disaster-related Tweet datasets show that our approach outperforms other baseline approaches of automated annotation in terms of classification performance. The technique presented in this study can be applied in business to create an automated system to analyze and continuously monitor users' real-time reactions.