Short text clustering method using Naïve Bayes and normalized Word2Vec vectors with dynamic data periods

Abstract

In order to improve clustering accuracy by dynamic referencing of data corpus, which is important in trend analysis with ever-changing vocabulary, we propose a short text clustering method using Complement Naïve Bayes and normalized Word2Vec vectors with dynamic data periods. We propose a short text clustering method using Complement Naïve Bayes and Normalized Word2Vec vectors with dynamic data periods. The proposed method enables accurate analysis of general trends over time, which can be compared with past trends and predict trends based on intergenerational perceptions such as generation gaps, thereby providing information for further marketing strategies. In an experiment to compare the clustering output of playboard short text data over a specified period of time with the specific trend of that time period, we will confirm the impact on accuracy by dynamically handling the data period. Trend analysis during arbitrary time periods that have the potential to create value in marketing will bring new perspectives to business and open the door to richer advertising communication.