SBSOM: Self-Organizing Map for Visualizing Structure in the Time Series of Hot Topics (Joint Workshop of Vietnamese Society of AI, SIGKBS-JSAI, ICS-IPSJ and IEICE-SIGAI on Active Mining) -- (Session 1: Text Mining 1)

In this paper, we propose a Sequence-Based Self-Organizing Map(SBSOM) that organizes clusters as series within the map to visualize their structure in terms of hotness, period and relations among topics. Principal Component Analysis(PCA) that is based on probabilistic document generation model is applied to extract hot topics from vast amount of documents, and these hot topics are used to label each document. Afterwhich, SBSOM is used to visualize these hot topics in a time series. SBSOM is also extended by defining label confidence for a more accurate labeling of its neurons. The initial experiments that use two kinds of news articles, the largest expands across ten years, validate that in addition to SOM showing only hotness of topics and relations among topics throughout whole period, SBSOM shows hotness within certain times, relations among topics, and period of topics.