Extraction of Characteristic Frequent Visual Patterns by Distributed Representation
Saki Kawanobe, Tomonobu Ozaki · 2017
Frequent pattern mining is one of the most important and fundamental tasks in data mining. While a large number of sophisticated techniques on frequent pattern mining are proposed, two essential drawbacks on frequent pattern mining, i.e. the explosion of discovered patterns and less comprehensibility of patterns, are still remained unsolved. In this paper, we propose an application of distributed representation to the area of frequent pattern mining in order to alleviate the drawbacks and to derive characteristic patterns with high understandability. More precisely, given a set of visual patterns derived from databases on tagged thumbnail images in social media, we attempt to identify characteristic visual patterns by the cluster analysis in the vector space obtained by distributed representation. In addition, to help understanding of the meanings of visual patterns, we associate plural tag patterns having similar vector representations with each visual pattern. A series of experiments are conducted to assess the effectiveness of the proposed framework using real tagged thumbnail images in Nicovideo(nicovideo.jp). The results confirm that the proposed framework can provide better vector representation of visual patterns compared with other dimensionality reduction techniques to identify characteristic patterns with high understandability.