Best Paper Award

  Online media is now one area where a variety of social and economic activities are happening. Huge amount of data generated there have been providing us with expanding opportunities of user activity analysis. Focusing on such large-scale online activity data, this paper proposes a non-linear time-series analysis method, named EcoWeb. So far, researchers have intensively studied time-series data analysis, but only a few have explored the flexible modeling of feature patterns based on domain knowledge. Inspired by the nature of the natural world where different species are scrambling for food resources, the authors propose a modeling method based on dynamic non-linear systems that imitate species population transitions observed in the natural ecosystem. The proposed modeling enables the flexible expression of time-series patterns, such as potential competitions and seasonal activities, and retrieval of important features that are closer to our intuition. This paper formalizes the proposed modeling and proposes learning algorithms based on the modeling. The authors also disclose the experiment that the authors conducted using time-series data including search occurrences regarding four keywords in a certain Web search engine. This experiment clarifies that the proposed method is capable of retrieving important features more efficiently and more precisely in comparison with conventional methods. As remarked above, the paper has introduced a novel non-linear time-series analysis method in which a high level of originality and effectiveness is recognized. Consequently, the paper surely deserves the Best Paper Award of IEICE.