Discovering and monitoring product features and the opinions on them with OPINSTREAM

Opinion stream mining encompasses methods for monitoring and understanding how people's attitude towards products changes over time. For many applications, though, not only a specific product is of interest but also the properties that people consider important for the whole category of products. Understanding which product features influence a buyer's choice positively or negatively allows decision makers to make well-informed decisions on improving their products or marketing them properly. In this study, we propose OPINSTREAM, a framework for the discovery and polarity monitoring of implicit product features deemed important in the people's reviews on different products. Our framework encompasses stream clustering, extraction of product features from the clusters, cluster adaptation and semi-supervised sentiment learning inside each cluster. These components build upon our earlier work on product feature discovery and monitoring (M. Zimmermann, E. Ntoutsi, Z.F. Siddiqui, M. Spiliopoulou, H.-P. Kriegel, Discovering global and local bursts in a stream of news, in: Proceedings of the 27th Annual ACM Symposium on Applied Computing, SAC'12, ACM, 2012., M. Zimmermann, E. Ntoutsi, M. Spiliopoulou, Extracting opinionated (sub)features from a stream of product reviews, in: Proceedings of the 16th International Conference on Discovery Science (DS'2013), Lecture Noteson Computer Science, vol. 8140, Springer, Singapore, 2013, pp. 340-355., M. Zimmermann, E. Ntoutsi, M. Spiliopoulou,Adaptive semi supervised opinion classifier with for getting mechanism (to appear), in: Proceedings of the 29th Annual ACM Symposium on Applied Computing, SAC'14, ACM, 2014. ), with emphasis on smooth cluster adaptation. We report on the performance of OPINSTREAM on two real datasets with product reviews, whereby we evaluate both the stream clustering approach for product feature monitoring and the semi-supervised polarity monitoring method.