Unsupervised Extraction of Graph-stream Structure for Purpose of Knowledge Retrieval and Information Fusion
Radosław Ziembiński
DOI: http://dx.doi.org/10.154392015288
Citation: Position Papers of the 2015 Federated Conference on Computer Science and Information Systems, M. Ganzha, L. Maciaszek, M. Paprzycki (eds). ACSIS, Vol. 6, pages 53–60 (2015)
Abstract. Big Data processing is in the age of Internet of Things one of the top research area. Inevitable introduction of various kinds of sensors to our life resulted in production of huge amount of data in a form of streams. In that world, an input to the knowledge retrieval process become available only by a short time. Additionally, it is often complex and encumbered by errors related to acquisition and transmission. These properties are even more evident if the sensor is simple but observed process is complex and contains many simultaneous threads. Then, improper acquisition of information may lead to errors caused by mixed data describing different processes threads. Some remedy may come from a proper representation of information collected by sensors. For this purpose, this paper introduces a graph-stream structure representing performance of complex multi-threaded process. The proposed network representation can separate information describing multiple threads and allows for modeling causal relationships between them. It gives separated and segregated information opening opportunity for development of qualitatively better and simpler knowledge retrieval algorithms. Further, the paper delivers a method for this representation extraction from multivariate data stream. It would be done by a clustering algorithm particularly designed for this purpose and evaluated quantitatively and qualitatively on example sets of data.