Chapter Preview
TopBackground
There already exists a plethora of interesting works that have been published to address data streams in the data mining community. These proposals have tried to adapt traditional data mining technologies to the data stream model. Clustering, as one of the most important parts of data mining process, has also gained research attention, regarding its usefulness when processing data streams. Data stream clustering is usually studied with the objective of minimizing the memory space, and processing time required. For such algorithmic designs use of single-pass algorithms that consume a small amount of memory is critical (Antonellis, Makris, & Tsirakis, 2009). Moreover, the real-time requirements and the evolving nature of data streams makes effective clustering a challenging research problem. There are many approaches in the literature, some of them trying to extend the classical clustering algorithms, such as k-median and k-means to data stream applications and some others trying to process the data streams using novel techniques, tailored to the challenges of data streams
An interesting variant of the problem of data stream clustering is when dealing with text streams. This version of the problem has a number of interesting applications such as topic detection and tracking, user characterized recommendation, trend analysis etc. Text stream analysis has attracted great interest and is of great practical value. Clustering text data streams is an important part of the data mining process and appears to have several practical applications such as news group filtering, text crawling, document organization and topic detection. There are many approaches in the literature trying to produce better results from the existing ones like adaptive feature selection (Gong, Zeng, & Zhang, 2011), efficient streaming text clustering (Zhong, 2005), topic models over text streams (Banerjee & Basu, 2007), categorical data stream clustering (Aggarwal & Yu, 2010) and semantic smoothing models (Liu, Cai, Yin, & Wai-Chee Fu, 2007; Xiaodan, Zhou, & Hu, 2006; Zhou, Hu, Zhang, Lin, & Song, 2006).
Key Terms in this Chapter
Web Mining: Is the application of data mining techniques (association rules finding, clustering, classification etc.) to web data, in order to discover useful patterns. According to analysis targets, web mining can be divided into three different types, which are web usage mining, web content mining and web structure mining, and an emerging area web opinion mining.
Clustering: Clustering is the process where given a set of objects and a similarity or distance function between pairs of them, we can partition these objects into groups such that similar objects are grouped together while dissimilar are under different groups, The groups are called clusters.
Text Streams: Text streams are a special kind of data streams where the data are text documents that are coming in a continuous fashion.
Data Mining: The nontrivial extraction of implicit, previously unknown, and potentially useful information, in the form of useful patterns from data. It is also known under the name of knowledge extraction from large databases, though the two notions are sometimes delicately separated; knowledge discovery usually refers to more formal methods of extracting knowledge.
Data Streams: A continuous byte-by-byte flow of data, that can represent different kinds of information such as computer network traffic, phone conversations, ATM transactions, web searches, and sensor data. A data stream can be distinguished in practice from a block transfer, although the moving of blocks could itself be considered a “stream” (of coarser granularity). More abstractly a data stream is an ordered sequence of instances that can be read only once or a small number of times using limited computing and storage capabilities.
Secondary Memory Algorithms: Algorithms for handling information on secondary media, like hard disks, CD-ROMs, etc., and try to minimize the number of page accesses in them.