Article Preview
TopIntroduction And Background
Given the problems faced by the supervised classification such as the need of many human, the initial classification should be reviewed when the number of documents increases, according to (Tan, 1999) where about 80% of documents are in text format. This huge volume of unstructured or semi structured gives rise to an act to find relevant information more difficult to achieve, that creates a problem known as the problem of information overload (Chen, 1997). The techniques and tools for knowledge discovery in texts (KDT; Feldman, 1995) or simply text mining (Tan. 1999) are being developed to address this problem. One such technique is clustering, a technique for grouping similar documents of a given collection by helping to understand its contents (Jain, 1999; Willet, 1988). One of his goals is to similar documents in the same group and placing documents in various different groups. The assumption is that through a process of clustering, similar objects remain in the same group based on the attributes they have in common. This assumption is known as the hypothesis of cluster described by Rijsbergen (1979).
In this paper, we introduced a new model from nature in this case the social spiders to solve a critical problem of data mining or text mining more precisely what the clustering of data by which we hope to contribute about solving this problem because the grouping or clustering of textual documents, especially Web pages, is one of the challenges of current research.
State Of The Art
A well designed clustering algorithm generally follows the four phases of design: data representation, modeling, optimization, and validation (Buhmann, 2003). Phase representation of the data structures predetermine what kind of cluster can be found in the data. Based on data representation, the modeling phase defines clusters and the criteria that separate the desired group structures of these unwanted or unfavorable. In this phase, a measure of quality, which can be either optimized or approximate when searching hidden structures in the data is produced (see Figure 1).
Figure 1. The biomimetic algorithms that have been solving the problem of clustering