Enhanced Self Organizing Map Alogrithm for Web Usage Mining through Neural Network

Abstract

Web mining is an essential part of data mining. Web mining adopts a great part of the data mining mechanisms to discover potentially useful information. Web mining analysis depends on three common set of information such as patterns, shared content and inter-memory association link structure relating to three subsets in web mining namely Web usage mining, Web content mining, Web structure mining respectively. Data grouping or clustering is a standard mechanism for statistical data analysis, which is utilized as a part of numerous fields, consisting machine learning, data mining, pattern recognition, image analysis, and bioinformatics. newlineClustering or Grouping aims to discover essential structures in data or document and arrange them into vital subgroups for further study and analysis. Existing techniques greedily select the following frequent item set which illustrates the following group to constrain the covering among the documents or data that comprise both the item set and some remaining item sets. As it were, the grouping or clustering outcome depends on the demand of grabbing the item sets, which in turns based on the avaricious heuristic. The technique does not take after a subsequent request of selecting groups or clusters. newlineTo overcome the above issues, a novel approach Enhanced Self-Organizing Map (ESOM) is proposed for document clustering which offers highest effectiveness and performance. The proposed system is estimating similarity between documents or data and subsequently formulates a new criterion functions for a document or data clustering. The principle of this analysis is to verify how much a data or document similarity measure overlaps with the real class labels and investigate useful similarity measure for data clustering. The ESOM algorithm is mainly focused on analyzing and generating usage of cluster overlapping phenomenon to plan cluster-integrating criteria. The system improves the effectiveness of clustering and saves computing time of clustering process. newlineiv newlineThe ESOM algorithm automatically (self-organizing) clusters documents or data for large scale of data. The ESOM clustering process groups a data over various levels by generating a cluster tree. The ESOM method identifies the document or data similarities among each pair of vectors in the clustering process. The ESOM method improves 0.304 F-measure, 1.05 the data or document similarity prediction, and minimizes entropy 0.118 for web usage log files. newline