Multi-Terms Association Approach to Generate Patterns for Efficient Data Categorization for Web Information Mining
The demand for heterogeneous data classification for text, image, music, movie, and medical datasets is growing in real-world applications. The complexity of learning classes for a single object associated with multiple term sets is a key issue for multi-term data sets. Existing methods learn based on the feature discrimination observed for similar term sets, but the discrimination measures the deviation of class values rather than associations. Such a method may not be suitable for classification, because each term contains a specific feature.
This paper proposes a Multi-Term Association (MTA) Approach that uses term features to describe and use association rules to discover term correlations for data classification. The MTA aims to find accurate classes for data objects using a "knowledge class" structure that suggests binary associations between them to build terminology patterns to handle multi-term database classification. A comprehensive experimental analysis was performed on a set of MULAN datasets to verify the efficiency of MTA relative to other recent classifiers. The analysis of the results indicates that improvisation has been done in different case studies for accurate measurement.
Keywords: Multi-Terms, Association Rules, Pattern, Categorization, Information mining.