International Journal of Science, Engineering and TechnologyAuthors: Mr. Sakshi Siva Ramakrishna, Dr. T Anuradha
Abstract: Business transactions generate a large amount of data that requires thorough analysis to provide vital insights for businesses' decision-making.A transaction is linked with a good number of record attributes. Traditional transactional data analysis treated all attributes equally, but a subset of attributes can distinguish fine-grained transactions. This discrimination is achievable through attribute and transaction weighting. User surveys or empirical data are the sources for weighting data attributes. If user views are unavailable or the empirical study is missing, weighting becomes tough. Some effective methods exist for weighting. When attributes are binary-valued, the weighting process should rely on transaction-item relationships. The HITS (Hypertext Induced Topic Search) algorithm can perform this weighting. A new algorithm called “Bijective HITS” is proposed, capable of weighing transactions and items by mapping transaction-item relations to item-feature relations. This two-level processcan identify important transactions and items. A new distance measure named “W-distance” is derived from this weighting process. Additionally, a link and density-based hierarchical clustering method is proposed to cluster transaction data using only binary information. Experiments conducted with real and hypothetical datasets compare the results of this approach with those of existing well-known methods. The findings indicate that the proposed models outperform the compared processes, offering better tools for implementing recommendation systems.
