Wipro Limited
System and method of data cleansing for improved data classification

Abstract:

A system and method for data cleansing are disclosed. The method includes receiving by a data classifier, a data corpus comprising one or more words. The method further includes grouping the one or more words into a plurality of pre-defined categories and determining an overlap ratio between each pair of the plurality of pre-defined categories based on a comparison between each of the plurality of pre-defined categories. A normalized overlap ratio is computed for each of the plurality of pre-defined categories based on a combination of each overlap ratio. At least one of the plurality of pre-defined categories is determined based on a comparison between each of the normalized overlap ratio and an ideal overlap ratio. Finally, one or more common words from the selected at least one of the plurality of pre-defined categories is removed based on a frequency of occurrence of the one or more common words.