SPARCL: An Improved Approach for Matching Sinhalese Words and Names in Record Clustering and Linkage

Quality of data residing in a database gets degraded and leads to misinterpretation due to a multitude of factors. Such factors vary from poor database design, lack of standards for recording database fields to typing mistakes (lexicographical errors, character transpositions). In such a case it is important to identify duplicates and merge them into a single entity. In doing so, one problem that arises is, the way in which string attributes are to be compared. Even though there are different methods in the literature that address the issue of approximate string matching, they all fall short in terms of accuracy when encountered with words from the Sinhalese language written in English. In this paper, it is intended to propose the development of an improved phonetic matching algorithm which improved the accuracy of approximate string matching remarkably. This modified algorithm outperforms the phonetic matching algorithms available in the literature, when applied on datasets containing Sinhalese names and words written in English. In addition, it demonstrates a computational time comparable with phonetic matching algorithms available in the literature. Thus, the modified algorithm which we name “SPARCL” outperforms other phonetic matching algorithms and is illustrated with a real life application.