THESIS
2015
viii, 43 pages : illustrations ; 30 cm
Abstract
Spreadsheets are error-prone. Although various techniques are proposed to detect errors in
terms of smells, they suffer from two issues. First, they cannot uniformly characterize and
detect smells. Each technique targets some specific smell types, and fails to leverage
information derived by previous works to improve detection accuracy. Second, smells are
often detected as violations of pre-defined rules, thus failing to adapt to diverse user practices.
In this thesis, we propose to derive cell clusters automatically using a two-stage technique
based on strong and weak features that capture different user practices. Smells can then be
detected as outliers of these clusters in feature space. We implemented our technique and
applied it to 70 spreadsheet files randomly sampled from...[
Read more ]
Spreadsheets are error-prone. Although various techniques are proposed to detect errors in
terms of smells, they suffer from two issues. First, they cannot uniformly characterize and
detect smells. Each technique targets some specific smell types, and fails to leverage
information derived by previous works to improve detection accuracy. Second, smells are
often detected as violations of pre-defined rules, thus failing to adapt to diverse user practices.
In this thesis, we propose to derive cell clusters automatically using a two-stage technique
based on strong and weak features that capture different user practices. Smells can then be
detected as outliers of these clusters in feature space. We implemented our technique and
applied it to 70 spreadsheet files randomly sampled from EUSES Corpus. Experiment results
show that our technique is effective to cluster cells and capable of detecting multiple types of
smells with a precision 0.72, recall 0.70, F-measure 0.71 compared with existing work 0.60,
0.51, 0.55 respectively.
Post a Comment