Calculating Similarity of Arbitrary Reports

Abstract Reporting is an essential part of business, today, and people spend a lot of time creating meaningful visualizations of their most important data. Surprisingly, the reuse of reports (i.e., applying the same visualization or query on di↵erent data) is not common. The recommendation of proven, existing queries represents one part of this reuse. Since there are several report formats and the reports target di↵erent data sources, the task of matching report queries for recommendation is very complex and has not been addressed, yet. Recent works that aim at query recommendation – usually to support the user in database add-ons – focus on collaborative approaches or target query completion. The idea behind this study is to follow a content-based, combined approach to match report queries. The use of both an abstract and a concrete representation of the queries allows the application of di↵erent, well-known matching techniques in parallel. The focus of this work is to evaluate the impact of similarity search and schema matching for matching queries. For that, query matching algorithms based on the two techniques are developed: an e cient, index-based comparison using similarity search and a fine-grained matching of the parse trees of the queries with schema matchers. Next to the presentation of an e↵ective combination of those algorithms, a major contribution of this thesis is the creation of an empirical data set of 150 queries with the corresponding similarity ratings for all 22,500 query pairs for a comprehensive evaluation.Reporting is an essential part of business, today, and people spend a lot of time creating meaningful visualizations of their most important data. Surprisingly, the reuse of reports (i.e., applying the same visualization or query on di↵erent data) is not common. The recommendation of proven, existing queries represents one part of this reuse. Since there are several report formats and the reports target di↵erent data sources, the task of matching report queries for recommendation is very complex and has not been addressed, yet. Recent works that aim at query recommendation – usually to support the user in database add-ons – focus on collaborative approaches or target query completion. The idea behind this study is to follow a content-based, combined approach to match report queries. The use of both an abstract and a concrete representation of the queries allows the application of di↵erent, well-known matching techniques in parallel. The focus of this work is to evaluate the impact of similarity search and schema matching for matching queries. For that, query matching algorithms based on the two techniques are developed: an e cient, index-based comparison using similarity search and a fine-grained matching of the parse trees of the queries with schema matchers. Next to the presentation of an e↵ective combination of those algorithms, a major contribution of this thesis is the creation of an empirical data set of 150 queries with the corresponding similarity ratings for all 22,500 query pairs for a comprehensive evaluation.