(A)generalization of band joins and the merge/purge problem

목차 (Table of Contents)

• CONTENTS
• Table of Contents = ⅰ
• List of Figures = ⅳ
• List of Tables = ⅶ
• 1 Introduction = 1
• 2 Previous Work = 8
• 2.1 Record Linkage = 9
• 2.2 Heterogenous Multi-Databases = 11
• 2.3 Sorting with Duplicate Elimination = 14
• 2.4 Band-Joins = 15
• 3 The Merge/Purge Problem = 20
• 3.1 The Sorted Neighborhood Method = 21
• 3.1.1 The naive Sorted-Neighborhood Method = 24
• 3.1.2 The Duplicate Elimination sorted-Neighborhood Method = 27
• 3.2 Clustering the data first = 31
• 3.2.1 Alternative Clustering Strategies = 33
• 3.3 Equational theory = 34
• 3.4 Computing the transitive closure over the results of independent runs = 38
• 3.5 A General-Purpose Merge/Purge Engine = 42
• 4 Experimental Results = 45
• 4.1 Generating the databases = 45
• 4.2 Pre-processing the generated database = 47
• 4.3 Initial results on accuracy = 49
• 4.4 The Duplicate Elimination Method = 52
• 4.5 The Clustering Method = 57
• 4.6 Analysis = 60
• 5 Parallel Implementation = 67
• 5.1 Single and Multi-pass sorted-neighborhood method = 67
• 5.2 Single and Multi-pass clustering method = 72
• 5.3 Scaling Up = 74
• 6 A Real-Word Test Case = 76
• 6.1 Database Description = 77
• 6.2 using Merge/Purge over OCAR's data = 79
• 6.3 A Purge Application = 84
• 6.4 Results over the entire OCAR database = 90
• 7 Optimizations = 96
• 7.1 Link Minimization = 96
• 7.2 A State-Saving Multi-pass approach = 103
• 7.2.1 Experimental Results = 107
• 7.3 Incremental Merge / Purge = 113
• 7.3.1 Initial experimental results = 117
• 8 Conclusion = 121
• 8.1 Summary = 121
• 8.2 Contributions = 124
• 8.3 Future Work = 125
• Bibliography = 127
• A Person Database Programs = 135
• A.1 Original OPS5 version of the equational theory = 135
• A.2 Schema Definition = 140
• A.3 Pre-processing Code = 141
• A.4 Equational Theory Code = 142
• B. OCAR's Database Programs = 144
• B.1 Schema Definition = 144
• B.2 Pre-processing Code = 145
• B.3 Equational Theory Code = 147
• B.4 purge Rules = 154
• C. Accuracy Results = 156
• C.1 Number of Duplicates per Record = 157
• C.2 Size of Input Database = 160
• C.3 Level of noise = 162
• D. Relevant Communications = 166
• D.1 Timothy Clark's request in KDD-Nuggets 95 : 7 = 166
• D.2 Data Cleaner's performance assessment over OCAR's data = 168
• D.3 Gratitute letter from OCAR = 170