國立虎尾科技大學 |

Effective Record Linkage Techniques for Complex Population Data.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Effective Record Linkage Techniques for Complex Population Data./
Author:	Nanayakkara, Charini.
Description:	1 online resource (229 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 84-03, Section: B.
Contained By:	Dissertations Abstracts International84-03B.
Subject:	Active learning. -
Online resource:	click for full text (PQDT)
ISBN:	9798845456649

Effective Record Linkage Techniques for Complex Population Data.
Nanayakkara, Charini.

Effective Record Linkage Techniques for Complex Population Data. - 1 online resource (229 pages)

Source: Dissertations Abstracts International, Volume: 84-03, Section: B.

Thesis (Ph.D.)--The Australian National University (Australia), 2022.

Includes bibliographical references

Real-world data sets are generally of limited value when analysed on their own, whereas the true potential of data can be exploited only when two or more data sets are linked to analyse patterns across records. A classic example is the need for merging medical records with travel data for effective surveillance and management of pandemics such as COVID-19 by tracing points of contacts of infected individuals. Therefore, Record Linkage (RL), which is the process of identifying records that refer to the same entity, is an area of data science that is of paramount importance in the quest for making informed decisions based on the plethora of information available in the modern world. Two of the primary concerns of RL are obtaining linkage results of high quality, and maximising efficiency. Furthermore, the lack of ground-truth data in the form of known matches and non-matches, and the privacy concerns involved in linking sensitive data have hindered the application of RL in real-world projects. In traditional RL, methods such as blocking and indexing are generally applied to improve efficiency by reducing the number of record pairs that need to be compared. Once the record pairs retained from blocking are compared, certain classification methods are employed to separate matches from non-matches. Thus, the general RL process comprises of blocking, comparison, classification, and finally evaluation to assess how well a linkage program has performed. In this thesis we initially provide a holistic understanding of the background of RL, and then conduct an extensive literature review of the state-of-the-art techniques applied in RL to identify current research gaps. Next, we present our initial contribution of incorporating data characteristics, such as temporal and geographic information with unsupervised clustering, which achieves significant improvements in precision (more than 16%), at the cost of minor reduction in recall (less than 2.5%) when they are applied on real-world data sets compared to using regular unsupervised clustering. We then present a novel active learning-based method to filter record pairs subsequent to the record pair comparison step to improve the efficiency of the RL process. Furthermore, we develop a novel active learning-based classification technique for RL which allows to obtain high quality linkage results with limited ground-truth data. Even though semi-supervised learning techniques such as active learning methods have already been proposed in the context of RL, this is a relatively novel paradigm which is worthy of further exploration. We experimentally show more than 35% improvement in clustering efficiency with the application of our proposed filtering approach; and linkage quality on par with or exceeding existing active learning-based classification methods, compared to our active learning-based classification technique. Existing RL evaluation measures such as precision and recall evaluate the classification outcome of record pairs, which can cause ambiguity when applied in the group RL context. We therefore propose a more robust RL evaluation measure which evaluates linkage quality based on how individual records have been assigned to clusters rather than considering record pairs. Next, we propose a novel graph anonymisation technique that extends the literature by introducing methods of anonymising data to be linked in a human interpretable manner, without compromising structure and interpretability of the data as with existing state-of-the-art anonymisation approaches.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798845456649Subjects--Topical Terms:

555387
Active learning.
Index Terms--Genre/Form:

554714
Electronic books.

Effective Record Linkage Techniques for Complex Population Data.
LDR:04833ntm a22003497 4500 001 1147231
005 20240909100731.5
006 m o d
007 cr bn ---uuuuu
008 250605s2022 xx obm 000 0 eng d
020 $a 9798845456649
035 $a (MiAaPQ)AAI29289056
035 $a (MiAaPQ)AustNatlU1885264165
035 $a AAI29289056
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Nanayakkara, Charini. $3 1472901
245 1 0 $a Effective Record Linkage Techniques for Complex Population Data.
264 0 $c 2022
300 $a 1 online resource (229 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 84-03, Section: B.
500 $a Advisor: Christen, Peter.
502 $a Thesis (Ph.D.)--The Australian National University (Australia), 2022.
504 $a Includes bibliographical references
520 $a Real-world data sets are generally of limited value when analysed on their own, whereas the true potential of data can be exploited only when two or more data sets are linked to analyse patterns across records. A classic example is the need for merging medical records with travel data for effective surveillance and management of pandemics such as COVID-19 by tracing points of contacts of infected individuals. Therefore, Record Linkage (RL), which is the process of identifying records that refer to the same entity, is an area of data science that is of paramount importance in the quest for making informed decisions based on the plethora of information available in the modern world. Two of the primary concerns of RL are obtaining linkage results of high quality, and maximising efficiency. Furthermore, the lack of ground-truth data in the form of known matches and non-matches, and the privacy concerns involved in linking sensitive data have hindered the application of RL in real-world projects. In traditional RL, methods such as blocking and indexing are generally applied to improve efficiency by reducing the number of record pairs that need to be compared. Once the record pairs retained from blocking are compared, certain classification methods are employed to separate matches from non-matches. Thus, the general RL process comprises of blocking, comparison, classification, and finally evaluation to assess how well a linkage program has performed. In this thesis we initially provide a holistic understanding of the background of RL, and then conduct an extensive literature review of the state-of-the-art techniques applied in RL to identify current research gaps. Next, we present our initial contribution of incorporating data characteristics, such as temporal and geographic information with unsupervised clustering, which achieves significant improvements in precision (more than 16%), at the cost of minor reduction in recall (less than 2.5%) when they are applied on real-world data sets compared to using regular unsupervised clustering. We then present a novel active learning-based method to filter record pairs subsequent to the record pair comparison step to improve the efficiency of the RL process. Furthermore, we develop a novel active learning-based classification technique for RL which allows to obtain high quality linkage results with limited ground-truth data. Even though semi-supervised learning techniques such as active learning methods have already been proposed in the context of RL, this is a relatively novel paradigm which is worthy of further exploration. We experimentally show more than 35% improvement in clustering efficiency with the application of our proposed filtering approach; and linkage quality on par with or exceeding existing active learning-based classification methods, compared to our active learning-based classification technique. Existing RL evaluation measures such as precision and recall evaluate the classification outcome of record pairs, which can cause ambiguity when applied in the group RL context. We therefore propose a more robust RL evaluation measure which evaluates linkage quality based on how individual records have been assigned to clusters rather than considering record pairs. Next, we propose a novel graph anonymisation technique that extends the literature by introducing methods of anonymising data to be linked in a human interpretable manner, without compromising structure and interpretability of the data as with existing state-of-the-art anonymisation approaches.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Active learning. $3 555387
650 4 $a Data science. $3 1472902
650 4 $a Clustering. $3 1413637
650 4 $a Education. $3 555912
650 4 $a Bibliographic records. $3 1472903
650 4 $a Markov analysis. $3 1413647
650 4 $a Pedagogy. $3 1148703
655 7 $a Electronic books. $2 local $3 554714
690 $a 0515
690 $a 0796
690 $a 0456
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a The Australian National University (Australia). $3 1186624
773 0 $t Dissertations Abstracts International $g 84-03B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29289056 $z click for full text (PQDT)