초록

Abstract ▼

Summary of the report
The Korea Institute of Science & Technology Information (KISTI) has been carrying out data collection and analysis on publications across all scientific domains at the national level in Korea. Their goal has been to create added value from this data for science and technolog

Summary of the report
The Korea Institute of Science & Technology Information (KISTI) has been carrying out data collection and analysis on publications across all scientific domains at the national level in Korea. Their goal has been to create added value from this data for science and technology originating from scholars in Korea – in terms of new knowledge, new scientific theories, and novel and usable technology.
As the focus of the academic information service paradigm shifts from keywords or text-based sources to scientist (i.e. the agents of science), methods and technologies are needed that enable the construction and analysis of social networks of scientists. Analysis of such academic communities can be based on authorship data. The practical objectives for the project on that we report herein were (1) to obtain a deeper understanding of mechanism of scientific collaboration among scholars based on authorship networks from KISTI’s authorship data and (2) to investigate how data quality in terms of author name disambiguation can affect research findings.
For this purpose, we constructed coauthorship networks among Korean scholars and traced the collaboration formation pattern over 20 years. Specifically, we focused on “triadic closure.” In collaboration network studies, a network structure of three nodes being fully connected has received special attention from network scholars as it may indicate that two authors who have not collaborated previously but shared a third author tend to collaborate later.
- Using four large-scale coauthorship networks including that of KISTI, we show that such a tendency, named “triadic closure,” can be represented in different ways for the same dataset depending on what measures are used: Newman (2001b)’s in one-mode networks (NCC), Opsahl(2013)’s in two-mode networks (OCC), and Time-Considered measure (TCC) in two-mode networks.
- Results show that triadic closure can be found to be prevalent by NCC or OCC, e.g., between 25% and 40%, while, by TCC, it rarely happens, e.g., usually between the lower bound of 1~3% and the upper bound of 4~7% in each dataset. Although Korean scholars were shown to have a slightly higher tendency to form triadic closure than others in field-level data, the ratio of triadic closure was still very low.
- These findings suggests that, depending on measures, follow-up research questions need to be framed from a different angle: according TCC, scientists can be better modeled to gain competitive advantage from bridging others who do not collaborated with each other. Another important issues related collaboration network studies is that author identities in collaboration networks can be merged or split if their identities are only indexed by names. Two authors may share the same name and be regarded to be the same person (merging).An author may use several name variants,which can lead us to regarding each name variant to represent different author identities (splitting). In the second part of our research, we show how name disambiguation in social network data can misrepresent network properties.
- For this, we use two datasets: an email exchange network extracted from Enron email corpus and a collaboration network of scholars in brain research constructed from MEDLINE.
- Results show that “organizational communication networks appear to be less coherent and integrated as well as bigger than they truly are,” if the email data is not properly de-duplicated(Jana Diesner, Evans, & Kim, 2015).
- Also, a collaboration network can be interpreted to be “more dense and cohesive than it really is and individual authors appear to be more productive, collaborative, and diversified than they actually are,” when the network dataset was incorrectly disambiguated by first or all given name initials (Jana Diesner et al., 2015).
- In both communication and collaboration networks, erroneous entity disambiguation are shown to lead to a false positive finding of a power law distribution in degree centrality of nodes.
The aforementioned findings can lead us to extend our understanding on data quality and its impact on findings from big data. In the third part of our research, we review studies that address the issue of how data quality affect knowledge discovery from data. The review suggests that “scrutinizing big social data for accuracy and integrity issues, systematically fixing them, and diligently documenting these processes can have another positive side effect beyond boosting the reliability of results and reusability of material: once we start to closely interact with the data, thereby forcing ourselves to understand their idiosyncrasies and patterns, can help us to move us from being able to precisely model and formally describe effects in society to also understand and explain them” (Jana Diesner, forthcoming)
We believe our project not only provides practical insights into the mechanism of scientific collaboration at a national scale, but also shows the importance of data quality control in network data, which highlights why the authority control of KISTI on authorship data is meaningful and should be maintained.

목차 Contents

• COVER ... 1
• Table of Contents ... 2
• Summary of the report ... 3
• Part1: A Time-Sensitive Analysis of Triadic Closure in Scientific Collaboration ... 5
• 1. Introduction and background ... 6
• 2. Methodology ... 9
• 3. Analysis ... 12
• 4. Conclusion and Discussion ... 19
• Acknowledgements ... 20
• References ... 20
• Part2: Impact of Entity Disambiguation Errors on Social Network Properties ... 24
• Introduction ... 25
• Background ... 26
• Data and Disambiguation ... 26
• Metrics ... 31
• Results ... 33
• Discussion and Conclusions ... 38
• Acknowledgment ... 39
• References ... 39
• Part3: Small Decisions with Big Impact for Data Analytics ... 42
• 1. Introduction ... 43
• 2. Accuracy of Social Interaction Data ... 45
• 3. Accuracy of Auxiliary Lexical Resources ... 46
• 4. Conclusions ... 47
• Acknowledgement ... 47
• References ... 47
• End of Page ... 50