This paper proposes a procedure that easily extracts a feature that helps differentiate between similar researcher names in articles. We examined email patterns and their timelines to identify researchers. Our statistical analysis results show multiple email address usage patterns are found in the case of approximately 43% researchers, and 5% of the patterns are overlapped. Base on the statistics, we conclude that the identification of researchers is still required to enhance performance of the researcher-centric analytics systems and applications.
With ever-increasing amounts of research data and advancements in technology in
big-data environments, a paradigm shift is required. Accordingly, studies on new
business intelligence services are being conducted and forecasting and analysis
methods are being developed. Prescriptive analytics first appeared in 2013 among
several analytical methods and offers diverse strategies for achieving the objectives of
and improving business competence. The 2014 Gartner Hype Cycle Special Report
predicted that prescriptive analytics will advance rapidly and reach a technology
maturity stage within the next ten years1). InSciTe Advisory is a service developed in
2013 for strengthening researcher research skills by using the 5W1H method with
prescriptive analytics [
The amount of academic literature published on the World Wide Web is
evergrowing [
Figure 1 shows an example of researchers who can have the same name. The
authors might be divied two different researchers or might be merged as a researcher.
Ensuring the accuracy of classification is difficult when the network automatically
classifies Researcher 1 as the same person associated with data collected on specific
research papers. Therefore, methods for automatically identifying researchers are
necessary when largescale literature data is considered. In addition, a correct answer
set with high accuracy and an experimental data set are required when researchers
conduct studies based on researcher data. To achieve this, accurate identification of a
researcher is required for certain works which are part of researcher data. Therefore,
currently operating authentication services such as Elsevier SciVal Expert and
ORCID are designed so that researchers can provide relevant information directly and
manage it by themselves [
Therefore, studies on researcher identification based on researcher meta
information extracted from papers and scientific literature data have been proposed.
Studies exist that examine the use of researcher email and affiliation information. The
study in [
In this paper, we introduce the extracting procedure for certain features, such as email address and affiliation name, which can play an important role in identifying a researcher. Further, we propose an analysis method based on the timeline, and state our experimental results.
This chapter explains the feature extraction procedure from researcher data.
The feature extraction procedure involves four stages as shown Figure 2. The first stage involves the collection of researcher data; we collect the meta information of the published papers that are on the web to identify disambiguation of a researcher’s name. To do so, we collect the Digital Object Identifiers (DOIs) of the papers; using these DOIs, we collect the published information on the pertinent sites. Since the websites where a paper is published are structured in different forms, we develop customized crawlers to collect data, taking into consideration the structure of each web page. The second stage is the feature extracting stage; email addresses of researchers are extracted. During this stage, the year when an email address was generated is extracted together with the email address to obtain the timeline information for the email addresses extracted. The third stage is the refining stage; during this stage, we remove the unnecessary data that may exist in the pertinent feature. For example, in addition to an affiliation name, the address and postal code of that organization are mentioned in some scientific papers; the same organization may be mistaken as different organizations owing to a different address or postal code mentioned in the papers, and therefore, such unnecessary information is removed. The last stage is the pattern extraction stage; during this stage, the unique pattern of a researcher is derived from the extracted pattern information, and this unique pattern can be used to accurately identify a researcher. the usage period of the email address can be defined. To estimate this period, the start and end time of a particular email address in use are defined as the time of first usage and the time of the latest usage (the last appearance of the pertinent email address), respectively. For example, more than 2 email addresses appeared for a particular researcher, and the periods during which each email address was used constitute a coprime relationship, i.e., the usage period for both the email addresses do not overlap like Case 1 as shown in Figure 3. The case 2 in Figure 3 shows that email addresses of researchers with the same name are different from each other but the appearance periods are overlapped; in the case, it is necessary to identify the researchers as same or different because it is possible that they are actually different researchers although their names are the same.
Metadata (title of the paper, the publishing year, coauthors, DOI, etc.) of a paper, as well as the researcher’s name, are included in the researcher data extracted by DBLP. However, it does not include the email address and affiliation information of a researcher. Therefore, we implemented an experimental data set according to the procedure explained in Section 3.1. The number of researchers present on DBLP as of September, 2014 is about 1,465,700, and the number of papers is about 4,122,000. To obtain an experiment data set from the pertinent data, we collected website contents using the DOI of papers. The email information is collected automatically by a crawler, so the email addresses of all authors are collected. Therefore, it is not necessary that the n-th email address is the n-th author’s email address. Thus, we considered only those email addresses for our experimental data, which had the number of authors equal to the number of email addresses. As a result, 64,802 researchers were extracted for the experimental data. To extract the first author, we compared the pertinent researcher’s name with the co-author list, located it at the first instance, and deleted the overlapped name; finally an experimental data set including 18,867 researchers was implemented. We found that these researchers whose names were extracted using the aforementioned process, published 3,790 papers, which is 46.64% of the entire experimental data set. 4.2
Through the statistical analysis of the results, we found that the number of overlapped email addresses whose appearance frequency are more than twice is 5.28%. In the data set, 3,162 researchers published at least two papers and a total of 8,126 papers were published by them. We set the minimum number of paper publications at two as a condition because the number of email addresses can be considered as one when the number of paper published is one. Further, 1,371 authors’ emails appeared at least twice in the data set. Among the researchers, a total of 167 researchers showed the pattern depicted in the right side of Figure 4, and they published 574 papers (7.06% of the entire experimental data set). The result about the overlapped email address is lower than 2% of the result about the overlapped papers.
It means that the productivity of researchers who have overlapped email patterns become high, and we awared that additional methods are needed to classify them because 43% of the researchers who have more than 2 email addresses.
In the big data environment, academic data are generated at a very fast rate. In this vein, researchers need to quickly obtain and accurately grasp the information presented in studies related to their research, the possible co-author network, and the trend in a specific research field to strengthen researchers’ research competence. Accordingly, prescriptive analytics are required for accurate analysis and establishing strategies. However, these services have to be implemented based on accurate data to establish customized strategies for researchers. To this end, the accurate identification of a researcher’s name and improvement in the information credibility with regard to researcher data becomes important.
This paper proposed an extraction procedure for important features from researcher data to identify ambiguous researcher’s names. To improve researcher identification, we defined the email address usage pattern by considering the timeline characteristic of the researcher’s email information and carried out experiments based on the DBLP data set; we verified that our identification method based on email addresses and that considers timeline characteristic is effective, and can be used as an important factor for identifying a researcher. As a future study, we will find a unique pattern representing a researcher by collecting and extracting the affiliation information considering the timeline characteristic from researcher data, and will research on an automated researcher identification system method by applying the obtained pattern to identify researchers; to verify its effectiveness, we will implement an accurately refined data set and compare its performance with the experimental data set.
This work was supported by the IT R&D program of MSIP/KEIT. [2014-044-024002, Developing On-line Open Platform to Provide Local-business Strategy Analysis and User-targeting Visual Advertisement Materials for Micro-enterprise Managers]. 6 7