Understanding the Advisor-advisee Relationship via Scholarly Data Analysis
Jiaying Liu, Tao Tang, Xiangjie Kong, Amr Tolba, Zafer AL-Makhadmeh, Feng Xia
SScientometrics manuscript No. (will be inserted by the editor)
Understanding the Advisor-advisee Relationship viaScholarly Data Analysis
Jiaying Liu · Tao Tang · Xiangjie Kong · Amr Tolba · Zafer AL-Makhadmeh · Feng Xia
Received: date / Accepted: date
Abstract
Advisor-advisee relationship is important in academic networks dueto its universality and necessity. Despite the increasing desire to analyze the ca-reer of newcomers, however, the outcomes of different collaboration patternsbetween advisors and advisees remain unknown. The purpose of this paperis to find out the correlation between advisors’ academic characteristics andadvisees’ academic performance in Computer Science. Employing both quan-titative and qualitative analysis, we find that with the increase of advisors’academic age, advisees’ performance experiences an initial growth, follows asustaining stage, and finally ends up with a declining trend. We also discoverthe phenomenon that accomplished advisors can bring up skilled advisees. Weexplore the conclusion from two aspects: (1) Advisees mentored by advisorswith high academic level have better academic performance than the rest; (2)Advisors with high academic level can raise their advisees’ h-index ranking.This work provides new insights on promoting our understanding of the rela-tionship between advisors’ academic characteristics and advisees’ performance,as well as on advisor choosing.
Keywords
Academic networks · Scholarly data · Social network analysis · Advisor-advisee relationship · Collaboration patterns
Corresponding author: Xiangjie Kong; E-mail: [email protected]. Liu · F. Xia · X. KongKey Laboratory for Ubiquitous Network and Service Software of Liaoning Province, Schoolof Software, Dalian University of Technology, ChinaT. TangChengdu College, University of Electronic Science and Technology of China, ChinaA. Tolba · Z. AL-MakhadmehComputer Science Department, Community College, King Saud University, Saudi ArabiaA. TolbaMathematics Department, Faculty of Science, Menoufia University, Egypt a r X i v : . [ c s . D L ] A ug Jiaying Liu et al.
Scholarly data analytics has become a vital part of the scientific research (Letch-ford et al., 2015; Lehmann et al., 2006; Gl¨anzel et al., 2017). Relationshipextraction is a major area of interest in the scholarly social network analy-sis. The structure of the network usually contains a large number of nodes,with different relationships such as friendship, kinship, and hostility. Analyzingsuch relations can strengthen our understanding of the evolution and develop-ment of the scholarly society (Xia et al., 2017). One of the main relations is thementorship because mentors are important for their prot´eg´es. Researchers havestudied various aspects of mentorship including types, phases, outcomes andmobilities as early as 30 years ago (Chao et al., 1992; Kram, 1983). Althoughthere is little doubt that the study of academic mentorship is necessary, ithas been neglected because there is no complete advisor-advisee dataset. Eachscholar needs guidance from the advisor when she/he enters academia as anewcomer. It is widely accepted that different advisors influence advisees dif-ferently (Murphy, 2015). However, where these differences are originated andhow significant they are, both are worth exploring. The lack of quantitativeacademic mentorship analysis leads to the striking lack of specific guidingsignificance for advisors choosing.The mentorship is universal and complex. Traditionally, a mentorship isdefined as “A relationship between the inexperienced mentee and an expe-rienced senior member of a particular field” (Dobson, 2013). It has differenttypes and forms in different fields. For instance, it can be regarded as therelationship between employees and managers in public utility companies andadvisor-advisee relationship in academia. There is a large volume of studiesdescribing mentorship in various aspects. According to the formation of therelationship, Chao et al. (1992) first categorized mentorship into two types:formal mentorship and informal mentorship. Informal mentorship usually hasneither compact structure with reasonable management, nor been recognizedby the organization formally. In contrast to informal mentorship, formal men-torship occurs with external intervention from the organization. Usually it isled and managed by the organization. Mentorship is a temporary relationshipwith almost fixed duration, for example, the average length of advisor-adviseerelationship in academic networks is five years (Kram, 1983). The phases ofthe mentorship can be defined as initiation, cultivation, separation, and re-definition on the basis of the particularly effective experience (Kram, 1983).In recent years, there has been an increasing amount of literature describingspecific elements of mentorship, including behaviors of mentors and prot´eg´es,mentoring functions, structure of cooperation in the mentorship (Johnson andRidley, 2015; Bozionelos et al., 2014; Borders et al., 2012), and so on. Thesestudies provide complete insights into the development and implementation ofthe mentorship.No matter what the form is, the ultimate goal of mentorship is to enhancethe personal and professional development of prot´eg´es (Chao, 1997). There-fore, mentors have been taken as beacon lights in the development of prot´eg´es. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 3
Based on inherently dyadic attributes of the mentorship, the existing literaturedemonstrates that mentorship is positive and beneficial for both mentors andprot´eg´es. In a positive mentorship, the prot´eg´es will have better performanceand more opportunities for promotion because their mentors provide themwith personal and career assistance. Meanwhile, the mentors can receive ful-fillment and gain satisfaction by improving prot´eg´es’ welfare (Ghosh and Reio,2013; Hu et al., 2014). Furthermore, organizations benefit as well because thereis a real possibility that prot´eg´es would like to be devoted to their organiza-tions after graduation (Malmgren et al., 2010; Florea et al., 2013). Moreover,as early as 1989, Fagenson (1989) suggested that mentoring can enhance workvalidity and the career success. Singh et al. (2009) perceived to the point thatcompared with those who do not have a mentor, individuals who have obtainmentors embrace more chances to promote in their careers, and more likely toconsider their future prospects ahead of time in more positive ways. Wanberget al. (2006) proposed sometimes a mentor may change a prot´eg´e’s vocationaltendency which can lead the prot´eg´e to the different lifeline.The recent trends in mentorship analysis have led to a proliferation of stud-ies focusing on the association between mentorship and the career developmentof prot´eg´es. Since Kram (1983) presented the psychosocial functions and ca-reer development functions to examine the mentors’ influence on the prot´eg´es,almost every paper focusing on the outcome of mentorship includes contentsrelated to it. The mentoring functions have received considerable critical at-tention. On this basis, Chao et al. (1992) compared the mentoring functionsamong formal mentorships, informal mentorships, and non-mentored counter-parts. Scandura and Ragins (1993) investigated the link between mentoringfunctions and the career mobility outcomes in terms of promotions and salaries.In a study conducted by Singh et al. (2009), it showed the correlation of risingstar attributes measured at different periods, i.e., no-mentored period and ayear after mentored. In the same vein, Chao (1997) examined linkages amongdifferent mentorship phases, functions and outcomes of prot´eg´es in order tointegrate different aspects of mentoring into a more comprehensive theory.According to Young and Perrewe (2000), there were specific behaviors suchas trust, related to career and social support the exhibited throughout thementoring process. Furthermore, Tuesta et al. (2015) found the evidence thatthere exists a positive correlation between time of advisor-advisee relationshipand advisee’s productivity in the area of Exact and Earth Sciences.Although research has been carried out on the mentors’ influence, with theexception of Tuesta et al. (2015) who measured scientific productivity by thenumber of publications in journals, all others (Chao et al., 1992; Scandura andRagins, 1993; Singh et al., 2009; Young and Perrewe, 2000; Chao, 1997) remainnarrowly focus on dealing only with analytic study based on the feedbackfrom the questionnaires. There are few data-based investigations studying theoutcomes of choosing different advisors in academia. This gives rise to thequestion of how to choose an advisor when newcomers enter academia. Inresponse to the question “What kind of advisors you will choose when youdecide to pursue the PhD degree?”, most of the answers are non-specific and
Jiaying Liu et al. all-embracing. Almost all advisees want to choose advisors who make a greatand positive influence on themselves. Young advisees in academia often seek towork in collaboration with top advisors in their field in pursuit of a successfulcareer. However, “top advisors” in academia can be defined differently.In this work, we analyze the relationship between advisees’ performance(i.e., productivity and impact) and advisors’ academic characteristics fromthe perspective of Computer Science. It is different from the existing re-search which only employs questionnaires. The dataset of advisor-advisee re-lationships is extracted from Digital Bibliography & Library Project (DBLP)dataset (Ley, 2009). We use the improved stacked autoencoder method basedon the deep learning with the highest accuracy reaching up to 94% to getthe complete dataset (Wang et al., 2016, 2017). Then we make a quantitativeanalysis of the dataset. We calculate the productivity and impact of adviseesincluding the number of publications, citations, and h-indices in 1-12 yearsafter first collaborating with their advisors. We separate the advisors intogroups according to their academic age to observe the relation between ad-visees’ outcomes and advisors’ academic ages. We find that advisees’ number ofpublications, citations, and h-indices follow the same trend with the increasingof advisors’ academic ages, exhibiting an initial growth, remaining stationaryfor a duration, finally with a declining trend. Based on the observation, weexamine the correlation between the ranking of advisors and their advisees’h-indices.Furthermore, by combining quantitative analysis with qualitative analysis,we find that advisors with high academic levels will bring up advisees withhigh academic performance. We conclude the findings from two main aspects:(1) Advisees mentored by the advisors with high academic level generallyhave better academic performance than the rest. We divide the advisors intotwo groups: one with a high ranking in terms of publications/citations/h-indices and the other without. By comparing the academic performance of alladvisees mentored by different advisors, we observe that the phenomenon alsoholds for advisees with high academic level (if a scholar ranks the top 10%in the number of publications/citations/h-indices, then she/he has the higheracademic achievement). (2) Advisors with high academic level can increase theprobability of their advisees ranking the top 10% in h-index. We calculate theprobability of an advisee ranking the top 10% in different cases. These casesare based on advisors’ different h-indices and academic ages. The outcomesof this quantitative research, which compare advisees’ achievements coachedby advisors with the different academic performance we present here, can beutilized for advisor recommendation. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 5 is a website whichaims to provide comprehensive search and mining services for researcher so-cial networks. It is considered as a widely used and one of the best-curateddatabases for Computer Science articles (Gollapalli et al., 2011; Moreira et al.,2011; Tang et al., 2008; Amjad et al., 2017). Currently, the system consists ofmore than 6,000 conferences, 3,200,000 publications, and 700,000 researcherprofiles before 2016. In this website, developers use Microsoft Graph SearchAPI to query each AMiner paper’s title and obtain candidate matching pa-pers for each AMiner paper. For the computational accuracy, they randomlysampled 100,000 linking pairs and evaluated the matching accuracy. The num-ber of truly matching pairs is 99,699 and the matching accuracy can achieve99.70%.Finally, the analysis is on the grounds of 15,559 advisors whose academicage is 5 (23,473 advisees), 20,859 advisors whose academic age is 10 (36,841advisees), 17,028 advisors whose academic age is 15 (33,652 advisees), 11,522advisors whose academic age is 20 (24,883 advisees), and 7,352 advisors whose academic age is 25 (11,305 advisees). The whole process of the experiment isillustrated in Fig. 1. Shifu: A deep learning based advisor-advisee mining model
Train and optimize the model
Construct collaboration ego-network
Normalize
DBLP
A scholar's publication information Collaborators' publication information
Select the advisor-advisee pairs whose identification accuracy exceed 90%
Combine
Advisor-advisee pairs Personal information
Clean the network Extract personal properties Extract collaboration properties
Advisor-advisee Dataset
AMiner
CitationAuthorPublication
Advisee's name Advisor's name
Match
Fig. 1: Dataset acquisition process.2.2 Ranking the scholarsWe take three variables into consideration to rank a scholar: number of publi-cations, citations, and h-indices. H-index of a scholar is calculated on the basisof the year of publications and referenced information each year. A scholar hasan h-index h if at least h of his/her publications attract h or more citations.In order to measure the advisors’ influence on advisees during different collab-oration periods, the indicators for each year are calculated.2.3 Analysis method When dividing the type of advisors, we take advisors’ academic ages intoconsideration because the academic characteristics such as citations and thenumber of publications are accumulated over time. Therefore, advisors withdifferent academic ages should not be put together for comparison. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 7
Definition 1
AA is defined as the academic age of scholars, i.e., AA = Y c − Y f where Y f is the year scholars published the first article and Y c is the investi-gated year.In this paper, we calculate advisees’ and advisors’ academic ages of theyear in which they began cooperating with each other. In other words, Y c isthe year when they commenced collaboration. In order to observe the association between advisees’ academic performanceand advisors’ academic level, we calculate the possibility that advisees men-tored by different types of advisors with different h-index rankings and aca-demic ages. The steps are described as follows:(1) Divide the advisors according to the academic age when they first collab-orated with their advisees. For advisors with the same academic age, dividethem into
T op group in which advisors’ h-index ranking is in the top 10%and Res group for the rest.(2) Calculate the total number of
T op advisors’ advisees N t t and Res advi-sors’ advisees
N t r .(3) Rank all the advisees with the same academic age according to their h-indices.(4) Calculate the minimum of the top 10% advisees’ h-index h min .(5) Calculate the number of T op advisors’ advisees N h t whose h-indices areabove h min . In the same way, calculate N h r of Res advisors.(6) The probability of
T op advisors’ advisees whose h-index rank top 10%can be calculated as N h t /N t t and N h r /N t r for Res advisors.
In this section, we present the basic analysis of the advisor-advisee datasetin subsection 3.1. Subsection 3.2 elaborates the correlation between advisors’academic characteristics and advisees’ academic performance. Subsection 3.3shows the phenomenon that an accomplished advisor could bring up a skilledadvisee. Finally, the analysis of the increasing publication rate is highlightedin subsection 3.4.3.1 Statistical analysis of the advisor-advisee relationshipsTable 1 summarizes the statistics of the advisors at the beginning of col-laborating with their advisees. It illustrates the mean of advisors’ academiccharacteristics including the number of publications (
N P ), citations (
N C ), Jiaying Liu et al. and h-indices (
N H ). Here we separate advisors according to their exact aca-demic age ( AA ) when they begin guiding the advisees, to mitigate the bias ofaccumulated citations for advisors with different AA .Table 1: Academic characteristics of advisors Attributes AA = 5 AA = 10 AA = 15 AA = 20 AA = 25 Average NP NC NH In order to carry out the data-based analysis of the advisor-advisee rela-tionships, we keep a tally of relevant information of advisors and advisees inthe dataset. Fig. 2 shows the range of each personal characteristic and numberof advisors/advisees distribution of these characteristics.The red points represent advisors and blue ones represent advisees. Fig.2(a) shows the academic age distribution of advisors and advisees since theirfirst collaboration. As shown in Fig. 2(a), most advisors’ AA are 5-20 years.After that, as AA increases, the number of advisors decreases. Fig. 2(c), Fig.2(e), and Fig. 2(g), respectively illustrate the distributions of advisees (10 yearsafter first collaborating with their advisors) and advisors (when they beganmentoring a certain advisee) in terms of N P , N C , and
N H . We only considerthe advisees whose career spans at least 10 years since they first collaboratedwith their advisors. It can be found from Fig. 2(c), Fig. 2(e), and Fig. 2(g), thatthere is a clear decreasing trend in the number of scholars with the increasingof
N P , N C , and
N H . We also plot the survival function, that is P ( x > X )for these graphs, which will get rid of the noise and observe the differencesbetween various groups. The results are presented in Fig. 2(b), Fig. 2(d), Fig.2(f), and Fig. 2(h). In these figures, the horizontal axis represents scholars’ AA , N P , N C , and
N H , respectively. The vertical axis is the survival rate.These survival curves have a declining trend. The steeper decline slope is onbehalf of the lower survival rate. From these figures, we can see that all curvesare concave. It illustrates that most scholars have relatively lower academicperformance, few scholars have higher academic achievements, which can beevidenced by “Long Tail Effect” (Anderson, 2006).Graphs in Fig. 3 are the results obtained from fitting methods for Fig. 2.Specific fitting functions are summarised in Table 2. In Table 2, y represents thenumber of advisees/advisors and x represents AA / N P / N C / N H . In order tocarry out effective fitting, we use different exponential functions to determinethe relationship. In fact, the existing literature (Newman, 2001) has pointedout that the degree distribution of the scientists’ networks is between theexponential and the power-law. Obviously, our results are in consistent withaforementioned law. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 9
A A number of advisors/advisees a d v i s o r s a d v i s e e s (a) Academic ages surviving proportion
A A a d v i s o r s a d v i s e e s (b) Survival function for AA advisors advisees nu m b e r o f a d v i s o r s / a d v i sees NP (c) Number of publications a d v i s o r s a d v i s e e s surviving proportion N P (d) Survival function for NP advisors advisees nu m b e r o f a d v i s o r s / a d v i sees NC (e) Citations N C surviving proportion a d v i s o r s a d v i s e e s (f) Survival function for NC advisors advisees nu m b e r o f a d v i s o r s / a d v i sees NH (g) H-indices a d v i s o r s a d v i s e e s surviving proportion N H (h) Survival function for NH Fig. 2: Academic performance distribution and survival function foradvisors and advisees. The horizontal axis represents (a) academicages, (c) the number of publications, (e) citations, and (g) h-indices,respectively. The vertical axis represents the number of scholars. In(b), (d), (f), and (h), the vertical axis represents the surviving pro-portion. number of advisees
A A a d v i s e e s E x p o n e n t i a l f i t o f a d v i s e e s (a) AA of advisees a d v i s o r s E x p 3 P 2 f i t o f a d v i s o r s number of advisors A A (b) AA of advisors number of advisees N P a d v i s e e s E x p 2 P M o d 1 f i t o f a d v i s e e s (c) NP of advisees number of advisors N P a d v i s o r s E x p G r o 1 f i t o f a d v i s o r s (d) NP of advisors number of advisees N C a d v i s e e s G a u s s f i t o f a d v i s e e s (e) NC of advisees number of advisors N C a d v i s o r s E x p G r o 3 f i t o f a d v i s o r s (f) NC of advisors - 2 0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0 2 2 2 4 2 601 0 0 0 0 02 0 0 0 0 03 0 0 0 0 04 0 0 0 0 05 0 0 0 0 06 0 0 0 0 0 number of advisees N H a d v i s e e s G a u s s f i t o f a d v i s e e s (g) NH of advisees number of advisors N H a d v i s o r s E x p G r o 1 f i t o f a d v i s o r s (h) NH of advisors Fig. 3: Fitted curves plot for AA , N P , N C , and
N H . The horizontalaxis represents (a) advisees’ AA , (b) advisors’ AA , (c) advisees’ N P ,(d) advisors’
N P , (e) advisees’
N C , (f) advisors’
N C , (g) advisees’
N H , and (h) advisors’
N H , respectively. The vertical axis representsthe number of scholars. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 11
Table 2: Exponential fitting of AA , N P , N C , and
N H distributionfor advisors and advisees
Attributes Fitting FunctionAdvisees’
AA y = y + Ae R x Advisors’
AA y = Ae − AX Advisees’
NP y = ae bx Advisors’
NP y = y + A e xt Advisees’
NC y = y + Aw √ π e − ( x − xc )2 w Advisors’
NC y = y + A e xt + A e xt + A e xt Advisees’
NH y = y + Aw √ π e − ( x − xc )2 w Advisors’
NH y = y + A e xt AA andaverage N P , N C , and
N H of their advisees 10 years after first collaboratingwith their advisors. It can be seen that the indicators have the same trendwith the increasing of AA because these indicators have a certain dependence.All of them experience an initial growth, then remain stationary, and finallyreach a decline phase. When advisors’ academic ages are in the range 1-9,advisees’ academic performance shows a linear growth trend. When advisors’academic age reaches 28, advisees’ performance shows a downward trend. Inthe period when advisors’ academic age is between 10 and 27, advisees’ averageachievements reach the highest level and basically remain unchanged.It is interesting to consider the possible factors underlying this phenomenon.Prior studies have noted the importance of advisors’ career coaching and so-cial support in advisees’ development. The current study (Wang et al., 2017)finds that the productivity of scholars is dynamic over the course of their ownscientific career. For scholars whose AA <
12, the annual productivity rises upslowly. If their
AA >
24, their productivity will decline slightly with the AA .In the range of 12 < AA <
24, their annual productivity are nearly fixed. Ad-visees’ average productivity/impact shows a linear growth trend because theiradvisors’ productivity in the range 1 ≤ AA ≤ AA advisors and the future career performance of advisees thus need deeperexploration. Here are some possible reasons for our findings: (a) Academic research isa long-term accumulation process. Junior advisors need time to accumulateexperience in conducting research and mentoring advisees. Compared withsenior ones, their capacity and resources are limited. Senior advisors have cu-mulative advantages, both in academic performance and resources. (b) Whenthe academic age of advisors is at the intermediate level (in the range of 10-27years), their academic capability may reach a certain level as well. With suffi-cient resources, they can provide their advisees with the most helpful guidance.In addition to that, they are also willing to strive to instruct their advisees,because their own academic performance can be improved in this way. (c)Few senior advisors have a relatively long academic career (as shown in Fig.2(a)), that’s why senior advisors of academic age above 35 have poor perform-ing advisees. The random fluctuation may cause this phenomenon, and thusfurther study is required to untangle the underlying reasons. In general, thecorrelation between advisees’ high performance and advisors’ academic agesis certainly reasonable, but it is not the causation of scholars’ success. Hardwork and passion are still essential to scientists’ success in the scientific career. a d v i sees ’ N P o r NC advisors’ AA 01234 publications citations h-index a d v i sees ’ NH (a) Academic ages p u b l i c a t i o n s c i t a t i o n s h - i n d e x advisees' NC or NP advisees' NH a d v i s o r s ' N H (b) H-indices Fig. 4: Relationship between the advisors’ academic characteristicsand advisees’ academic performance for 10 years since collaborat-ing with their advisors. The horizontal axis represents (a) advisors’academic age, (b) h-indices. In the subgraphs, the left vertical axisrepresents advisees’ average number of publications and citations.The right vertical axis represents their average h-index.Similarly, in Fig. 4(b), the abscissa represents the advisors’
N H . In thissubgraph, the left vertical axis represents advisees’ average
N P , N C and rightaxis represents their average
N H . Comparing with advisees’
N C and
N H , ad-visees’
N P has the least obvious trend with the increase of advisors’
N H . Thephenomenon elicits the truth that quantity is not equal to quality especiallyin academia. The maximum value of advisors’ h-index we have considered is25 because there are few advisors with high
N H values (see Fig. 2). The value nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 13 of indicators for advisees’ fluctuates widely along with the growth of abscissa.So the data is binned in order to make the trend more clear.3.3 An accomplished advisor brings up a skilled adviseeAccording to the influence of advisors’ academic ages on advisees, Fig. 5 in-dicates advisees’ h-indices whose advisors’ academic ages are 5, 10, 15, 20,25, and the ranking of their h-indices begins at the top 5%, 5%-10%, 10%-15%, 15%-20%, 20%-25%, respectively. The polylines represent the trends ofall advisees’ average h-index over time. It can be seen from the data in Fig. 5,advisors with different rankings in h-index can bring their advisees differenth-indices. It is commonly perceived that high-achieving advisors literally causetheir advisees higher academic productivity and impact. Selective bias is onepossible reason as top advisors tend to select quality students. It is interestingto observe that as the advisors’ academic age rises, the gap among adviseesnarrows.The advisees supervised by advisors whose h-index rank the top 10% al-ways have the highest h-index. But it is not clear how significant the differencesamong these advisees are. On the basis of the definition of scholars’ impact,here we regard the top ten percent of advisors in h-index as
T op advisors. Res advisors represent the rest of advisors. We try to find out the differencesbetween advisees mentored by
T op advisors or Res advisors. We convert theproblem from a qualitative analysis to the quantitative analysis. Fig. 6 reflectsadvisees’ average
N P , N C , and
N H over time since advisees first collabo-rated with their advisors. In Fig. 6, advisors are differentiated with academicages and h-indices. In the figures, T n corresponds to T op advisors whoseacademic ages are n while R n corresponds to Res advisors whose academicages are n . From the figures, we can see that advisors’ N P results in the leastobvious discrepancy between advisees supervised by
T op advisors and Res advisors. The most striking observation to emerge from the data comparison isthe growth rate of each indicator. Advisees mentored by
T op advisors havehigher N P , N C , and
N H than the others. Moreover, their growth rate ishigher. This result may be explained by the Matthew Effect (Langfeldt et al.,2015). In academia, while excellent scholars’ accomplishments and reputationtend to snowball, those with modest accomplishments have greater difficultyto improve their impact.In Fig. 7, similarly, which is consistent with assessment standards of ad-visors, the advisees whose h-index rank the top 10% of each indicator areconsidered as the
T op advisees. The differences between T op advisees’ N P , N C , and
N H , who are mentored by
T op advisors or Res advisors arehighlighted in Fig. 7(d), Fig. 7(e), and Fig. 7(f). To the same extent as theprevious results, comparing with
N C and
N H , the discrepancy between ad-visees’ average
N P mentored by
T op advisors or Res advisors is not obvious.However, considering academic ages of the advisors,
T op advisees mentoredby T op advisors whose academic age is 30 have the lowest average N H . For h-index t ( y e a r s ) (a) 5 h-index t ( y e a r s ) (b) 10 h-index t ( y e a r s ) (c) 15 h-index t ( y e a r s ) (d) 20 h-index t ( y e a r s ) (e) 25
Fig. 5: Advisees’ h-indices whose advisors’ academic ages are (a) 5,(b) 10, (c) 15, (d) 20, (e) 25, as well as their h-indices are ranking atthe top 5%, 5%-10%, 10%-15%, 15%-20%, 20%-25%. The polylinesrepresent the trends of all advisees’ average h-index over time. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 15 advisees' NP T T T T T T R R R R R R t ( y e a r s ) (a) Number of publications A A = 3 0A A = 2 5A A = 2 0A A = 1 5A A = 1 0 advisees' NP
A A
A A = 5 (b) Publications’ difference T T T T T T R R R R R R advisees' NC t ( y e a r s ) (c) Citations A A = 3 0A A = 2 5A A = 2 0A A = 1 5A A = 1 0A A = 5 advisees' NC
A A (d) Citations’ difference T T T T T T R R R R R R advisees' NH t ( y e a r s ) (e) H-indices A A = 3 0A A = 2 5A A = 2 0A A = 1 5A A = 1 0A A = 5 advisees' NH
A A (f) H-indices’ difference
Fig. 6: Average academic performance of all advisees supervised by
T op advisors and Res advisors. T n corresponds to T op advisorswhose academic ages are n . R n corresponds to the rest whose aca-demic ages are n . (a), (c), and (e) show the advisees’ number ofpublications, citations and h-indices coached by different advisors,respectively. Details of their statistical mean and deviation are shownin (b), (d), and (f), respectively. Res advisors, the same scenario occurs to advisors whose academic ages are5. It is different from advisors’
N C and
N H . This phenomenon gives us inspi-ration that only considering the number of publications to evaluate scholarsis insufficient because quantity is not always equal to quality. As a result, anumber of comprehensive indices have appeared to evaluate scholars. In Fig.6 and Fig. 7, it can be seen from the subgraphs that comparing with all of theadvisees,
T op advisees mentored by T op advisors or Res advisors makemore distinct values in the average
N P , N C , and
N H .The results of the correlational analysis among advisor-advisee’s collabora-tion duration, advisors’ academic ages, proportion of advisees’ h-index rankingthe top 10% are shown in Fig. 8. The data in first three years of collabora-tion with advisors whose academic age is 5 are missing because of the slowgrowth rate in this period. It is difficult to calculate the top 10% advisees’ h-indices because the values of their h-indices are relatively low and similar. For
Res advisors, the proportion of advisees ranking top 10% is shown in the leftsubgraph in Fig. 8(a), and the result of
T op advisors is summarized in theright subgraph. Fig. 8(b) shows the proportion of differences between adviseesmentored by T op advisors and Res advisors. The values of advisors whoseacademic ages are 30 in Fig. 8(b) are always below 0.1. For others, their valuesare all above 0.1, even the maximum value can reach 0.28. It indicates thatmost
T op advisees are coached by T op advisors.Selective bias is a possible explanation for these phenomena. The advisor-advisee relationship is a two-way choice relationship, quality advisees tend toseek top advisors and vice versa. Moreover, we have also found that the gapbetween these advisees is closely related to the advisors’ academic age and thetime duration of their collaboration. In the first three years of collaboration,the gap between advisees is relatively small. Since the fourth year, the gapbegins to grow wider.3.4 Publication rates analysisAs mentioned before, we have separated advisors according to their AA in theexperiment. However, considering the publication rates increased over time, itmay have a biased effect on the scholars’ number of publications, citations, andh-indices. So we carry out the experiment to verify whether the publicationsrates impact our findings.Fig. 9(a) displays the number of new entries in the database per year. Fromthe data in Fig. 9(a), it is apparent that the growth rate of the publication isdifferent at different stages. The number of new records increases sharply in2003 and 2011. To be more specific, there are less than 25,000 new publicationsper year in 1999-2002 but over 50,000 in 2003-2007, almost two times morethan the first period. In order to ensure all of the advisees have plenty oftime to accumulate enough publications, we choose advisors whose academicage is 20 in 1996-2002 and 2003-2007, respectively. We compare the averageacademic performance (h-index) of their advisees. In Fig. 9(b) and Fig. 9(c), nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 17 T T T T T T R R R R R R advisees' NP t ( y e a r s ) (a) Number of publications A A = 3 0A A = 2 5A A = 2 0A A = 1 5A A = 1 0 advisees' NP
A A
A A = 5 (b) Publications’ difference T T T T T T R R R R R R advisees' NC t ( y e a r s ) (c) Citations advisees' NC A A
A A = 5 A A = 1 0 A A = 1 5 A A = 2 0 A A = 2 5 A A = 3 0 (d) Citations’ difference advisees' NH t ( y e a r s ) T T T T T T R R R R R R (e) H-indices A A advisees' NH
A A = 5 A A = 1 0 A A = 1 5 A A = 2 0 A A = 2 5 A A = 3 0 (f) H-indices’ difference
Fig. 7:
T op advisees’ academic performance supervised by T op advisors and Res advisors. (a), (c), and (e) show the
T op advisees’number of publications, citations and h-indices coached by differentadvisors. Differences between T op advisees’ N P , N C , and
N H whoare mentored by
T op advisors or Res advisors are highlighted in(b), (d), and (f). t (years)
5 10 15 20 25 30 d i ff ere n ce (b) Proportion differences Fig. 8: Correlations among advisor-advisee’s collaboration duration,advisors’ academic age, the proportion of advisees’ h-index rankingthe top 10%. (a) Advisees’ probability of h-index ranking the top10% mentored by
Res and
T op advisors. (b) The proportion ofdifferences between advisees mentored by T op advisors and Res advisors.the polylines show the advisees’ average h-index mentored by different h-indexranking advisors. Data in these figures can be compared with the data in Fig. 5,which is the basis of our findings. A positive correlation can be found betweenadvisors’ and advisees’ academic performance. So the publication rates havelimited influence on the phenomena we have found in this work.
Moreover, we have processed our dataset and do experiments on the part ofthe real dataset to explore the correlation between the advisor and their aca-demic grandchildren. We extract 4,256 advisor-advisee pairs (645 advisors) nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 19 nu m b e r o f r ec o r d s year publications (a) New records per year h-index t ( y e a r s ) (b) 1996-2002 h-index t ( y e a r s ) (c) 2003-2007 Fig. 9: Advisees’ h-indices whose advisors’ AA is 20 in different timeperiods, as well as their h-indices are ranking at the top 5%, 5%-10%, 10%-15%, 15%-20%, 20%-25%. (b) Advisees’ average h-indexmentored by advisors whose AA is 20 in 1996-2002. (c) Advisees’average h-index mentored by advisors whose AA is 20 in 2003-2007.and compile the statistics on the data. For advisors, we still divide them intotwo groups according to their h-indices: T op advisors and Res advisors. Andthen we analyze their grandchildren’s impact. We discover that the average h-index of
T op advisors’ grand-advisees is 40% higher than the grand-adviseesmentored by Res advisors. It illustrates that there may be also a correlationexisting between them and it needs in-depth research. Usually, PhD studentsare guided directly by their advisors, thus we only focus on the direct advisor-advisee relationships. Through this analysis, a comparatively deep analysisshould also be made on the relationship between the advisor and their aca-demic grandchild. We will carry out an in-depth study of this issue in futurework.There are still a few limitations in this work. Firstly, the conclusions areefficacious only for Computer Science, which is a rapidly developing discipline and has the characteristic of spreading its knowledge in conferences. It wouldbe interesting to explore the anatomy of the relationship in other traditionalareas of knowledge if we can get the large-scale advisor-advisee relationshipsdataset for the discipline. Secondly, we only show the correlation betweenadvisors’ academic characteristics and advisees’ academic performance. Otherfactors may also produce these phenomena, such as selective bias and the rankof institutions. It is necessary to further explore the cause of these phenomena.Thirdly, it is unfortunate that the study only illustrates the advisors’ benefitsderived from advisees on the formal relations. In terms of relevant experience,the informal advisor-advisee relationship is also an important part of mentor-ship. In future work, through the continuous improvement of the dataset, wewill conduct the in-depth study on the informal advisor-advisee relationships. This study presents advisees’ potential academic performance of choosing dif-ferent advisors with different academic ages or academic levels in the field ofComputer Science. It has identified the relationship between advisors’ aca-demic age and advisees’ performance, which experiences an initial growth,follows a sustaining stage, and finally ends up with a declining trend. The sec-ond major finding is that accomplished advisors can bring up skilled advisees,which is evidenced by advisees’ academic performance advised by differentadvisors. Taken together, these findings suggest a significant role for advisorsin promoting their advisees. This research extends our knowledge of scholars’career success and will serve as a base for future studies.In summary, through the analysis of the relationship between advisors andadvisees, the findings have significant implications to understand the relation-ship between advisors’ academic characteristics and advisees’ performance.Moreover, it can shed light on the advisor recommendation. However, it shouldbe noted that the results may differ in different research fields, it would beinteresting to explore the anatomy of the relationships in other areas of knowl-edge.
Acknowledgment
The authors extend their appreciation to the Deanship of Scientific Researchat King Saud University for funding this work through research group NO(RGP-1438-27).
References
Amjad, T., Ding, Y., Xu, J., Zhang, C., Daud, A., Tang, J. and Song,M. (2017), ‘Standing on the shoulders of giants’,
Journal of Informetrics (1), 307–323. nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 21 Anderson, C. (2006),
The long tail: Why the future of business is selling lessof more , Hachette Books.Azoulay and Pierre (2012), ‘Research efficiency: Turn the scientific method onourselves.’,
Nature (7392), 31–2.Borders, L. D., Wester, K. L., Granello, D. H., Chang, C. Y., Hays, D. G.,Pepperell, J. and Spurgeon, S. L. (2012), ‘Association for counselor educa-tion and supervision guidelines for research mentorship: Development andimplementation’,
Counselor Education and Supervision (3), 162–175.Bozionelos, N., Bozionelos, G., Polychroniou, P. and Kostopoulos, K. (2014),‘Mentoring receipt and personality: Evidence for non-linear relationships’, Journal of Business Research (2), 171–181.Chao, G. T. (1997), ‘Mentoring phases and outcomes’, Journal of VocationalBehavior (1), 15–28.Chao, G. T., Walz, P. and Gardner, P. D. (1992), ‘Formal and informal mentor-ships: A comparison on mentoring functions and contrast with nonmentoredcounterparts’, Personnel Psychology (3), 619–636.Dobson, L. (2013), ‘Effective practices of formal mentoring programs’, MathAlliance Research Study , 1–3.Fagenson, E. A. (1989), ‘The mentor advantage: Perceived career/job experi-ences of prot´eg´es versus non-prot´eg´es’, Journal of Organizational Behavior (4), 309–320.Florea, L., Cheung, Y. H. and Herndon, N. C. (2013), ‘For all good reasons:Role of values in organizational sustainability’, Journal of Business Ethics (3), 393–408.Ghosh, R. and Reio, T. G. (2013), ‘Career benefits associated with mentoringfor mentors: A meta-analysis’,
Journal of Vocational Behavior (1), 106–116.Gl¨anzel, W., Heeffer, S. and Thijs, B. (2017), ‘Lexical analysis of scientificpublications for nano-level scientometrics’, Scientometrics pp. 1–10.Gollapalli, S. D., Mitra, P. and Giles, C. L. (2011), Ranking authors in digitallibraries, in ‘Proceedings of the 11th annual international ACM/IEEE jointconference on Digital libraries’, ACM, pp. 251–254.Hu, C., Wang, S., Yang, C.-C. and Wu, T.-y. (2014), ‘When mentors feelsupported: Relationships with mentoring functions and prot´eg´es’ perceivedorganizational support’, Journal of Organizational Behavior (1), 22–37.Johnson, W. B. and Ridley, C. R. (2015), The elements of mentoring , Macmil-lan.Kram, K. E. (1983), ‘Phases of the mentor relationship’,
Academy of Manage-ment Journal (4), 608–625.Langfeldt, L., Benner, M., Sivertsen, G., Kristiansen, E. H., Aksnes, D. W.,Borlaug, S. B., Hansen, H. F., Kallerud, E. and Pelkonen, A. (2015), ‘Ex-cellence and growth dynamics: A comparative study of the matthew effect’, Science and Public Policy p. scu083.Lehmann, S., Jackson, A. D. and Lautrup, B. E. (2006), ‘Measures for mea-sures’,
Nature (7122), 1003–1004.
Letchford, A., Moat, H. S. and Preis, T. (2015), ‘The advantage of short papertitles’,
Royal Society open science (8), 150266.Ley, M. (2009), ‘DBLP: Some lessons learned’, Proceedings of the VLDB En-dowment (2), 1493–1500.Malmgren, R. D., Ottino, J. M. and Amaral, L. A. N. (2010), ‘The role ofmentorship in prot´eg´e performance’, Nature (7298), 622–626.Moreira, C., Calado, P. and Martins, B. (2011), Learning to rank for expertsearch in digital libraries of academic publications, in ‘Portuguese Confer-ence on Artificial Intelligence’, Springer, pp. 431–445.Murphy, T. H. (2015), ‘On great teachers’, Journal of Legal Studies Education (1), 223–227.Newman, M. E. (2001), ‘The structure of scientific collaboration networks’, Proceedings of the National Academy of Sciences (2), 404–409.Owens, B. (2013), ‘Research assessments: Judgement day.’, Nature (7471), 288–90.Scandura, T. A. and Ragins, B. R. (1993), ‘The effects of sex and genderrole orientation on mentorship in male-dominated occupations’,
Journal ofVocational Behavior (3), 251–265.Sinatra, R., Wang, D., Deville, P., Song, C. and Barab´asi, A.-L. (2016),‘Quantifying the evolution of individual scientific impact’, Science (6312), aaf5239.Singh, R., Ragins, B. R. and Tharenou, P. (2009), ‘Who gets a mentor? Alongitudinal assessment of the rising star hypothesis’,
Journal of VocationalBehavior (1), 11–17.Tang, J., Jin, R. and Zhang, J. (2008), A topic modeling approach and itsintegration into the random walk framework for academic search, in ‘DataMining, 2008. ICDM’08. Eighth IEEE International Conference on’, IEEE,pp. 1055–1060.Tang, J., Zhang, J., Yao, L., Li, J., Zhang, L. and Su, Z. (2008), Arnetminer:Extraction and mining of academic social networks, in ‘Proceedings of the14th ACM SIGKDD International Conference on Knowledge Discovery andData Mining’, ACM, pp. 990–998.Tuesta, E. F., Delgado, K. V., Mugnaini, R., Digiampietri, L. A., Mena-Chalco,J. P. and P´erez-Alc´azar, J. J. (2015), ‘Analysis of an advisor-advisee rela-tionship: An exploratory study of the area of exact and earth sciences inbrazil’, PloS one (5), e0129065.Wanberg, C. R., Kammeyer-Mueller, J. and Marchese, M. (2006), ‘Mentorand prot´eg´e predictors and outcomes of mentoring in a formal mentoringprogram’, Journal of Vocational Behavior (3), 410–423.Wang, W., Liu, J., Xia, F., King, I. and Tong, H. (2017), Shifu: Deep learningbased advisor-advisee relationship mining in scholarly big data, in ‘Proceed-ings of the 26th International Conference on World Wide Web Companion’,International World Wide Web Conferences Steering Committee, pp. 303–310.Wang, W., Liu, J., Yu, S., Zhang, C., Xu, Z. and Xia, F. (2016), Miningadvisor-advisee relationships in scholarly big data: A deep learning ap- nderstanding the Advisor-advisee Relationship via Scholarly Data Analysis 23 proach, in ‘Proceedings of the 16th ACM/IEEE-CS on Joint Conferenceon Digital Libraries’, ACM, pp. 209–210.Wang, W., Yu, S., Bekele, T. M., Kong, X. and Xia, F. (2017), ‘Scientificcollaboration patterns vary with scholars academic ages’, Scientometrics (1), 329–343.Xia, F., Wang, W., Bekele, T. M. and Liu, H. (2017), ‘Big scholarly data: Asurvey’,
IEEE Transactions on Big Data (1), 18–35.Young, A. M. and Perrewe, P. L. (2000), ‘What did you expect? an examinationof career-related support and social support among mentors and prot´eg´es’, Journal of Management26