What makes a great PhD supervisor (for recommender-systems and machine-learning research)?

Published by Joeran Beel on

Successful PhD Candidates at Trinity College Dublin (Machine Learning and Recommender Systems)

Successful PhD Candidates at Trinity College Dublin. Source: https://www.tcd.ie/medicine/physiology/assets/img/IMG_0858.jpg

From time to time, I get asked by students “how can I find a good PhD supervisor?” And sometimes I sit on PhD evaluation panels and I am surprised how little thought the PhD students put into choosing their supervisor. Therefore, I decided to write this blog post for potential PhD candidates about choosing a good PhD supervisor. I hope it provides guidance to those who plan to do a PhD. In my description, I focus on potential PhD students in information retrieval, machine learning and recommender systems because these are the fields I am working in. However, most of the information probably applies to PhD studies in other fields.

Before I explain the important characteristics of a good PhD supervisor, one note: If you have a potential PhD supervisor in mind, ensure that this person will be really your day-to-day supervisor. Only because you are talking to a professor about a PhD supervision doesn’t mean that this professor will be your day-to-day supervisor. If the professor is leading a large research group, chances are that one of the postdoctoral researchers will be your day-to-day PhD supervisor. This is not necessarily bad, but beware that all the following points should then primarily apply to the postdoctoral researcher who is supervising you.

So, let´s get started. The perfect PhD supervisor has/provides:

    1. A good track record of successful PhD students
    2. Expert Knowledge / Authority in the Field / Large Network
    3. Sufficient Time for your PhD supervision
    4. Opportunities for collaborations
    5. A permanent and stable position
    6. Sufficient, Secure, and Flexible Funding
    7. Good office space
    8. A position at a reputable university
    9. Additional Mentorship Skills

1. A good track record of successful PhD students

The single-best indicator of a good PhD supervisor is the success of the supervisor´s previous and current PhD students. Therefore, do some research on the previous and current PhD students. Probably the students are listed on the supervisor´s homepage. If not, ask your potential supervisor.

To evaluate the success of PhD students, you should look at publication output and post-PhD positions.

PhD-Students´ Publication Output

On average, a PhD student in machine learning and recommender systems, who was well supervised, and worked full-time, should have roughly the following publications after four or five years: 1 very good journal or conference article; 3 full-papers at good conferences; 6 other publications at average conferences or workshops (mix of full papers, short papers, demos/posters). On some of these publications, the PhD student should be the first author. Ideally, there are also one or two publications without the PhD supervisor as co-author. The number of publications may vary strongly depending on the type of paper and the venue. For instance, one full-paper at a conference has probably the same value as two or three short papers. And one short paper has the about same value as two or three posters/demos. One full-paper in a (very) good journal or conference may easily equal two or even three average conference publications. Consequently, a PhD student who has “only” three publications, but all at top-conferences and journals, may also have had a very good supervisor. However, if many PhD students of your potential supervisor have, for instance, their first publication only in the third year of their PhD, and even after five years the best publication is only at a mediocre conference, you should be sceptical.

Post-PhD Positions

A PhD student who has a good supervisor will develop a passion for academia. Consequently, a large proportion of the previous PhD students should have academic positions (ideally professorships) or work in research departments of major IT companies after completing their PhD. If most previous PhD students of your potential supervisor work as software engineers, project managers or consultants, then the supervisor probably did not do a good job.

2. Expert Knowledge / Authority in the Field / Large Network

Ideally, your PhD supervisor is an authority in the field of your PhD topic, i.e. the supervisor has lots of expertise, is well known in the community and has a large network. Some key indicators for the knowledge and network of a supervisor are the publications, publication venues, citations, co-authors, invited talks, and committee memberships as explained in the following.

Publications & Venues

Visit the supervisor’s homepage or Google Scholar profile and check if some of the recent publications relate to your PhD topic. Just read the titles and maybe abstracts to get a rough idea. Also look at the publication venues (journals and conferences) where the potential supervisor has published. Use http://www.core.edu.au/conference-portal (conferences), http://portal.core.edu.au/jnl-ranks/ (journals) and http://www.scimagojr.com (journals) to find out, how good the venues are. The best is A*, then A, then B, then C. Be careful though about the ratings. For instance, Core lists the ACM Conference on Recommender Systems (RecSys) as a B conference. However, in the field of recommender systems, RecSys is undoubtedly the premier conference.

Citation Counts

A good first indication of a researcher´s reputation and impact is the researcher´s citation counts. Because citation counts vary among disciplines, you should compare your potential supervisor with other researchers in the same field. A good starting point is Google Scholar to learn about the most-cited researchers in your field of interest.  On Google Scholar you find the most-cited researchers for any research field, using this URL https://scholar.google.de/citations?view_op=search_authors&hl=en&mauthors=label:your_research_field. For instance, the most cited researcher for machine-learning has 280,891 citations (see screenshot).

Most-Cited Machine-Learning Researchers on Google Scholar

Most-Cited Machine-Learning Researchers on Google Scholar

In contrast, the most-cited researcher for recommender-systems has 59,476 citations.

Most-Cited Recommender-Systems Researcher (Google Scholar)

Most-Cited Recommender-Systems Researcher (Google Scholar)

When looking at these numbers, keep in mind that these are absolute citation counts.  Comparing citation-counts of a 60-year old professor with a 30-year old postdoctoral researcher doesn´t make sense. Rather, you may want to look at the researcher’s profile and look at the citations per year (and the trend).

George Karypis´ Google Scholar Profile (Most-Cited Recommender-Systems Researcher)

George Karypis´ Google Scholar Profile (Most-Cited Recommender-Systems Researcher)

Co-Authors

Look at the researchers your potential supervisor cooperates with, i.e. co-authored publications. You can do this easily via Google Scholar. A large number of different co-authors indicates a large network. And, ideally, the co-authors are from different countries and are well-known researchers as well.

More…

Ideally, your potential PhD supervisor is also organizing (high-impact) conferences and workshops in the field of your research and is on editorial boards of journals. Giving keynotes on conferences, and invited talks is also a good indication that your potential supervisor is an authority in the community.

3. Sufficient Time for your PhD supervision

A fundamentally important factor for the success of your PhD is the time that your supervisor has for you. Having the authority in your field as PhD supervisor is useless if your supervisor has no time to supervise you. Unfortunately, the time of a PhD supervisor is typically the inverse of his or her knowledge and reputation. The reason is simple. An authority usually needs to give keynotes, serve on editorial boards, advice IT companies, maybe lead an entire research centre and supervise other staff including postdocs. That leaves little to no time for supervising PhD students (I am sorry to tell you but while PhD students are important, they are not the most important thing to a professor). So, you probably should try to find a supervisor with the right balance between time and expertise/reputation. A few other factors that may impact a supervisor’s time are listed below.

Supervisor´s Obligations Besides PhD supervision

Beware of what your potential supervisor does besides supervising PhD students and doing research. Is the supervisor having a high teaching load? Or leading a research centre? Then there is probably little time left for supervising PhD students.

Personal Interest in Your PhD Topic

Professors typically work on many projects, but that does not mean that they like all projects equally, and invest the same amount of time for each project. For instance, a professor may get some “easy” money from an industry partner or funding organization. This means a company may have approached a professor and offered generous funding for a joint project. Hardly any professor will say ‘no’ to this as long as the project somehow relates to the professor’s interests. However, that does not mean that the project is the professor’s favourite research topic that he or she is really passionate about. Consequently, the professor would probably spend less time on that project than on other projects. You should try to find out how much personal interest your potential PhD supervisor has on your PhD topic.

Personal Interest in Succeeding PhD students

Supervisors have different personal interests in successful PhD students. Young assistant professors typically will have an inherent interest that you succeed. Assistant professors typically want a promotion, and one metric for promotions are successfully completed PhD students. Similarly, assistant professors need publications for promotions, and the main source for publications often are PhD students. So, they will tend to push you toward publishing, which may be stressful but eventually positive. Although, in the worst case, they may exploit you and e.g. always add their names first publications. In contrast, more senior professors do not need you that much. This may be good or bad. For instance, they may have altruistic motives to support you as good as possible (while more junior researchers are probably less altruistic). So, in summary, having a more senior or junior supervisor is neither generally good or bad. But you should try to find out how much personal interest your potential supervisor has in having you succeed.

Also, beware that in the worst case, supervisors have an interest in you not completing your PhD, or at least delaying the completion. Such supervisors exploit their PhD students as cheap labour for teaching and administration and want to keep them as long as possible. Fortunately, only very few supervisors are like that. A good indication to find out is to ask how long other PhD students in the group needed to complete their PhDs.

4. Opportunities for collaborations

During your PhD, you should ideally supervise a few Master students and collaborate with other PhD students, more senior researchers and maybe some industry partners. Some research visits at other universities or industry placements may also be beneficial. The advantages of such collaborations are manyfold: You will gain valuable knowledge, broaden your own network, increase your publication output, and polish your CV.

Collaborations are typically within the supervisor´s lab, within the department, or within the network of your supervisor (colleagues, industry partners, …). Hence, you should look for a supervisor with a sufficiently large lab and a large network. Good indicators for a large network are the number of co-authors and, less importantly, the number of contacts on Linkedin. Supervisors who are active in the community, in terms of e.g. organizing conferences, are typically also well connected. Also, ask your supervisor how open he or she is of ‘letting you go’ for a few months to work abroad.

5. A permanent and stable position

You will spend the next 3 to 5 years doing your PhD. Hence, you should be sure that your potential supervisor will be available for the next 3 to 5 years. You may not be aware of this but the fluctuation in academia is quite high. So, you should look at a few indicators to be able to estimate the chance that your potential supervisor will still be at the university in a few years. Or, you should be willing to move where your supervisor may move. A few indicators are as follows

Contract Type / Duration

Your safest bet is with a professor who is tenured, i.e. a professor who has a permanent position. Your second-best option is a tenure-track professor, i.e. a professor who will be tenured in a few years if he or she performs well. Supervisors with a temporary contract (postdoctoral researchers and non-tenured professors) are riskier. There is a quite high chance that they will take on another position. However, keep in mind that it is not necessarily a disaster if your supervisor changes the university — at least not if you would be willing to follow your supervisor. Often, when professors take on a position at a different university, they are glad if some of their PhD students follow them. So, if you are willing to relocate, a supervisor with a temporary contract shouldn’t be a no-go.

Settled

Try to find out how deeply rooted your potential PhD supervisor is at the university and the city. If your potential supervisor has a partner in town, owns a house or apartment, and has children who go to school, the supervisor probably will not want to move. However, in contrast, if the potential supervisor is rather young, very ambitious, and has no family — or even worse, a family in a different town or even country — chances are quite high that this supervisor will apply for positions elsewhere.

Sabbatical

In many countries, professors have the right to take sabbaticals. Sometimes, these sabbaticals last a year, sometimes even longer. While this is quite awesome for the professor, this may not be so awesome for the PhD students the professor supervises. So, it might be a good idea to ask your potential supervisor, if he or she has any plans for a long absence.

6. Sufficient, Secure, and Flexible Funding

The source for your PhD funding may have a significant impact on how your PhD project and supervision develops. If your funding was given to do a specific project, then you do need to do this specific project and you will have little freedom to move the focus. In contrast, if you receive a scholarship with no specific project goal, you will be much more flexible.

You should ask your potential supervisor

  1. Whether the funding is secure for the entire period of your PhD, or if it could happen that funding runs out after e.g. one or two years.
  2. Whether there is sufficient funding to send you to conferences at least once a year, better twice (costs for conference attendance is typically between 500 and 1700 Euros including hotel etc.).
  3. How flexible the funding is. If the funding is e.g. coming from an industry partner, this partner probably wants a very specific project to be done. However, be ensured that one thing is certain: During your PhD, you will want to adjust or even change your topic. Hence, ideally, the funding is completely flexible and allows you to do whatever is best for completing your PhD.

7. Good office space

Ask your potential PhD supervisor where you would be working. Would you have your own office (probably not)? Share an office with a few others? Sit in an open space? Also, have a look at the office, and if you see that the tables and chairs are very old, ask whether you could get new equipment (chances for success are higher before you sign a contract).

8. A position at a reputable university

The reputation of the university where you do a PhD is important. Whether you attend a conference, apply for a research grant, or just talk with colleagues: being at a reputable university will give you advance laurels. This may be unfair, but can be very helpful.

9. Additional Mentorship Skills

A PhD supervisor should be first and foremost a mentor. But don´t be mistaken and assume that only because your potential PhD supervisor is an excellent researcher, and maybe even has plenty of time for you, he or she is also an excellent mentor. A good mentor has excellent communication skills and gives you honest and critical feedback; motivates you and shares his or her knowledge, and a good mentor should have good organizational and project management skills. Actually, I believe that these skills are probably even more important than the general expertise of a supervisor because usually, after one or two years, you should know more than your supervisor anyway about your particular PhD topic. If you had to choose between a supervisor with excellent skills in your PhD topic but horrible organizational/management skills, or a supervisor with little expertise in your field but excellent organizational/management, the latter one would probably be the better choice. Of course, ideally, you find a supervisor who is excellent in all aspects.

 

 

 

 

 


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.