This semester (Fall 2020), we kicked off the inaugural SCS PhD Coffee Chats program! The goal of the program is to encourage spontaneous connection, friendship, and mentorship throughout SCS. In light of the pandemic, we felt it especially important to connect students of all years and departments in the School of Computer Science. Over the course of the semester we’ve received a lot of positive and constructive feedback, and heard from students who’ve wanted to implement smaller-scale version of coffee chats in their own departments. Thus, we figured we’d explain the details of the program in a reproducible way, including the matching algorithm, lessons learned, and some statistics about participation.
To be matched for the following week, students fill out a 2-minute form like this (this is a publicly accessible link instead of the CMU-accessible link; edit access to a blank copy given upon request). Students must be signed into their school email account in order to fill out the form.
The form asks for the student’s name, pronouns, department, year, availability, whether they prefer zoom or in-person interactions, and whether they want to be in a pair or placed in a group of 3 or 4 students. Then, they select which type of match they are looking for: friendship outside of work, research topic, and mentorship.
Depending on the type of match they select, students (optionally) enter different supplementary information which they are subsequently matched on:
Below is an example email that a participant would receive:
Hi Alice and Bob, Thank you for participating in the SCS Coffee Chats program! This week, we've matched you on the theme(s) of: Research topic. Alice (she/her/hers) is in year 1 of PhD in the Machine Learning Department, and Bob (he/him/his) is in year 3 of PhD in the Computational Biology Department. You are all available at: Friday Morning, Wednesday Evening, and can meet Over Zoom Note: Please be respectful of each other's time, and try to confirm with your partner(s) in a timely manner. If you don't receive a response and would like to be re-matched, let us know and we can see if we can rematch you. Alice's research topics/interests include: time series, causality, healthcare Bob's research topics/interests include: neuroscience, time series, computer vision To help kick off the conversation, here are some optional icebreaker questions/activities for when you meet: 1. What's the best piece of advice you've ever heard? 2. If you had 25 hours a day, how would you use your extra time? 3. Do a short workout together (e.g. 10 jumping jacks, 10 pushups, 10 sit ups). 4. Take a creative selfie/screenshot and send it to us! (feel free to use this email thread) Finally, once you've had your coffee chat we'd love to hear how it went & any feedback/suggestions you might have: [link] Happy coffee-chatting! Social Connectedness Working Group (part of the SCS PhD Advisory Committee)
Code for the full matching algorithm is available at: https://github.com/scs-phd-deans-committee/coffee-chats-public
Greedy matching algorithm: Within each match type (friendship, research, mentorship, or random) we use a greedy score-based algorithm where the scoring function depends on the match type. At a high level, given the scoring function, the matching algorithm works by computing a score for every possible grouping of 2, 3, or 4 students. Next, the algorithm acts greedily, selecting the highest-scored match first and removing its members from the remaining pool of potential groups. Then, the next highest-scored match is selected and removed from the pool. This process continues until everyone is matched. If there are any remaining unmatched people, then they are matched manually.
Scoring functions: For each match type, we define a scoring function that takes in a pair of individuals and outputs a score. To get scores for larger groups, we add the pairwise scores for all possible pairs in that group and rescale the sum.
Across all match types, a logistical compatibility score is computed based on overlapping availability, location (zoom vs. in-person), and group size (pair vs. 3-4 people). In addition to logistical compatibility, we include the following considerations into the scoring functions for each match type (see the code for complete details):
At the end when all participants have been matched, a manual spot check is done for the matchings, and we check the “anything else” freetext to see if there are any issues with the matchings. If so, manual adjustments are made.
Finally, emails are sent to each matched group using an automated script that populates the body of the email with basic information about each person, as well as recommended logistics based on their responses.
Over the course of the semester, we went through several iterations of the program as we received feedback from participants. Here’s a summary of the changes we made in response to feedback:
In addition to constructive feedback, we got a lot of positive feedback. Here are some excerpts:
Here are some numbers-driven takeaways from our program:
Thank you to all of you who participated in and/or provided feedback on the program! If you’re interested in getting involved or have any questions, comments, or concerns, please direct them to hlzhou[at]andrew.cmu.edu.
Written by Helen Zhou
The PhD Coffee Chats program was made possible by the collective hard work and insight of the Social Connectedness Working Group; thank you to Catherine King, Abhinav Adduri, Tobias Durschmid, Ritam Dutta, Roger Iyengar, Shilpa George, Leo Chen, and anyone else who helped at various stages of ideation, creation, and deployment!