Friday, October 14, 2022 - 14:00 to 15:30
SCHOOL OF COMPUTER SCIENCE
The School of Computer Science is pleased to present…
MSc Thesis Proposal by: Seyed Sobhan (Arman) Vagheh Dashti
Date: Friday October 14th, 2022
Time: 2:00PM - 3:30PM
Location: Essex Hall, Room 122
Reminder: Two-part attendance required: Part I (Scan the QR code, fill in online form) and Part II (sign in sheet). *Make sure you are in the room at least 5-10 minutes BEFORE the presentation starts – once the presentation has begun, latecomers will not be admitted.
Predicting future successful teams of experts who can synergistically work with each other is challenging due to 1) the enormous pool of plausible expert candidates with diverse backgrounds and skills, and 2) the drift and variability of interest and the level of expertise of candidates in time. Indeed, prior works in team formation have neglected the fact that experts’ skill sets and interests change over time. Specifically, 1) operations research (OR)-based methods wherein multiple objective functions must be optimized with respect to constraints via integer programming overlooked the temporal nature of human collaborations; 2) graph-based search represented the expert network as a static graph and did not take into account the dynamics of the expert network in time and the constant emergence of new collaborations; 3) neural-based methods which map experts and skills in a latent space and bring vectors of experts and their skills close to each other fail to recognize the possible drift and variability of experts’ skills and interest in time and its impact on the prediction of future successful teams. Moreover, neural models are prone to overfitting when training data suffers from the long-tail phenomenon, i.e., few experts have a lot of successful collaborations and the majority have participated sparingly. To overcome the aforementioned shortcomings, we propose (i) a streaming scenario training strategy for neural models where we train the models in an orderly manner to grasp the changes in experts’ skills and interests within time, and (ii) an optimization objective that leverages both successful and virtually unsuccessful teams via various negative sampling heuristics to address the long-tail distribution of experts over teams. We empirically benchmark our proposed objective functions and training method against the state-of-the-art in terms of effectiveness and efficiency on four large-scale datasets with varying domains and distribution of skills and experts, namely, dblp (computer science researchers), imdb (movies cast and crews), uspt (US patents inventors), and github (open-source software developers).
Keywords: Temporal Neural Team Formation, Artificial Intelligence, Machine Learning, Neural Networks, Negative Sampling
MSc Thesis Committee:
Internal Reader: Dr. Robin Gras
External Reader: Dr. Mohammad Hassanzadeh
Advisor(s): Dr. Hossein Fani and Dr. Saeed Samet
MSc Thesis Proposal Announcement
5113 Lambton Tower 401 Sunset Ave. Windsor ON, N9B 3P4 (519) 253-3000 Ext. 3716 email@example.com