About Me
I am a PhD student in Computer Science at the University of Colorado Boulder, advised by
Prof. Robin Burke.
My research lies at the intersection of recommender systems and interpretable machine learning, with a focus on understanding, explaining, and controlling how recommendation models make decisions.
My current work makes trained recommendation models interpretable and controllable through post-hoc concept-based representations and steering mechanisms that enable meaningful interventions while preserving recommendation quality. I also work on provider-side transparency and on counterfactual and comparative explanations.
Projects
IN PROGRESS
Exploring Social Media Interfaces with Content Creators
Co-author
In progress, targeting ACM CSCW
An interview study with content creators on Instagram, TikTok and YouTube, including walkthroughs of their analytics dashboards, examining how platform analytics shape creators’ theories about recommendation algorithms and their content strategies.
UNDER REVIEW
Steerable and Interpretable Recommendation via Post-Hoc Concept Bottlenecks
Meysam Varasteh, Masoud Mansoury, James Caverlee, Robin Burke
Under review at The Web Conference (WWW) 2027
A model-agnostic, post-hoc framework that adds human-interpretable concept bottlenecks to pretrained recommenders, enabling concept-level steering that is more targeted than SAE-based approaches while preserving ranking quality.
Counterfactual Explanations for Recommendation: Are We Hitting the Target?
Meysam Varasteh, Masoud Mansoury, James Caverlee, Robin Burke
Under review at The Web Conference (WWW) 2027
An evaluation of state-of-the-art recommendation explainers showing that standard fidelity metrics reward explanations that reshuffle the whole recommendation list, and proposing list-aware metrics that credit an explanation only if it moves the target item while largely preserving the rest of the list.
2026
Why Didn't More People See It? Recommendation Transparency for Providers
Meysam Varasteh, Robin Burke
IntRS Workshop @ ACM RecSys 2026
A surrogate-model approach that explains, at the system level, why providers’ items do or do not get exposure in recommendation lists, revealing which factors drive exposure and how they differ across recommenders and domains.
Why This, Not That? Mining User Profiles for Pair-wise Counterfactuals
Meysam Varasteh, Veronika Bogina, Noam Koenigstein, Robin Burke
IntRS Workshop @ ACM RecSys 2026
Introduces pairwise explanation of rankings, answering “Why is item A ranked above item B?”, and a counterfactual approach that identifies the items in a user’s history responsible for the relative ranking of two items.
Counterfactual Profile Perturbations for Recommendation Explanations: A Reproducibility Study
Meysam Varasteh, Elizabeth McKinnie, Robin Burke
arXiv preprint
A reproducibility study of the LXR counterfactual explainer that extends it beyond the top-ranked item, shows that simple baselines such as popularity and latent-factor similarity can work surprisingly well, and estimates a lower bound on optimal explanation size.
2024
Comparative Explanations for Recommendation: Research Directions
Meysam Varasteh, Elizabeth McKinnie, Amanda Aird, Daniel Acuña, Robin Burke
IntRS Workshop @ ACM RecSys 2024
A position paper that proposes a taxonomy of comparative explanations for recommender systems, highlighting their value for providers, and outlines directions for future research.
2023
An Improved Hybrid Recommender System: Integrating Document Context-Based and Behavior-Based Methods
Meysam Varasteh, Mehdi Soleiman Nejad, Hadi Moradi, Mohammad Amin Sadeghi, Ahmad Kalhor
31st International Conference on Electrical Engineering (ICEE 2023)
A hybrid recommender that addresses data sparsity by combining user-based and item-based collaborative filtering with NLP-processed textual context such as reviews, outperforming baselines on three real-world datasets.
Fairness Auditing Through the Lens of Counterfactual Explanations
Yashar Deldjoo, Meysam Varasteh, Yara M. Bahram, Marco Antonio Insabato, Tommaso Di Noia
2022
Designing a Sequential Recommendation System for Heterogeneous Interactions Using Transformers
Mehdi Soleiman Nejad, Meysam Varasteh, Hadi Moradi, Mohammad Amin Sadeghi
arXiv preprint
A Transformer-based sequential recommender that models different types of user interactions as a sequence of events and builds the dependencies between event types into its architecture, with educational content recommendation as a motivating use case.
Honors & Awards
2026
Outstanding Teaching Assistant Award, CS Department, University of Colorado Boulder
2026
WSDM NSF Conference Travel Grant, National Science Foundation
2025
Departmental Summer Research Fellowship, CS Department, University of Colorado Boulder
2024
SIGIR Conference Student Travel Grant, SIGIR
2023
Dean’s and Departmental Excellence Fellowship, CS Department, University of Colorado Boulder
2023
Early Career Professional Development Fellowship, CS Department, University of Colorado Boulder