WatchLens open-source platform links video rec policies to playback behavior

WatchLens, an open-source platform created by researchers Deogyong Kim and Dongha Lee, addresses a critical gap in video recommendation research: linking playback behavior to the recommendation conditions that produced it. Existing user-study tools often track either user behavior or recommendation policies, but rarely both in a single workflow. WatchLens solves this with a modular architecture that keeps user interfaces, content sources, and recommendation policies independently configurable. Researchers can assign different policies to the feed and watch page separately, and a standardized logging layer attaches the specific policy and ranking position to every recorded event. This design eliminates the need to reconstruct the connection between recommendations and outcomes after the fact, enabling precise analysis of how policies affect playback behavior, session continuation, and navigation patterns.

The platform's capabilities are demonstrated in a short-form video case study where the interface, feed policy, and content pool remain constant while only the watch-page policy varies. This setup allows researchers to perform session-level comparisons of recommendation effects on real viewing behavior. WatchLens is released as a publicly available, single-server deployable system, making it practical for reproducible online video recommendation research. The paper, accepted to the ACM Recommender Systems conference (RecSys 2026), is available on arXiv with DOI 10.48550/arXiv.2608.04807. By providing a flexible, well-instrumented experimentation environment, WatchLens gives the research community a standard tool to test how different recommendation policies influence user engagement and viewing habits, potentially accelerating progress in understanding and improving video recommender systems.