"If community notes are the answer, what is the question?" & "Competitive search with multiple queries and corpus expansion"

Time & Location: Friday, March 14, 1:30pm - 2:30pm at SEC LL 2.221

Speaker: Andreas Haupt (Postdoctoral Fellow - Stanford)

Title: If community notes are the answer, what is the question?

Abstract: One of the primary challenges in crowdsourced fact-checking is that users must write notes clarifying the context of misleading social media posts, often missing the critical window for impact. A potential solution is to leverage large language models to generate these notes, and let users merely vote on their helpfulness. Such generation requires a deeper examination of community notes’ objective. I analyze the properties of X’s “bridging” community notes algorithm as a social choice mechanism—focusing on static incentive compatibility, shill-proofness, and its vulnerability to broader selection biases—and outline an approach for the automatic generation of community notes.

Meeting times: Andreas has graciously allocated some time to meet with anyone interested in chatting about his work or research. Please sign-up on the google sheets link here.

Speaker: Haya Nachimovsky (MSc Student - Technion)

Title: Competitive search with multiple queries and corpus expansion

Abstract: The Web is a canonical example of a competitive search setting that includes document authors with ranking incentives: their goal is to promote their documents in rankings induced for queries. The incentives affect some of the corpus dynamics as the authors respond to rankings by applying strategic document manipulations. This well-known reality has deep consequences that go well beyond the need to fight spam. We present two contributions to the study of competitive search. We first analyze a setting where authors opt to improve their documents' ranking for multiple queries in contrast to a single query, which was the case in past work. We use game theoretic analysis to prove that equilibrium does not necessarily exist, yielding potential instability in the search ecosystem. We then empirically show that it is more difficult for authors to improve their documents' rankings for multiple queries with a neural ranker than with a state-of-the-art feature-based ranker. Our second contribution is a novel paradigm for improving the stability of the competitive search setting. As lack of convergence of authors' content might lead to poor publisher welfare, and as a result to poor users' welfare, novel content ranking algorithms have been proposed in the past. We offer an alternative approach: corpus enrichment with a small set of fixed dummy documents. It turns out that, with the right design, such enrichment can lead to pure Nash equilibrium and even to the convergence of best-response dynamics to a high welfare result, where we still employ the classical/current content ranking approach. We show two such corpus enrichment techniques with tight bounds on the number of documents needed to obtain the desired results. Interestingly, our study is a novel extension of Borel's Colonel Blotto game.