Information Retrieval Meets Game Theory
Abstract: In competitive search settings such as the Web, authors of documents may have an incentive to have their documents highly ranked for certain queries. This can drive corpus dynamics as documents may be manipulated in response to induced rankings (e.g., by applying search engine optimization). Such post-ranking corpus effects are not directly modeled in ad hoc retrieval models and, more generally, are not accounted for by the formal foundations of retrieval paradigms. In this talk I will discuss how (algorithmic) game theory can be used to analyze some aspects of the competitive search setting. I will first discuss the probability ranking principle (PRP) [3] which is the theoretical underpinning of most ad hoc retrieval methods. As it turns out, the PRP is sub-optimal in competitive settings [1]. In addition, I will discuss some initial theoretical and empirical results regarding the strategic behavior of document authors in competitive retrieval settings, specifically with respect to the foundations of classical ad hoc retrieval models [2]. I will then discuss future directions.
Loading