ProbKS: Keyword Search on Probabilistic Spatial Data

Abstract

Many applications, like Twitter, Yelp, or Facebook, produce documents that are tagged with geolocations. For example, when a user tweets using Twitter, the tweets are tagged with the user’s location (inferred using the user’s IP address, or mobile GPS). These locations, however, are computed with inherent uncertainty. In such scenarios, it is desired to support search queries that take into account both text relevancy and location proximity. In this paper, we study the problem of text retrieval queries on probabilistic spatial data. We consider top-($c$, $k$) queries to capture semantics of both textual relevance and probabilistic location proximity. A top-($c$, $k$) query returns $k$ tuples which have the highest probability of being in the top-$c$ query results under the possible world semantics. We propose a framework to answer such queries. Our framework integrates two components: scoring textual similarity based on the query text; and the document text and calculating top-$c$ confidence based on the probability of the document falling within the query region. We develop an IRTree-based Incremental Scoring Approach (ISA) that returns an iterator over tuples in decreasing order of text similarity. Our parameterized probabilistic ranking algorithm $PRank^c$, consumes the output of ISA interactively and calculates top-$c$ confidence of these tuples in linear time. We also provide a heuristic optimization to terminate the $PRank^c$ algorithm earlier without compromising on result quality. We conduct experiments on real data to show the efficiency of this framework.