Web Stream Reasoning Using Probabilistic Answer Set Programming ⋆ Matthias Nickles 1,2 and Alessandra Mileo 1 1 INSIGHT Centre for Data Analytics National University of Ireland, Galway {matthias.nickles,alessandra.mileo}@deri.org 2 Department of Information Technology National University of Ireland, Galway Abstract. We propose a framework for reasoning about dynamic Web data, based on probabilistic Answer Set Programming (ASP). Our approach, which is proto- typically implemented, allows for the annotation of first-order formulas as well as ASP rules and facts with probabilities, and for learning of such weights from ex- amples (parameter estimation). Knowledge as well as examples can be provided incrementally in the form of RDF data streams. Optionally, stream data can be configured to decay over time. With its hybrid combination of various contem- porary AI techniques, our framework aims at prevalent challenges in relation to data streams and Linked Data, such as inconsistencies, noisy data, and probabilis- tic processing rules. Keywords: Web Reasoning, Uncertainty Stream Reasoning, Answer Set Program- ming, RDF, Probabilistic Inductive Logic Programming, Machine Learning 1 Introduction & Related Work Many real-world applications on the Web involve data streams (e.g., messaging events, web searches, or sensor data), but while stream processing and data stream mining are already established research areas, stream reasoning [28] is still a very young research field. Challenges in this regard are not only incremental reasoning in the presence of rapidly changing dynamic information, but also provisions for inconsistencies, inco- herence and noise in stream data, and stream reasoning using probabilistic background knowledge (e.g., probabilistic rules). Probabilistic logic programing, and the ability to learn facts and rules from possibly incomplete or noisy data, can provide an attractive approach to stream reasoning, since it combines the deduction capability and declar- ative nature of logic programming with probabilistic inference abilities traditionally modeled using Bayesian or Markov networks. In particular nonmonotonic probabilistic (inductive) logic programming [18, 1, 25, 4, 16] is promising in this regard, as it already provides for concepts useful for dealing with dynamic knowledge by means of, e.g., default reasoning. In this paper, we present a novel approach to probabilistic inductive logic programming based on Answer Set Programming (ASP) [9]. In contrast to ex- isting approaches, it provides a unified framework for probabilistic inference as well ⋆ This research is sponsored by Science Foundation Ireland (SFI) grant No. SFI/12/RC/2289