The problem of belief fusion, or the “infoglut problem,” is discussed in this paper in terms of multimodal logics. The main intuition is that each modal operator corresponds to an agent (or source of information), and that agents are arranged in a total order according to their level of reliability. Modalities are also given for groups of agents. Given the multiplicity of sources of information, mechanisms to prevent inconsistent beliefs when merging pieces of information from different sources must be provided. The paper presents two mechanisms: level cutting fusion and level skipping fusion. In level cutting fusion, all levels following a level introducing an inconsistency are not included in the fusion, thus the level introducing the inconsistency induces a threshold on the hierarchy of sources of information. In level skipping fusion, levels introducing inconsistencies are not included in the fusion. Examples for different applications, including information security and diagnosis, illustrate the two fusion methodologies.
The approaches are semantically based, where the semantics is given in terms of Kripke possible worlds semantics. Two sound and complete axiomatizations, one for each fusion technique, are given. A decision procedure and automated reasoning system is defined using prefix tableaux [1,2]. The tableaux proof system is also used to provide the complexity of the logics (polynomial space (PSPACE) complete).
The information fusion framework presented in the paper will be of interest to scholars working on epistemic and doxastic logics and multiagent systems, as well as researchers studying information retrieval and information agents.