A framework for multiset merging

Information fusion is a research area that investigates how to combine information provided by independent sources into one piece of information. Several aspects of this topic have already been studied leading to, amongst others, aggregation operators in bounded lattices and merge functions of propositional belief bases. In this paper, information fusion is investigated in the context of coreferent objects, which are pieces of data in an information system, that refer to the same real world entity. The fundamental operator in our approach is a merge function that maps a multiset of coreferent objects onto a single object, which is called the solution. We investigate the specific case where objects themselves are multisets, which can be applied to problems such as Multi-Document Summarization (MDS) and fusion of duplicate graphs (e.g. XML documents). Our approach involves the definition of quality measures that express the correctness and the completeness of a given solution. We show how a solution can be found that optimizes a balance between correctness and completeness. Merge functions that result in such a solution are called f-optimal merge functions and we investigate their properties.