|•||User relevance feedback proves to be very effective in improving retrieval performance. This means that the natural interactivity of the retrieval process can be exploited using additional knowledge about the distribution of relevant and non-relevant information objects. This knowledge can be used to modify query terms as well as their weights in order to place the relevant material at the beginning of the ranking list (cf. Belkin 1993).|
|•||Many systems have led to comparable quality, however the overlap of their result sets has been rather small. Some researchers have attempted to exploit this fact by introducing poly-representation techniques of retrieval objects and achieved good results. The so called data fusion is based on the idea that different object representation corresponds better to individual users' interests and perspectives (cf. Fox & Shaw 1994).|
MIMOR integrates these two approaches. It can be seen as an additional layer within an IR system managing the poly-representation of queries and documents by selecting appropriate methods for indexing and matching. Through learning from relevance feedback information, the model adapts itself during an initial phase by assigning weights to the different method-object combinations. MIMOR is not limited to text documents as many common IR applications, but rather it is open to data types like structured data and multimedia objects.
Each Information Retrieval System calculates a retrieval status value (RSV) for each document by comparing it to the query.
|Doc. 1||Doc. 2||Doc. 3||Doc. 4||...|