Personalization in Collaborative Fusion based Enterprise Information retrieval

Due to data spread across various heterogeneous data stores, information retrieval in Enterprise data stores is always challenging compared to web based retrieval systems. We have proposed a collaborative fusion based information retrieval in [1] using the observation on similar users tends to prefer similar search results. The solution applied three dimensions of user similarity, document similarity and user to documents affinity to a collaborative information fusion based retrieval. The work also proposed active feedback based search result revision to get highly relevant results. But the work did not have any provision for personalization and could not handle the cold start problems. Without consideration for cold start problems, the user to document affinity cannot be modeled accurately as the result, the collaborative fusion process is affected. In this work, we improve our earlier solution of collaborative fusion based information retrieval with consideration for user personalization and solution for cold start problems. The solution is based on query refinement using the information hidden in enterprise messaging systems. A user profile is built as vector of concepts using the information in enterprise messaging systems and this user profile concept vector is used to refine the query in way to personalize the results and avoid cold start problems. Compared to approach in [1], the proposed query refinement based personalization is able to increase the relevancy accuracy by 10% as obtained from experimental results.