Personalization and Cold Start Problems in Information Retrieval

The retrieval of information within enterprise data stores is consistently challenging due to the dispersion of data across diverse and heterogeneous storage systems, in contrast to the relatively simpler web-based retrieval systems. This paper introduces an enhanced approach to information retrieval, building upon our prior work on collaborative fusion-based retrieval presented in [1]. Our initial solution leveraged the insight that similar users exhibit preferences for comparable search results. It incorporated three key dimensions: user similarity, document similarity, and user-to-document affinity, forming the basis for collaborative information fusion in retrieval processes. The research proposed a methodology involving active feedback for the revision of search results, aiming for highlighted relevance. However, a notable limitation of the previous work was the absence of provisions for personalization and an inability to effectively handle cold start problems. This absence impedes the accurate modeling of user-to-document affinity, thereby impacting the collaborative fusion process. In response, this paper addresses these limitations by refining our collaborative fusion-based information retrieval solution. The enhancements specifically focus on incorporating user personalization and providing a resolution for cold start challenges. Through these improvements, we aim to augment the accuracy and adaptability of our information retrieval system in enterprise settings.