Recommending similar patients in a social network of patients using medical knowledge bases

Abstract

The ever-increasing volume and complexity of information flowing into our daily lives newlinevia web and social media challenge us with our limited information-processing newlinecapabilities to adapt to such abundant information so as to reach at the most interesting newlineproducts/services of relevance. Also called as Information Overload problem , newlineRecommender Systems pose as a solution to this. Recommender Systems suggest users newlineabout products, services, or similar users they may be interested in, by taking into newlineaccount or predicting their profiles, tastes, priorities or goals. Recommender Systems newlinehave found their way into the Healthcare domain targeting patients and caregivers. newlinePatients of today are information savvy, ever curious to know more about what they are newlinesuffering from, and how to overcome it by getting the right information at right time. In newlinethis process, they seek solace in online communities as well as social networks which newlinehelps them to know about other patients with similar conditions sharing their newlineexperiences, and providing emotional support to deal with the sufferings. Such patient newlinecentric social networks use Recommender Systems in the background to recommend newlinepatients to other target patients. However, current Recommender Systems have many newlinelimitations such as handling unstructured data, sparsity in user profile element spaces, newlinelack of flexibility to incorporate contextual factors into the recommendation processes, newlineand most importantly, the validity of the recommendation by a proven knowledge base. newlineIn the case of a patient-centric social network, the user profile data contains newlinedemographics, type of diseases, symptoms, drugs, and laboratory tests, etc. which are newlinefed by the patients themselves without the help of any medical professionals. The type newlineof data is mostly unstructured due to the use of imprecise terms, colloquial terms, newlinemisspellings, alternative terms, and localized terms. The use of exact medical terms in a newlinepatient s profile cannot be guaranteed. This directly affects the quality of the newlinerecommendation.