Evaluation of similarity measures for knowledge profiles from an expert directory: A field study

Expert directories describe the knowledge of the staff. From field data of an implementation of such an expert directory we calculated weighted connections between employees based on their shared topics. These relations will be used for recommending people with similar profiles. Taking into account that the field data has unequal distributions in various manifestations we looked at four different similarity measures to refine our recommendations. Our findings show the importance of using mathematical inverse calculations when dealing with unequal distributions. We found a way of benchmarking our similarity measures using group membership as distinctive indicator.