Similarity learning in nearest neighbor and application to information retrieval

摘要: Many people have tried to learn Mahanalobis distance metric in kNN classification by considering the geometry of space containing examples. However, similarity may an edge specially while dealing with text e.g. Information Retrieval. We proposed online algorithm, SiLA (Similarity learning algorithm) where aim is a (e.g. cosine measure, Dice and Jaccard coefficients) its variation eSiLA we project matrix learnt onto cone positive, semidefinite matrices. Two incremental algorithms been developed; one based on standard rule other symmetric version. can be used Retrievalwhere performance improved using user feedback.