The present invention relates to a tensor-based meta recommendation method, which expresses existing reports as a tensor which is multi-dimensional mode data, performs a matrix decomposition for tensor data of the reports, and recombines a decomposed matrix to recommend a new report, comprising the steps of: (a) expressing existing reports as a tensor, which is multi-dimensional mode data; (b) performing a matrix decomposition for tensor data of the reports using a coordinate descent for tensor factorization (CDTF); (c) calculating similarity between metas through a feature vector learned through a recombination process, and clustering semantically similar metas; and (d) recommending a new report by recombining a decomposed matrix for a user query, and simultaneously recommending the report and the metas. Accordingly, the tensor data of the reports is performed with the matrix decomposition, and the decomposed matrix is repeatedly recombined, thereby increasing efficiency in case of repeated update, and more accurately searching for metadata.;COPYRIGHT KIPO 2018
展开▼
