The present invention is related to a method for identifying in a given semantic graph of linked concepts a set of concepts matching the content of a text, the semantic graph of linked concepts being stored in at least one equipment connected to a network, the methods comprising of the following steps: inputting a sample text in natural language at a workstation connected to the network; performing a semantic analysis of the sample text to extract a set of initial concepts belonging to the given semantic graph; accessing the equipment to identify a list of candidate concepts in the given semantic graph, the candidate concepts being the concepts of the semantic graph within a given maximum distance according to a given distance function away from an initial concept; calculating for each candidate concept a proximity score representative of the distances between the candidate concept and each initial concept; identifying as matching concepts the candidate concepts with the highest proximity score.