Ontologies are known as a quality and functional model, allowing meta data representation and reasoning. However, their maintenance plays a crucial role as ontologies may be misleading if they are not up to date. Currently, this work is done manually, and raises the problem of expert subjectivity. Therefore, some works have developed maintenance tools but none has allowed a precise identification of the relations that could link concepts. In this paper, we propose a new fully generic approach combining sequential patterns extraction and equivalence classes. Our method allows to identify terms from textual documents and to define labelized association rules from sequential patterns according to relevance and neighborhood measures. Moreover, this process proposes the placement of the found elements refined by the use of equivalence classes. Results of various experiments on real data highlight the relevance of our proposal.