Key Phrase to Text Similarity, Clustering, and
   Interpretation in Hierarchical Ontologies

                                    Boris Mirkin

Applied Mathematics and Informatics, National Research University Higher School of
                              Economics Moscow
    Computer Science and Information Systems, Birkbeck University of London
                               BMirkin@hse.ru



      Abstract. Scoring similarity between key phrases and unstructured texts
      is an issue which is important in both information retrieval and text
      analysis. Researchers from the two fields use different scoring functions,
      although clear delineation between the two still is lacking. We use suffix
      tree based score expressing the average conditional probability of a sym-
      bol in a common substring. Usually, a domain taxonomy serves as the
      source of key-phrases. Given a set of entities, such as texts or projects or
      working groups, one can derive clusters of key-phrases using key-phrase-
      to-entity scores. The clusters represent common themes in the meaning
      of texts or in activities of working groups. To interpret them, the domain
      ontology should be used. If the ontology is a rooted tree, a lifting method
      is proposed to find the most parsimonious interpreting head subject(s),
      up to a few gaps and offshoots. Some applications and application is-
      sues are considered. The work is being conducted jointly with T. Fenner
      (London), S. Nascimento (Lisbon) and E. Chernyak (Moscow).