Figure 1

Illustration of USI annotation process according to four variants. During a first phase (A), documents similar to the document d _i to be annotated are identified. Here, this is done using PMRA*. The set of the k-NNs of d _i , denoted K, is identified from this ordered list. It will support the annotation calculus for d _i . Here k=3. This selection may be done by only taking the k top ranked documents obtained by PMRA* (B top) or after interaction with the user on a semantic map (B bottom). The set A ₀ of candidate concepts for d _i annotation is obtained either by taking the union of concepts annotating at least one document (A ₀) or two documents (A _{0+f i l t e r} ) of K. This candidate set is then processed to find the medial annotation of K that will be proposed to annotate d _i (C).