Document Annotation by Semantic Roles

Abstract

This study examines the annotation of documents with respect to semantic roles. The data for thematic models is formed based on the standard bag-of-words hypothesis. The question of the number and uniqueness of themes in modeling remains unresolved. A key reason cited is the lack of unified criteria for evaluating the quality of clustering. There have been attempts to prove the uniqueness of themes through experimental evidence using a set of regularization criteria based on the results of document grouping. The absence of such proof is explained by the fact that the number of groups is initially set as a free parameter. Currently, five types of annotation are applied for the unification of documents: metatextual, morphological, syntactic, accentual, and semantic. This paper employs semantic annotation by identifying actants within sentences—noun phrases that denote participants in a situation and their semantic roles. It is proposed that, in defining semantic roles in the Uzbek language, the unification of terms should be conducted in relation to specific subject areas. An example of document annotation using semantic roles is provided, along with a determination of their similarity based on cosine metric.