Proposta de uma métrica para avaliação do nível de concordância de classificações paralelas de entidades e decisão sobre revisão final a partir de revisões paralelas de um mesmo texto

Abstract

A major problem of recognizing named entities is the extraction of accurate information from texts that help humans in their daily processes. One such process of information extraction is the revisions of texts that occur in some domains, such as educational and police. These reviews take time to be done and are prone to errors. This term paper aims to decide about a final revision using the best ratings of parallel revisions of the same text made by one or more reviewers. An important step is to create a metric that assists in choosing the ratings and generates an agreement value that can be analyzed. A bibliographic study of existing metrics that may assist in the creation of the new metric is also made.