Desenvolvimento de um modelo de representação temporal para coleções de documentos

Abstract

This work aims to present the development of a new vector representation model for document collections, which captures the temporal aspect of the variation in the occurrence of terms over time. As a proof of concept, this new temporal vector representation model was included in the interactive visual framework ScienceView , which has as one of its features to generate dynamic multidimensional projections of scientific articles collections. The collections of scientific articles evolve over time, thus it is possible to extract the temporal aspect, and using it to improve the analysis applied to the projections. The importance of the terms was dynamically measured, based on the variation in the occurrence of the term over time, by extracting the time series of the occurrences of each term in the collection of scientific articles. The new model developed was tested in two collections of scientific articles: the CBR-ILP-IR document collection, which presents scientific articles from three research areas; and the base formed by the articles published by the researcher Alessandro Vespiginiani. Thus, a comparative study was carried out, both with the classic model already existing in ScienceView and the new temporal model. Investigating how the approach employed by the new vector representation model developed performs on the projection. Since the projection uses a representation model in its generation, the evaluation is possible through empirical and numerical analysis on the final projections, indirectly inferring how the new vector representation model performs and impacts the final projection.