Uso de contextos temporais para classificação de documentos

Abstract

Due to the increasing amount of information being stored and accessible throughthe Web, Automatic Document Classification (ADC) has become an important research topic. ADC usually employs a supervised learning strategy, where we first build a classification model using pre-classified documents and then use it to classify unseen documents. One major challenge for ADC in many scenarios is that the characteristics of the documents and the classes to which they belong may change over time, since new documents are created, new information rise, new terms also are introduced and, consequently, the class definitions may change. Despite the potential quality reduction in the classification models associated with temporal-related changes, most of the current techniques for ADC are applied without taking into account the temporal evolution of the collection of documents. As we will see in this work, an important challenge in building classifiers is to deal with this temporal evolution.The two main hypotheses of this dissertation are: (1) the temporal evolution of the document collections significantly affects the performance of automatic document classifiers, and (2) the dimensions that compose this temporal evolution can be taken into account in order to build better classifiers, which are more efficient and effective. Thus, the goal of this thesis is to characterize, quantitatively and qualitatively, the impact of the temporal evolution of document collections on automatic document classifiers, identifying the dimensions that compose this impact. Moreover, based on this characterization, the goal is to propose alternate strategies that can be used to minimize the challenges associated with the temporal evolution of the collections on automatic document classifiers.The main expected contributions of this thesis are: (1) demonstration and quantification of the temporal evolution and how this affects automatic document classifiers, (2) identification of the effect dimensions that compose the temporal evolution, (3) quantification and qualification of each dimension identified, (4) design of a model to select contexts of the training set that minimize the temporal effects, and (5) the validation of this model using different document collections and classification algorithms.