Distributional Term Representations for Short-Text Categorization

Abstract

Abstract. Everyday, millions of short-texts are generated for which effective tools for organization and retrieval are required. Because of the tiny length of these documents and of their extremely sparse representations, the direct application of standard text categorization methods is not effective. In this work we propose using distributional term representations (DTRs) for short-text categorization. DTRs represent terms by means of contextual information, given by document occurrence and term co-occurrence statistics. Therefore, they allow us to develop enriched document representations that help to overcome, to some extent, the small-length and high-sparsity issues. We report experimental results in three challenging collections, using a variety of classification methods. These results show that the use of DTRs is beneficial for improving the classification performance of classifiers in short-text categorization.

Hybrid Attention Networks for Chinese Short Text Classification

“…Cabrera et al [2] proposed a semi-supervised short text classification method, Distributional Term Representations (DTRs), which enriched document representations based on contextual information that overcame the small-length and high-sparseness short text to some extent.…”

Generación y enriquecimiento automático de recursos léxicos para el análisis de sentimientos

“…La idea de esta representación distribucional es que la semántica de un término t i puede revelarse a partir de otros términos con los que co-ocurre dentro de un documento de la colección [2,9]. De esta forma, cada término en el conjunto de términosúnicos (vocabulario total) de la colección de documentos T , t i ∈ T , es representado con un vector de pesos t i =< w 1 , w 2 , ..., w n > donde w j representa la contribución del término j a la descripción semántica de t i como lo indica la siguiente fórmula:…”

Early text classification: a Naïve solution

“…These extensions aim at helping naïve Bayes to deal with missing information, usually, at the attribute level. For instance by equipping the classifiers with mechanisms to work under highlysparse representations (e.g., in short text categorization) [15,2,7,19]. These methods are mostly based on smoothing attribute-class probabilities and often use co-occurrence statistics.…”

“…This paper motivates further work on extending this model for early text classification. For instance, one can define/modify adaptive priors that change as the value of t increases; we can implement the same idea with methods that take into account termdependencies (see e.g., [6,17,20]) in order to increase the predictive power of the classifier; also one can adopt advanced/alternative smoothing techniques to account for partial and missing information properly [15,2,7]; as well as many other possibilities. The main goal of this paper is to show that naïve Bayes can be used for early text classification and that its performance is competitive with the single existing solution to this problem.…”