Term Network Approach for Transductive Classification

Abstract

Abstract. Transductive classification is a useful way to classify texts when just few labeled examples are available. Transductive classification algorithms rely on term frequency to directly classify texts represented in vector space model or to build networks and perform label propagation. Related terms tend to belong to the same class and this information can be used to assign relevance scores of terms for classes and consequently the labels of documents. In this paper we propose the use of term networks to model term relations and perform transductive classification. In order to do so, we propose (i) different ways to generate term networks, (ii) how to assign initial relevance scores for terms, (iii) how to propagate the relevance scores among terms, and (iv) how to use the relevance scores of terms in order to classify documents. We demonstrate that transductive classification based on term networks can surpass the accuracies obtained by transductive classification considering texts represented in other types of networks or vector space model, or even the accuracies obtained by inductive classification. We also demonstrated that we can decrease the size of term networks through feature selection while keeping classification accuracy and decreasing computational complexity.

GA-TCTN: a framework for hyper-parameter optimization and text classification using transductive semi-supervised learning through term networks

“…Transductive Classification through Term Networks (TCTN) is an interesting approach for text classification using terms (words) of those texts. According to [13], TCTN can be divided into four main steps: (1) term network generation, (2) initial relevance score setting, (3) relevance score propagation, and (4) text classification. In the next subsections, we present details of these steps.…”

“…In this article, we consider the Top-k approach since networks with these characteristics have been obtained the best results in semi-supervised learning [21]. However, k is an important hyper-parameter that has impact in the classification performance [5,13].…”

“…Although most network representation for transductive learning relies on term frequency to build relations between documents and terms in bipartite or more complex networks, the term network approach proposed in [13] has the ability to represent useful relations among terms in a way to improve relevance scores of the terms and consequently improve the classification performance. This approach, named Transductive Classification through Term Networks (TCTN) provided results that surpass the classification performance of other transductive learning algorithms based on networks (document and bipartite) or based on vector space model.…”

“…On the other hand, TCTN has some hyper-parameters that needs to be tuned to produce better classification results. In [13], TCTN was carried out with manually defined parameters, which limits possibilities of maximizing TCTN's classification performance. Besides, this type of tuning requires a manually defined set of parameters, which can also impair obtaining better classification performances.…”

“…Therefore, to improve the benefits of TCTN and also obtain better classification performances, in this article, a method based on Genetic Algorithm (GA) to tune TCTN parameters (GA-TCTN) is proposed. Also, our method allows searching in a larger search space, i.e., while in [13] one of hyper-parameters is α ∈ [0.1, 0.3, 0.5, 0.7, 0.9], in our method α ∈ [0, 1]. Thus, GA-TCTN allows the use of more specific and non-trivial hyper-parameters values, being more flexible and practical than TCTN with manually defined parameters.…”