Learning When to Classify for Early Text Classification

Abstract

Abstract. The problem of classification in supervised learning is a widely studied one. Nonetheless, there are scenarios that received little attention despite its applicability. One of such scenarios is early text classification, where one needs to know the category of a document as soon as possible. The importance of this variant of the classification problem is evident in tasks like sexual predator detection, where one wants to identify an offender as early as possible. This paper presents a framework for early text classification which highlights the two main pieces involved in this problem: classification with partial information and deciding the moment of classification. In this context, a novel approach that learns the second component (when classify) and an adaptation of a temporal measurement for multi-class problems are introduced. Results with a classical text classification corpus in comparison against a model that reads the entire documents confirm the feasibility of our approach.

Early Classification of Time Series is Meaningful

“…One reason for using the UCR datasets is simply that everyone does this, and that it helps making comparisons between the various proposed approaches. Another reason is that, in fact, applications where ECTS is useful exist [30,2,24,10,28,14,8] but they have still to realize their full potential. Companies are becoming more and more aware of the problem and are now starting to integrate ECTS into their projects.…”

“…That the research carried out so far has not lead to applications is to our opinion a mistaken critic [30,2,24,10,28,14,8]. Real applications are: i) either hidden from the academic world because of their sensitive nature; ii) or under development, as this is a relatively new problem whose awareness is recent.…”

“…An increasing number of applications require to recognize the class of an incoming time series as quickly as possible without unduly compromising the accuracy of the prediction [30,2,24,10,28,14,8]. For example, in emergency wards of hospitals [17], in control rooms of national or international electrical power grids, in government councils assessing emergency situations, in all kinds of contexts, it is essential to make timely decisions in absence of complete knowledge of the true outcome (e.g.…”

Early Detection of Sexual Predators in Chats

“…TEASER (Schäfer and Leser, 2020) or ECTS (Xing et al, 2012). However, there exists a key difference that prevents us from using such methods directly: An eSPD System never classifies a chat as nongrooming as long as there are still messages left (or expected), while an eTSC system at some stage might decide that it is safe to stop controlling the chat (Loyola et al, 2018). This opens the door to malicious attacks by using long and harmless openings in grooming attempts.…”

A text classification framework for simple and effective early depression detection over social media streams

“…Finally, (Loyola et al, 2018) considers the decision of "when to classify" as a problem to be learned on its own and trains two SVMs, one to make category predictions and the other to decide when to stop reading the stream. Nonetheless, the use of these two SVMs, again, hides the reasons behind both, the classification and the decision to stop early.…”