Concept-Based Label Embedding via Dynamic Routing for Hierarchical Text Classification

Abstract

Hierarchical Text Classification (HTC) is a challenging task that categorizes a textual description within a taxonomic hierarchy. Most of the existing methods focus on modeling the text. Recently, researchers attempt to model the class representations with some resources (e.g., external dictionaries). However, the concept shared among classes which is a kind of domain-specific and fine-grained information has been ignored in previous work. In this paper, we propose a novel concept-based label embedding method that can explicitly represent the concept and model the sharing mechanism among classes for the hierarchical text classification. Experimental results on two widely used datasets prove that the proposed model outperforms several state-of-theart methods. We release our complementary resources (concepts and definitions of classes) for these two datasets to benefit the research on HTC.

Hierarchical Text Classification and Its Foundations: A Review of Current Research

“…CNs have been applied to HTC tasks for their ability to model latent concepts and, hence, capture the latent structure present in the label space. It is expected that a better understanding of the labels' organization, such as modeling the "team sports" and "ball sports" concepts, can be effectively exploited to improve decision-making during classification [87,88].…”

“…In methods able to categorize labels at different levels of the hierarchy separately, many authors choose to showcase the accuracy score at each separate level, as well as a single overall score [87,113]. The overall score is the one obtained by classifying the last level of the hierarchy given the (possibly incorrect) predictions of the parent classes.…”

“…As another example of the application of CNs to HTC, Wang et al [87] propose the Concept-based Label Embedding via Dynamic routing (CLED), in which a CN is used to extract concepts from text documents. Concepts can be shared between parent and child classes and can thus be used to support classification based on hierarchical relations.…”

Short text classification model based on dynamic routing and CNN with attention mechanism

“…However, Capsule-B used averaging pooling to fuse the three scale feature maps before classification, which seems to violate its design. Wang et al [6] exploited the dynamic routing algorithm of capsule networks to explore a concept-based dynamic routing label embedding method for hierarchical text classification, using dynamic routing to simulate concept sharing between subclasses and parents, which not only retains the advantages of capsule networks but also shortens the training time.…”

Hierarchical Label Generation for Text Classification

“…To solve the HTC task, recent work has focused on enhancing label embeddings with a taxonomic hierarchy (Cao et al, 2020;Zhou et al, 2020;Wang et al, 2021) or considering a sequential classification approach (Rivas Rojas et al, 2020;Yang et al, 2018Yang et al, , 2019 that leverages a Seq2Seq framework to capture the label hierarchy. Despite the previous methods being successful, their approaches classify labels sequentially by choosing them from the predefined label set in the training dataset.…”