Large-scale hierarchical text classification without labelled data

Abstract

The traditional machine learning approaches for text classification often require labelled data for learning classifiers. However, when applied to large-scale classification involving thousands of categories, creating such labelled data is extremely expensive since typically the data is manually labelled by humans. Motivated by this, we propose a novel approach for large-scale hierarchical text classification which does not require any labelled data. We explore a perspective where the meaning of a category is not defined by humanlabelled documents, but by its description and more importantly its relationships with other categories (e.g. its ascendants and descendants). Specifically, we take advantage of the ontological knowledge in all phases of the whole process, namely when retrieving pseudo-labelled documents, when iteratively training the category models and when categorizing test documents. Our experiments based on a taxonomy containing 1131 categories and widely adopted in the news industry as a standard for the NewsML framework demonstrate the effectiveness of our approach in these phases both qualitatively and quantitatively. In particular, we emphasize that just by taking the simple ontological knowledge defined in the category hierarchy, we could automatically build a large-scale hierarchical classifier with reasonable performance of 67% in terms of the hierarchy-based F-1 measure.

A new term‐weighting scheme for text classification using the odds of positive and negative class probabilities

“…In particular, as big data analysis continues to be a major research trend, unsupervised or semisupervised learning is playing an important role in TC (Gliozzo, Strapparava, & Dagan, 2005;Ha-Thuc & Renders, 2011;Ko & Seo, 2009). Many new research issues remain for TC; however, these also are based on the term-weighting schemes of supervisedlearning-based TC.…”

POWDER and the multi million-triple store

“…• Thematic annotation with terms from the IPTC hierarchy [5] • Thematic clustering of semantically homogeneous document fragments (hereafter, segments) into classes corresponding to the incident they report on.…”

“…5 The framework is fully extensible and configurable with respect to storage mechanisms, inference engines, RDF file formats, query result formats, and query languages.…”

Cloud-based Textual Analysis as a Basis for Document Classification

“…To solve this problem [6] developed a system to hierarchically classify unlabelled data. As already mentioned, classifying data manually is extremely expensive and slows the classification process down.…”

“…Additionally, it grows to be an inefficient approach as with larger datasets the number of categories can exceed to thousands, of which each needs to be represented by a sufficient number of labelled documents. The system solves this issue by using ontological knowledge and by searching 'pseudo-relevant documents on the Web' [6]. With the ontology it is possible to create a hierarchical model including the context of ancestors among different classes.…”