Learning Term Taxonomy Relationship from a Large Collection of Plain Text

In this paper, we present a heuristic for labeling a given term a taxonomy label. Specifically, for a given term, our goal is to construct a model for determining an "is-a" relationship between the given term and an inferred concept. Such term-labelling problem is not new, but the existing solutions require semi-supervised training processing, e.g., supervised LDA, or rely on lexicographers, e.g., wordnet. The model construction cost becomes burdens for employing such semantic understanding capability in various emerging applications. Aiming at these issues, in this study, we present a lightweight approach with the following features. First, the proposed approach is unsupervised and take only pain text as inputs. Second, the proposed approach allows incremental model construction. Third, the proposed approach is simple but effective and computationally efficient in comparison with the existing solutions. We demonstrate these results through experiments by comparing our approach with DBpedia and employ the popular search terms as test set. From experiment results, we see that 30 percent improvement in accuracy can be achieved by the proposed approach.

I. Introduction

In this paper, we present a heuristic for annotating a given term a concept label. Specifically, for a given term, our goal is to construct a model for determining an is-a relationship (hyperonym-hyponym relationship) between the given term and an inferred concept. For example, if we are given “New York Yankees”, our goal is to label it as “a team.” Yet another example is to label “the Beatles” as “a band.” Such term-labelling problem is not new, but become more and more important than ever, as such problem is essential to many emerging applications, such as context-aware computing, user preference understanding, intelligent search engine constructing, and etc. However, the models constructed by the existing solutions require supervised or semi-supervised training over semi-structured data, e.g., supervised LDA and DBpedia[2], or rely on lexicographers, e.g., wordnet. The model construction become burdens for employing such semantic understanding capability in various emerging applications.