Expertise Modeling Using Hierarchical Classification And Wikipedia Knowledge

Abstract: The traditional text classification approach involves classifying document into a class where classification hierarchy is limited to one level. However, the real-world problems are more complex and could be related to hierarchical structure and therefore there has been numerous research in hierarchical classification and algorithms. Millions of enthusiastic researchers contribute in the form of research articles, books and conference dissertations, and abundance of research data makes finding an expert a non-trivial task. We define expertise modeling as profiling an expert, a knowledgeable person in one or more domains, based on evidence from research articles into one or more research topics. A hierarchical classifier is trained on titles and abstracts of research articles and predicts the research topic(s) for a given document as a title and an abstract. We have used traditional Bag-of-Words (BOW) representations of the text which is enriched using a semantic knowledge from Wikipedia’s concepts and categories. The representations based on concepts and categories are called Bag-of-Concepts (BOC) and Bag-of-Categories (BOK) respectively. For each of these representation a hierarchical classifier is trained and an output from each of these hierarchical classifiers are combined using consensus methods to output a class. In reality, a research articles can belong to multiple research areas and therefore two methods to multi-label a research article is proposed. We evaluate and compare the performance of hierarchical model with a baseline traditional classifier, and using different training set amount and evaluation parameters such as precision, recall, and f-measure. The combined outputs from hierarchical classifiers, BOW, BOC, and BOK is compared with a traditional classifier and hierarchical classifier based on BOW. The results from various approaches, comparison of performance of different hierarchical classifiers and current issues are also discussed. Speaker Bio: Afiz Momin received his under-graduate major in Bioinformatics and before joining graduate school at Dalhousie University, he has professional experience on various projects such as web and database technologies, big data and text analysis. Currently, he is involved in research in Visual Text Analytics Lab under supervision of Dr. Evangelos Milios. His thesis research is focused on expertise modeling using hierarchical classification and semantics from external sources such as Wikipedia. Afiz’s research interests mainly include software engineering, data science, text analysis and visualization.

More information at: https://www.dal.ca/news/events/2016/12/15/expertise_modeling_using_hierarchical_classification_and_wikipedia_knowledge.html