In this paper, we address the problem of classifying documents available fromthe global network of (open access) repositories according to their type. Weshow that the metadata provided by repositories enabling us to distinguishresearch papers, thesis and slides are missing in over 60% of cases. Whilethese metadata describing document types are useful in a variety of scenariosranging from research analytics to improving search and recommender (SR)systems, this problem has not yet been sufficiently addressed in the context ofthe repositories infrastructure. We have developed a new approach forclassifying document types using supervised machine learning based exclusivelyon text specific features. We achieve 0.96 F1-score using the random forest andAdaboost classifiers, which are the best performing models on our data. Byanalysing the SR system logs of the CORE [1] digital library aggregator, weshow that users are an order of magnitude more likely to click on researchpapers and thesis than on slides. This suggests that using document types as afeature for ranking/filtering SR results in digital libraries has the potentialto improve user experience.
展开▼
