Research on the automatic construction of the resource space model for scientific literature

Abstract

The resource space model is a semantic data model to organize Web resources based on a classification of resources. The scientific resource space is an application of the resource space model on massive scientific literature resources. The construction of a scientific resource space needs to build a category (or concept) hierarchy and classify resources. Manual design suffers from heavy workload and low efficiency. In this thesis, we propose novel methods to solve the following two problems in the construction of a scientific resource space: 1. Automatic maintenance of a category hierarchy. A category hierarchy needs to evolve dynamically with new resources continually arriving so as to satisfy the dynamic re-quirements of the organization and management of resources. We propose an automatic maintenance approach to modifying the category hierarchy according to the hierarchical clustering of resources and show the effectiveness of this method by a series of comparison experiments on multiple datasets. 2. Automatic construction of a concept hierarchy. We propose a joint extraction model based on a deep neural network to extract entities and relations from scientific articles and build a concept hierarchy. Experimental results show the effectiveness of the joint model on the Semeval 2017 Task 10 dataset. We also implement a prototype system of the scientific resource space. The prototype system enables the comparative summarization on scientific articles. A set of novel comparative summarization methods based on the differential topic models (dTM) are proposed in this thesis. The effectiveness of the dTM-based methods is shown by a series of experimental results.