The next generation of educational applications need to significantly improve the way feedback is offered to both teachers and students. Simply determining coarse-grained entailment relations between the teacher's reference answer as a whole and a student response will not be sufficient. A finer-grained analysis is needed to determine which aspects of the reference answer have been understood and which have not. To this end, we propose an approach that splits the reference answer into its constituent propositions and two methods for detecting entailment relations between each reference answer proposition and a student response. Both methods, one using hand-crafted features and an SVM and the other using word embeddings and deep neural networks, achieve significant improvements over a state-of-the-art system and two alternative approaches.

To have a more meaningful impact, educational applications need to significantly improve the way feedback is offered to teachers and students. We propose two methods for determining propositional-level entailment relations between a reference answer and a student's response. Both methods, one using hand-crafted features and an SVM and the other using word embeddings and deep neural networks, achieve significant improvements over a state-of-the-art system and two alternative approaches.

Truly effective and practical educational systems will only be achievable when they have the ability to fully recognize deep relationships between a learner’s interpretation of a subject and the desired conceptual understanding. In this paper, we take important steps in this direction by introducing a new representation of sentences – Minimal Meaningful Propositions (MMPs), which will allow us to significantly improve the mapping between a learner’s answer and the ideal response. Using this technique, we make significant progress towards highly scalable and domain independent educational systems, that will be able to operate without human intervention. Even though this is a new task, we show very good results both for the extraction of MMPs and for classification with respect to their importance.

With the exponential growth of scholarly data during the past few years, effective methods for topic classification are greatly needed.Current approaches usually require large amounts of expensive labeled data in order to make accurate predictions.In this paper, we posit that, in addition to a research article's textual content, its citation network also contains valuable information.We describe a co-training approach that uses the text and citation information of a research article as two different views to predict the topic of an article.We show that this method improves significantly over the individual classifiers, while also bringing a substantial reduction in the amount of labeled data required for training accurate classifiers.

Given the large amounts of online textual documents available these days, e.g., news articles, weblogs, and scientific papers, effective methods for extracting keyphrases, which provide a high-level topic description of a document, are greatly needed.In this paper, we propose a supervised model for keyphrase extraction from research papers, which are embedded in citation networks.To this end, we design novel features based on citation network information and use them in conjunction with traditional features for keyphrase extraction to obtain remarkable improvements in performance over strong baselines.