Online course reviews have been an essential way in which course providers could get insights into students' perceptions about the course quality, especially in the context of massive open online courses (MOOCs), where it is hard for both parties to get further interaction. Analyzing online course reviews is thus an inevitable part for course providers towards the improvement of course quality and the structuring of future courses. However, reading through the often-time thousands of comments and extracting key ideas is not efficient and will potentially incur non-coverage of some important ideas. In this work, we propose a key idea extractor that is based on fine-grained aspect-level semantic units from comments, powered by different variations of state-of-the-art pre-trained language models (PLMs). Our approach differs from both previous topic modeling and keyword extraction methods, which lies in: First, we aim to not only eliminate the heavy reliance on human intervention and statistical characteristics that traditional topic models like LDA are based on, but also to overcome the coarse granularity of stateof-the-art topic models like top2vec. Second, different from previous keyword extraction methods, we do not extract keywords to summarize each comment, which we argue is not necessarily helpful for human readers to grasp key ideas at the course level. Instead, we cluster the ideas and concerns that have been most expressed throughout the whole course, without relying on the verbatimness of students' wording. We show that this method provides high and stable coverage of students' ideas.
Incorporating contrastive learning objectives in sentence representation learning (SRL) has yielded significant improvements on many sentence-level NLP tasks. However, it is not well understood why contrastive learning works for learning sentence-level semantics. In this paper, we aim to help guide future designs of sentence representation learning methods by taking a closer look at contrastive SRL through the lens of isotropy, contextualization and learning dynamics. We interpret its successes through the geometry of the representation shifts and show that contrastive learning brings isotropy, and drives high intra-sentence similarity: when in the same sentence, tokens converge to similar positions in the semantic space. We also find that what we formalize as "spurious contextualization" is mitigated for semantically meaningful tokens, while augmented for functional ones. We find that the embedding space is directed towards the origin during training, with more areas now better defined. We ablate these findings by observing the learning dynamics with different training temperatures, batch sizes and pooling methods.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.