Multi-View Multi-Label Online Learning for Education Topic Recommendation

In current education field, in order to learn and teach some important knowledge, students and teachers have to pay more attention to search some related articles. This is very time-consuming and is not conducive to improving efficiency. Indeed, fast to retrieve the knowledge, evaluate its trends, and extract the scope of topics is beneficial to students and teachers. This operation can also recommend the future scope to them. Furthermore, since the publication of these related articles is sustained, thus the topics and corresponding scopes should be updated ceaselessly. Thus, in this study, we treat the extraction results from title and abstract as two views (keyword-level semantic view and sentence-level semantic view) of the articles and regard scopes of articles (i.e., topics) as multiple labels firstly and then develop a multi-view multi-label online learning for education topic recommendation (MVML-ETR). Here, in MVML-ETR, we adopt an auto-encoder-decoder machine to handle the online stream of articles and in order to validate its effectiveness, we select ArXiv data set for experiments and some SOTA algorithms for comparison.