The work herein describes a system for automatic news category and keyphrase labeling, presented in the context of our motivation to improve the speed at which a user can find relevant and interesting content within an aggregation platform. A set of 12 discrete categories were applied to over 500,000 news articles for training a neural network, to be used to facilitate the more in-depth task of extracting the most significant keyphrases. The latter was done using three methods: statistical, graphical and numerical, using the pre-identified category label to improve relevance of extracted phrases. The results are presented in a demo in which the articles are pre-populated via News API, and upon being selected, the category and keyphrase labels will be computed via the methods explained herein.
View on arXiv