SJORS - A Semantic Recommender System for Journalists

Abstract

In recent years, the world of journalism has undergone a transformation motivated by the growing use of digital devices for news consumption, the change in the advertising model, and the financial crisis. This process has led to a scenario where the newspaper workforce has been reduced, with journalists being expected to undertake more complex work due to the tough competition among different media, and the immediacy expected by customers. In this context, we introduce SJORS, a recommender System that identifies the top-N most relevant wire news for a journalist at any given time, which he/she can use in composing news articles to publish. Regarding the contributions of this work: - We have provided an in-depth study of a particular domain of the recommendation of wire news to journalists in a newsroom. - We have proposed a wire news recommender system that considers the recency of news and language ambiguity. This is done through a detailed analysis of the journalist's past activity. The system combines the novelty of wire news in the media and their lifespan in the editorial system with a semantic analysis that utilises word embeddings and the Semantic Web. - We have conducted a study in a real newsroom context, generating a dataset with information about the work in a newsroom of 20 journalists during one month interacting with more than 150,000 wire news. The experiments have validated the performance of SJORS with 0.606 at nDCG and 0.816 at MRR.