This paper presents the underlying methods and algorithms that support a potential solution for automatic recommendation of in-context media content applied to the media sector, developed in the scope of the magnet project. The objective of magnet is to develop a software platform to support journalists in the early phase of an article production by automatically resurfacing older content from previous articles, relevant for the current moment. To reach this objective, the intelligence of the platform is twofold: it is capable of understanding the topic of the activity being developed by the journalist, and it should prioritize content that has a similar context with the current situation being handled. The similarity of context is closely derived from the activities developed by other journalists, giving a strong emphasis on aspects of collaborative filtering. Topic and context identification are the two vehicles enabling the platform to automatically recommend older content in the media archives, and timely offer it to the journalists. A key innovation aspect of magnet project is the context-based search on top of the topic-oriented approach. The platform uses previously tested algorithms based on Vector Space Models and Context Modelling. These have been selected because they also contribute to both the transparency and explainability of the proposed solution, as this is a very important aspect regarding journalistic activity. The paper presents the baseline concept, the core algorithms, the software platform and the validation results.