Abstract : Social media sites such as Twitter 1 and Facebook 2 have emerged as powerful means of communication that allow people to exchange information about their daily activities, latest news or real-world events. Aside social interactions among users, social medias are expected to provide added value services in a variety of domains (e.g sentiment analysis, trend analysis and event detection). Detecting events on social medias poses new challenges due to the sparsity and the informal nature of social media posts. One of the main challenges in detecting events in social media is to differentiate event and non event messages. To face this challenge, we propose to take advantage from the knowledge that can be extracted from the Linked Opened Data (e.g. DBpedia) to enrich the short textual messages with contextual information brought by the presence of named entities. We evaluate our approach on two gold-standard datasets and the preliminary results show that exploiting the ontological categories of the named entities has a positive impact on the classification output.