The tweet contextualization task aims at providing an automatic readable summary explaining a given tweet. As tweets are very short documents, bound to 140 characters, and not always written maintaining proper spellings, there is indeed a need for such a task. This article describes a semantic tweet expansion approach for the tweet con-textualization task based on Wikipedia as an external knowledge source. This approach consists in two major phases, namely: the first is the generation of the candidate terms, from Wikipedia. While the second is the selection of the most-related terms. To achieve this latter, we propose a semantic relatedness measure based on the Explicit Semantic Analysis and association rules mining. The e↵ectiveness of our approach is proved through an experimental study conducted on the INEX 2014 collection. Our results have outperformed of the runs issued from INEX 2014.