Nowadays, social networks are one of the most used means of communication. For example, the social network Twitter has nearly 100 million active users who post about 500 million messages per day. Sharing information on this platform is unique because messages are limited in characters number. Faced with this limitation, users express themselves briefly and use sometimes a hashtag that summarizes the general idea of the message. Nevertheless, hashtags are noisy data because they do not respect any linguistic rule, may have several meanings, and their use is not under control. In this work, we tackle the problem of hashtag context which may have useful applications in several fields like information recommendation or information classification. In this paper, we propose an original data cleaning method to extract the most relevant neighbor hashtags of a hashtag. We test our method with a dataset containing hashtags related to several topics (such as sport, music, technology, etc.) in order to show the efficacy and the robustness of our approach.