As the usage of social networks grows day by day, a single person can reach hundreds or thousands of people in a minute. Microblogging is the new era of social communication, which can be used anywhere thanks to mobile phones. People spend hours and use social networks extensively, expressing their feelings, interests and dislikes. If this data can be extracted and analyzed effectively; useful items, news or people can be recommended. There are high number of studies that extract keywords from texts in order to obtain such information, however, microblogs have noisy text blocks, and hence regular text extraction algorithms fail to produce successful results. In this work, we propose a new approach, sentiment supported hybrid TF-IDF, in order to extract keywords to represent a user's profile more effectively. According to experimental results conducted under 50 different twitter accounts with 3 human judges, the proposed approach outperforms previous similar techniques in terms of profile constructions through keywords.