The rapid and accurate identification of bot accounts in online social networks is an ongoing challenge. In this paper, we propose BOTTRINET, a unified embedding framework that leverages the textual content posted by accounts to detect bots. Our approach is based on the premise that account personalities and habits can be revealed through their contextual content. To achieve this, we designed a triplet network that refines raw embeddings using metric learning techniques. The BOTTRINET framework produces word, sentence, and account embeddings, which we evaluate on a real-world dataset, CRESCI2017, consisting of three bot account categories and five bot sample sets. Our approach achieves state-of-the-art performance on two content-intensive bot sets, with an average accuracy of 98.34% and f1score of 97.99%. Moreover, our method makes a significant breakthrough on four content-less bot sets, with an average accuracy improvement of 11.52% and an average f1score increase of 16.70%. Our contribution is twofold: First, we propose a unified and effective framework that combines various embeddings for bot detection. Second, we demonstrate that metric learning techniques can be applied in this context to refine raw embeddings and improve classification performance. Our approach outperforms prior works and sets a new standard for bot detection in social networks.