Many applications that allow text based communication between users are troubled with malicious content. This thesis presents a system for detecting such behaviour in an E-commerce application. The system is based on an algorithm for anomaly detection which is trained using messages sent between users in the application. Preprocessing of the text is performed using the NLP-toolbox Glove. The resulting word embeddings are used to create numerical representations of messages, which are then used as input to a clustering algorithm based on K-means. Vectors positioned far away from existing clusters were considered anomalies. This report assesses performance of this system, and relates this to the performance achieved with an existing approach.