This study proposes an outlier detection model in text data stream. Text stream is an important variant of data stream clustering. It has many useful implementations such as trend analysis, detection and tracking of topics, recommendation of user, and outlier detection. Outlier detection detects events which are interesting to the user and perhaps can be used to trigger some actions. One challenge in outlier detection in text stream is that normal behavior can change and thus it should be possible to adapt the models to the changes. Therefore, detecting outlier in text stream is not a trivial task. This paper proposes a conceptual model to detect outliers in the text stream. The model contains four main phases namely pre-processing, text representation, feature selection, and outlier detection phase. In the first phase, tokenization, stop words removal and stemming will be used. An incremental term weighting for text stream representation will be proposed in the second phase. An online feature selection will be improved on phase number three. Finally, in the fourth phase, one of the swarm intelligence techniques will be improved to detect outlier in the text stream.