Outlier detection is a primary step in many data mining applications. An outlier is an abnormal individual from a population, which usually leads poor accuracy in models. Medical literatures are the most reliable resources for researchers to know the progress in their research areas and latest contributions from others. Traditional keyword search retrieves all the text data that contain the keywords you have specified. That is great as far as it goes, but people still have to read all those literatures to find out whether they actually contain any information that is relevant to your search. While text mining is aware of real text meanings to identify facts, relationships and assertions that would otherwise remain buried in a mass of all literatures. Therefore, a strategy is planned that using keywords to collect as many keywordrelated literatures as possible, and then applying our proposed algorithm to eliminate the outlier literatures so that the data quality is improved. Experiments show that up to 80% outliers literatures are detected from the normal literatures.