In one aspect, a computer system automatically identifies style issues in a source code base. A reference set for a known style issue includes source code examples that exhibit the style issue. The source code examples in the reference set are compared to the source code base, for example using string convolution. Based on the comparison, locations in the source code base that are likely to exhibit the style issue are identified. Various steps in the processing may be implemented using machine learning models, clustering or other automated data science techniques.