Digging for Diamonds: Identifying Valuable Web Automation Programs in Repositories

Web automation programs offer a means for users to enhance the usability of the web. These programs can be published on a wiki or other repository, thereby making them available for use by other users. However, in addition to programs of broad usefulness to the community at large, these repositories also contain many programs that are unreliable or highly specialized to the needs of very small sub- communities. These less valuable programs clutter the repository and make it difficult to find the valuable web automation programs. In this paper, we evaluate a machine learning model that can distinguish between high-value and low-value web automation programs. We find that the model performs well for a wide range of different languages, purposes and configurations, indicating that the model could serve as an effective basis for future repository enhancements.