Senthil Kumar (Senthil Kumar) Kumarasamy Mani  Senthil Kumar (Senthil Kumar) Kumarasamy Mani photo       

contact information

STSM & Manager, Cognitive Application Support, Member - IBM AoT
India Research Laboratory, Bangalore, India
  +91dash080dash67043897

links

Professional Associations

Professional Associations:  ACM

more information

More information:  Publications  |  My Google Scholar Profile  |  My Research Gate Profile


NeedFeed: Fighting check-in notification spam based on code relevance

NeedFeed: Fighting check-in notification spam based on code relevance

Authors: Rohan Padhye, Senthil Mani and Vibha Singhal Sina

IBM Research India, New Delhi

Abstract: Modern software development almost always involves multiple developers concurrently modifying a common set of source code artifacts. Several platforms such as GitHub (based on Git) and Rational Team Concert (based on Jazz) offer features for collaborative development backed by a central software repository. In order to keep every member of the development team in-sync with the latest changes, such tools usually provide an option to send out code check-in notifications in the form of emails or pop-up alerts to every developer in the team. However, such notifications quickly cause an information overload for developers who are subscribed to this feed. Past research indicates that most developers spend a significant amount of time sifting through the feed to find changes that are relevant to their work, while some ignore the notification stream completely and are left with an inconsistent view of the system. In this paper we explore different strategies for reducing notification spam by filtering out changes that are irrelevant to a developer. Our definition of relevance is based on the observation that developers often revisit source code artifacts which they are responsible for. Experiments are performed on a set of 40 open-source Java projects hosted on GitHub. We start with a naive touch-based strategy that subscribes developers to all changes involving a file that they have modified at least once, and then try to improve on this by training a history-based model using parameters such as code ownership, relative contribution and recency of last change to a file. We also evaluate our technique at a finer granularity of changes to individual methods in Java files and discuss the feasibility of tool support for relevance-based change notifications

Data Set: We analyzed 40 open-source projects hosted on GitHub. The following data set contains a list of summary statistics about these 40 projects, and CSVs containing code metrics such as ownership, contribution, recency and longevity for each change for each developer: Download Data-Set