Learning to Describe Editing Activities in Collaborative Environments: A Case Study on GitHub and Wikipedia

Learning to Describe Editing Activities in Collaborative Environments: A Case Study on GitHub and Wikipedia Edison Marrese-Taylor1, Pablo Loyola2, Jorge A. Balazs1 and Yutaka Matsuo1 Graduate School of Engineering, The University of Tokyo, Japan1 {emarrese, jorge, matsuo}@weblab.t.u-tokyo.ac.jp IBM Research, Tokyo, Japan2 e57095@jp.ibm.com Abstract We propose to study the automatic generation of descriptions from content editing activities in collaborative environments. We define such task as identifying the changes associated to two consecutive versions of a document, and then producing a message in natural language that explains it, which should provide a com- pact description of the change while retain- ing its key informative elements. Our model is based on a sequence to sequence architec- ture that receives as input the representation of the change, and outputs a message. We pro- pose a framework to conceptualize the prob- lem and two instances for GitHub activity and Wikipedia contributions, two of the most im- portant collaborative systems on the Web. Our results indicate that the proposed approach is able to generate feasible descriptions, which are on average aligned with the semantic pur- pose of the editing activities. 1 Introduction One of the positive outcomes of the current perva- siveness of the Web has been the boost of collab- oration across several domains. Examples of this are platforms such as Wikipedia and GitHub, where self-organized and voluntary groups of individuals gather based on the common goal of crafting docu- ments and programs (Crowston et al., 2007). The outcome of these collaborative activities is usually the result of a series of incremental mod- ifications over time. For example, in the case of GitHub, incremental modifications are usually func- tional changes, which allow to incorporate new fea- tures or fix reported bugs. In the case of wiki-based platforms, contributors modify the content of a given article in order to reflect an update on the matter the article is dealing with. The transparency and openness of change management provides complete awareness of the state of the document being crafted, at any point in time (Dabbish et al., 2012). As collaboration is carried out in a decentralized fashion, the coordination between contributors plays a key role (Von Krogh et al., 2003). While there exist direct ways of communication, such as bug trackers and discussion forums, we are interested in studying the indirect ways in which contributors interact and coordinate. One of these elements are the short messages that the contributors provide at the time of submitting the change. This short message usually provides a description of the change and serves as a way of broadcasting it to the rest of the community, ideally clarifying the purpose and other technical aspects, and supporting the reviewing process (Guzman et al., 2014). Therefore, our goal is to use this set of change- message pairs to develop a model able to explain collaborative activities by automatically generating a short passage in natural language. In that sense, we visualize this task as being in-between summariza- tion and translation given the challenges it presents, …

Content truncated to 500 words