Détection de communautés dynamiques dans les réseaux évolutifs de développeurs au sein de Chrome

, , , and

BibTeX reference

The time-stamped database provided by Google traces the code contributions to Chrome browser on the period from September 2008 to January 2014. It describes the participation of 1045 developers in 46 offices in 22 countries with some 324 860 changelists (CLs).

An individual who contributes to a CL can be done either as "submitter", "approver", "Comment" "Comment unsollicited", "Disapprover", "reviewer" or "reviewer silent". We thus have a social network evolving over nearly six years in which the participation of developers in a CL is a collaborative relationship.

The network can be illustrated by a graph, that is to say a set of points called vertices connected by edge segments. In this case, participation in the same CL results in the existence of edges between the vertices representing the parties involved.

The quantitative study of the database proceeds according to two approaches: descriptive statistical analysis with the objective of drawing the dynamic profile of individuals; and, secondly, an automatic classification which aims to involve developers in sub-groups, also called communities more strongly interconnected than with the rest of the network.

In graph theory, the classification is equivalent to community detection in a graph and is a difficult combinatorial problem, even more so if every relationship is subject to temporal information.

The detectable communities vary over time. In addition, for a given community structure, quantitative measures provide an understanding of the internal organization of the group and measure the effectiveness of the collaborative work.

In light of our analysis, it appears that there are several profiles in the activity for an individual, characterized in particular by experience; the association of developers can be interpreted in part by membership in an office; and the effectiveness of a community is based on the structure and composition.

, 19 pages

Research application


G2160.pdf (1 MB)