GRASS mailing list community evolution

Watching how grass-dev develops (and grass-user is used)

DRAFT - work in progress - by A Giacomelli and M Neteler

Introduction

During the 10th GRASS GFOSS User meeting in Cagliari, Italy, a summary of the activities of the Italian GFOSS community was presented. Together with basic indicators on the activity of the Italian community, some simple yet intriguing statistics, derived from an analysis of the main discussion mailing lists were shown. (DARE DUE esempi SU QUESTO).

In the typical brainstorming atmosphere which permits events such as software user meetings, we considered the idea of replicating the same analysis on two other mailing lists with a much longer history, namely the grass developer (grass-dev) and the grass user (grass-user) mailing lists.

The outcome of the analysis provides a unique insight on the dynamics of the user and developer communities, over an extremely long time span, from 1991 through 2008.

How source data was collected

The story of the creation of a seventeen-year long archive of communications deserves some description, as it is representative of the effort spent in maintaining a historical record of the communications within developers (and users) through various phases of the GRASS project.

SHORT STORY ABOUT LONG CONVERSATION (uh, could be better title)

• US Army mailing lists launch 12/1991
• interfaced with deja news (http://en.wikipedia.org/wiki/Deja_News) in (check MN)
• Deja_News forum only (dovrei verificare ma ho gli mbox files delle liste, si fa preso con "mutt")
• 1995 (?) email spam nasce in Dejanews, carefully later polished manually from the list
• new mailing lists born in 1999 at University of Hannover as dejanews wasn't usable nor pratical (check MN)
• in 2001 lists migrated to Italy with MN and server migration
• missing emails recovered from dejanews and merged into original lists mbox files (which MN received from US Army, don't remember precisely)
• All email headers for many years had to be reconstructed since their format was broken.
• complete archive restored and online (check date MN)
• in 2007, lists migrated to OSGeo infrastructure

Analysis Methodology

The information extraction approach used leans on the KISS side: the core of the parsing is handled by a perl script, while the remaining post processing is carried out via standard queries and no-nonsense charting tools.

Time and space

The first core set of information extracted was the time zone reference of the messages, considering that time zone may be used to provide an approximate indication of longitude.

One of the drivers for our analysis was also to verify if/how the mailing lists provided an evidence of the shift of development activity from the initial US-based model to Europe, rather than providing a detailed spatial distribution of the developers or the users. This insured that simply considering the time zone reference would be an adequate proxy of location for the source of a given message.

For the grass-dev list, the results we obtained from a first pass with the scripts developed was able to parse correctly over 99% percent of the messages. It may be possible to obtain a greater completeness by refining the parsing algorithm to handle exceptions encountered in the process, but we considered the level of approximation obtained in the extraction of the time zone reference to be adequate for the quality objectives of our analysis.

For the grass-user mailing list, the number of messages with time zone not identified by the first pass of the parsing algorithm is higher (some 3%), but still considered satisfactory within the scope of the current analysis.

What do time and time zones tell

The charts (include numbers) show:

• absolute number of message postings by time zone and year (Figures 1 and 4, respectively for GRASS-dev and GRASS-user)
• the relative proportion of messages posted each year from a given time zone (Figures 2 and 5, respectively for GRASS-dev and GRASS-user)
• the cumulated proportion of messages deriving from different time zones, calculated assuming 100% to be the e-mail traffic generated from the beginning of the mailing list records through 2008 (Figures 3 and 6, respectively for GRASS-dev and GRASS-user)

• Figure 7: local time of posting on grass-dev: shows that most of the communication is done on business hours (and some in the evening)

(mettere qui vari spunti)

...and what about the contents ?

Another interesting analysis is represented by the text extraction of specific keywords from the message body. While it can be extremely intriguing to build dictionaries of words and expressions used within a mailing list, in the case of the GRASS lists, we decided to focus on GRASS commands. Matrices with the occurrence of GRASS commands by year were generated for both mailing lists. The clear limitation in this type of analysis is that the use of a term is not associated to context. Reference to a specific command may not indicate if this is associated to a coding problem, to issues in use, or to working examples. Another element which is neglected in the analysis is quotation: i.e. the occurrence of a term is counted as long as it appears in the body of a message.

The review of the entries reported by the parser (Figures xx and yy) (DOVE CI PORTA ?)

yay... La cosa deve un po' crescere. (AL LIMITE CI LIMITIAMO A SPIEGARE CHE SIAMO CONTENTI DI AVER FATTO UNA PRIMA ESTRAZIONE...)

Reference: worksheets 5 and 6 of ODS file, which will become Table 1 (for GRASS-dev terms) and Table 6 (for GRASS-user terms)

Poi

- RELEASES AND EMAIL HYPE (faccio io) - ANNI 90: depression and renewal - ...