Augmenting Transactive Memory Systems in Virtual Teams by means of Natural Language Processing and Machine Learning

A Transactive Memory System (TMS) is a mechanism that captures the ability of teams
to encode, store and retrieve knowledge collectively. TMS, thus, helps in locating “who
knows what”. Such knowledge enables team members in an organisation to solve
problems requiring knowledge beyond their own expertise, and has thus been suggested
as one of the microfoundations of dynamic capabilities – the ability of organisations to
renew themselves. TMS has been shown to be valuable for efforts to integrate and
renovate knowledge assets of the organisation. However, prior research on TMS has
focused mainly on face-to-face teams with only few studies considering the more
difficult case of distributed work arrangements. In this research, I will expand on this
proposition and present a computational framework that supports TMS in virtual teams.
The objective of the research was to broadly examine the ways in which machine
learning algorithms and natural language processing techniques could be employed to
provide support to TMS in virtual teams. Specifically, this research builds and evaluates
a computational framework that pushes the boundaries of knowledge on distributed
work arrangements through the lens of TMS.
The research methodology followed the design science research. The validation of the
computational framework has been done using data mined from archived mailing lists
of a real Free Open Source Software development virtual team. In order to identify who
knows what in the studied virtual team, I used mined data from experts’ conversations
and survey data. Based on these foundations, I built a computational framework that
involves two main components: The first component handles the mining of raw textual
data and the second handles the classification of this data into broad areas of expertise.
My findings highlight the impediments to TMS in virtual teams and prove the
usefulness of machine learning techniques and natural language processing in
identifying expertise. Also, these findings suggest that it is possible and beneficial to
support TMS through algorithmic means. From a theoretical point of view, this research
contributes to the TMS research with a novel framework for augmenting TMS in
distributed work arrangements. These findings are generalisable to a similar type of
virtual teams. Although, only a limited number of skills were considered, the developed
computational framework can be improved and extended to include a greater range of
skills and other types of communities.