"1. INTRODUCTION. Social media is described as a web-based service that allows individuals to construct a profile within an organized framework, to generate a list of other users with whom they share a connection, and to navigate their own list of connections and view those made by others within the system [1]. Thus social networks use the Web to facilitate their communications as one element of what has been termed ‘Web 2.0’ - a term coined by O’Reilly [2] and which has come to be used to describe a wide range of Internet-based information and communication technology (ICT) applications that offer the potential for significantly increased interactivity with a high degree of ‘communication, cooperation, collaboration and connection’ [3] between users. More recently attention has turned to how these new applications can be applied to facilitate educational attainment. It is often recognised that a constructivist perspective is most appropriate for understanding learning through social media. Insights on student behaviour and academic performance however are relatively limited in the research literature. Our intention here is to analyse access patterns to a discussion forum within a VLE (virtual learning environment) to determine a taxonomy of online learners. Often, naively, it is assumed that mere participation in an online discussion session – a form of social media – will lead to better learning, and therefore, raised performance. However, it remains to be seen whether social media in itself can aid learning and attainment through participation. Moreover it is a matter of debate what type of access pattern is most conducive to learning. Are relatively short yet frequent sessions involving directed engagement with a discussion forum to obtain a specific objective more productive in learning terms than relatively long yet infrequent sessions intended to develop understanding through exploration and sustained participation? 2. ONLINE IDENTITY and COMMUNITY. 2.1 Learning Theory. Social Learning Theory [4] sees environmental (i.e. social) and psychological factors influencing behaviour with ‘retention’ (remembering what one observes), ‘reproduction’ (recreating that behaviour), and ‘motivation’ (having good reason to recreate that behaviour) as critical cognitive aspects. ‘Motivation’ in these terms resonates with the ideas of Whyte [5] who advocated getting students to accept personal responsibility for their own learning, and, as long as the individual has some inherent ability, the result is that better academic progress is made. However some form of guidance or ‘scaffolding’ is required for learners and this needs to be built into the educational infrastructure, or online learning environment. Social Development Theory [6] argues that social interaction precedes intellectual development, with consciousness and cognition being the end product of social behaviour. Thus the connections between people and the sociocultural context in which they interact are critical for such development [7]. Thus Vygotsky challenges the traditional ‘instructionist’ model of learning whereby teachers ‘transmit’ information to students, and offers in its place learning contexts in which students themselves take an active role in enabling their learning. In this analysis of social constructivism, Vygotsky’s Zone of Proximal Development (ZPD) appears to chime to some extent with the Community of Practice (CoP) concept [8] whereby novices firstly undertake learning from a position of ‘Legitimate Peripheral Participation’ (LPP). Lave and Wenger argue that learning is situated such that it is embedded within activity, context and culture. However intentionality is not necessarily assumed in this model of learning and can be an incidental or vicarious outcome of the interactions occurring within a ‘community of practice’. 2.2 Empirical Evidence. While there are a number of proponents arguing for the use of ‘social media’ as an educative tool, empirical data on their use is not plentiful but is contentious. A study by Heo et al. [9] examining online interaction amongst of a set of students engaged in project-based learning showed that levels of academic achievement do not always reflect the quantity of online interaction, while the quality of interaction is shown to have a critical relationship with academic outcomes. The fact that social media can act as a distraction to learners is attested to by the work of Junco [10] who finds Facebook usage and scholarship to be negatively correlated: the frequency of engagement with Facebook has a negative correlation with the amount of time spent by students preparing for class. In the same vein Junco and Cotten [11] find that students who spend more time chatting online than their peers report greater levels of academic impairment. Such evidence shows social media holding back rather than supporting student learning. This view chimes with Kubey et al. [12] who find that heavy recreational Internet use highly correlated with impaired academic performance. Additionally loneliness, staying up late, tiredness, and failing to attend class also have correlations with self-reports of Internet-caused academic impairment. Such observations prompted further research and hence we wished to investigate student access patterns to an online discussion board and apparent learning approaches. 3. METHODOLOGY. The investigation took place over two years (2009-2011) on an undergraduate course on Project Management. Students on the course were enrolled on different degree pathways (Marketing and Management) so the class composition included diverse academic interests as well as diverse demographics (e.g. gender, race, age, etc.). Each session ran for one term (from January to April), with class lectures scheduled on a weekly basis. The task given to the student cohort was to form teams of 4-6 people in order to undertake a project assignment which ran from January until mid-March. Students were informed that the group activity did not attract academic credit per se, but was essential for a subsequent activity which did attract credit. This follow-on activity required students on an individual basis to produce short critical accounts reflecting on their team working experience in the first activity. Thus, without sufficient involvement in the group activity, students would find it difficult to complete the reflective account (the second activity). Communications amongst the team members in the group activity were to be conducted through an online discussion application (Blackboard), with instructions given to students to refer to the discussion forum frequently in order to receive updates from the class instructors as well as to raise queries and to read responses. The first year of the study (2009/10) involved 160 active students on the discussion forum, while the second year (2010/11) involved 143 active students. Only those students who submitted the second assignment were considered to be active. Analysis of the data derived from the discussion forum involved a hierarchical cluster analysis to identify subgroups amongst the class, as well as tests of difference amongst groups (t-tests) and tests of association among factors (Pearson correlation). 4. RESULTS. An initial hierarchical cluster analysis (Ward’s linkage on Euclidean distance) was performed on the data set for each year separately. Year 2010/11. This procedure distinguished several groups on dimensions related to variables extracted from the analytics data: the average interval between online sessions (AIn); and the average duration of an online session (ADu). We see that there are five major groups of students formed when differentiating on the variables introduced above, with the primary differentiator being ADu at a height of h=7.77 (i.e. a considerable difference amongst the two top-level clusters). The top-level clusters consisted of n=87 students with average durations per session of less than 9 minutes, and n=56 students with average duration per session of greater than 9 minutes (the cut-off point of 9 minutes was chosen from the dendogram by visual inspection). The second level clusters were divided on the basis of average interval between sessions. For the group with relatively shorter ADu, a first division at h=6.94 was made, though a subsequent division at h=4.05 serves to better define three distinct clusters. The height of h=4.05 indicates a relatively high distinction amongst clusters. The sizes of the sub-clusters were n=54, n=26 and n=7, divided at average interval per session of 1-5 days, 5-12 days, and more than 12 days respectively. The alternative group with comparatively longer ADu was divided at h=4.01, with cluster sizes of n=41 for 1-5 days, and n=15 for more than 5 days. No further divisions were made due to small height values which could not provide reliable sub-clusters. We can define the five clusters on the basis of their average interval between sessions and the average duration of a session. Cluster 1, n=54: Short interval between sessions and short duration per session. Cluster 2, n=26: Medium interval between sessions and short duration per session. Cluster 3, n=7: Long interval between sessions and short duration per session. Cluster 4, n=41: Short interval between sessions and long duration per session. Cluster 5, n=15: Long interval between sessions and long duration per session. Note that no clear cluster emerged with medium interval between sessions and long duration per session. A test of difference between means (independent samples ttest) confirmed no significant difference between clusters formed on final assignment mark. A one-way ANOVA test was conducted on the three sub-clusters divided at h=4.05 (Levene’s statistic F=.515, p=.599 confirms homogeneity of variance). Results indicate there is a statistically significant difference between groups (F (2.77) = 5.878, p = .004). A Tukey post-hoc test reveals that the final mark attained in the follow-up assignment is statistically significantly higher for average interval per session of between 1 and 5 days (1