1. INTRODUCTION. This paper describes the rationale and design of LASSIE (Learning Analytics for Social Systems in Institutional Education) an open source analytics tool for Athabasca Landing, a social site implemented at Athabasca University, which supports both formal and informal learning as well as many other social and practical applications. The paper begins by exploring salient differences between the Landing and a learning management system. We go on to discuss LASSIE’s objectives, design decisions, architecture and functionality, followed by descriptions of its data extraction, graphing, trend discovery, and statistical features. We describe the value of the system and discuss its limitations and future plans for development. 1.1 Learning Spaces. Learning management systems (LMSs) such as Moodle (www.moodle.org), BlackBoard (www.blackboard.com), and Desire2Learn (www.design2learn.com) have, historically, tended to model and replicate traditional classroom and institutional processes and consequently tended to embed institutional processes and forms such as courses, formal assessments, timetables, classes and hierarchies of control. They do not lend themselves well to different, more learner-centric approaches and are the cause of increasing dissatisfaction among educators (e.g.[1-4]). In recent years, alternatives that go beyond the LMS have begun to enter the field, with potential for greater learner control and rich tools for media creation and sharing. (e.g. [5-16]). These can present new challenges for learning analytics (LA) as less formal structure in a learning space leads to less easily analysable forms of data. Social media are soft technologies from which emerge patterns and usages that are not part of the hard design of the system but that are overlaid on top of it. Without clearly demarcated courses, lessons, learning outcomes and so on, there is a need for a flexible analytics toolset that can be adapted constantly to cater for many different structures of data. Athabasca University (AU) is an entirely distance-based university. To support its distributed population, we have built Athabasca Landing, an Elgg-based beyond-the-LMS social system. The Landing connects AU staff and students who are distributed over a vast geographical region, providing a soft space with rich tools and a comprehensive infrastructure for sharing and connecting with others, including blogs, wikis, bookmark sharing, photo sharing, event scheduling, group formation, social tagging, file sharing, podcasting, video sharing, profile creation, social networking and much more. It is more of a social construction-kit than a purpose-driven space. It supports many social forms including explicit groups, social networks, and set-based categorizations. It is as much as possible owned, shaped and controlled by the people who use it. It is a persistent space not defined by course or program temporal boundaries. There is no prescribed way of structuring courses in Landing. Most Landing users are not using it to support or attend a specific course and have joined it to communicate and network with other users, and those who are taking courses may engage with the site in many other ways. Thus the activity of a user taking a course on Landing is neither course specific nor is his or her course activity restricted to the course. 1.2 Learning Analytics. LMSs are, typically, course-centric and follow a rigid structure, thereby focusing and delimiting the scope of analytics. Courses have defined durations and students are only active within those time frames and can only interact within the confines of the course structure. Social learning environments such as the Landing provide enormous flexibility to instructors and learners alike and the analytics of this data can potentially explain the influence of extra-curricular activities on learner success such as the influence of a learner’s social network and extra-course activities. However, the blurred edges and flexible nature of the system presents a problem for those attempting to analyse how it is used. There is no strict definition of a course, course structure, a diverse set of learning objects can be used in innovative ways, and the instructor may choose to use the social space partially or fully for his course. Analytics applied to course delivery on the Landing poses numerous challenges when compared with that using a traditional LMS but also provides unique opportunities not possible in the formally bounded space. Students’ activity can be analyzed inside and outside the context of the course so, for example, friends network of a student and its correlation to student success can be analyzed and, in principle, it should be possible to observe diffusion of knowledge and connections beyond a fixed set of formal learning transactions. 1.3 Related Work. Data from a database can be extracted in several different contexts and tools have been proposed for different sorts of data extraction. [16, 27]. DeLeS [27] offers useful analytics but the analytics is limited to the functions programmed in the system. AAT [25] is not only limited to pre-programmed analytics queries but also allows users to specify data, create queries, and design reports using a graphical user-interface. Neither DeLeS not AAT is designed to work with social learning environments. SNA software (Social network analysis) is frequently used to analyse social systems data and many studies have been conducting on social systems using SNA. SNA numerically or visually describe features of a network for the purpose of quantitative or qualitative analysis. SNAs predate social networking and most are designed to analyse a network rather than a social network. Cytoscape (www.cytoscape.org) is a powerful network graphing and analysis tool that has been used in anything from analysing social networks and web networks to biological networks. Sonivis (www.sonivis.org) is geared more toward wiki spaces. Recently, SNAs geared towards social networking such as SocNetV [23], SNA-network [24], and Statnet (www.statnet.org) have also appeared on the market but they only provide analysis from a network perspective. To date, we have not found a software tool designed to analyse formal learning within a social network, leading to the decision to develop LASSIE. 2. LASSIE. LASSIE, available as a standalone software tool and as an Elgg plugin (with limited features), enables the analysis of user behaviour within and outside the context of formal learning. LASSIE is capable of extracting data, graphing it, and performing statistical calculations. It could be used to view data, trends, and correlate data to facilitate interpretation. Data are displayed in sortable tables and can also be downloaded in CSV format for further analysis using tools such as spreadsheet software, SNAs or R-language statistical packages. It can be used to observe activity of an individual user, a group of people, or the overall activity in different time slices. Furthermore, it supports time slicing, allowing users to view data from specific periods of interest. LASSIE also provides a REST-based web service support for future integration with SNAs and other software tools. Initially, LASSIE was designed as an Elgg plugin.. To cater for more data sources, LASSIE’s development was forked into a standalone application while maintaining the Elgg plugin as a cut-down version for simpler real-time and ad hoc queries. Figure 1 shows a screenshot of the standalone application. Figure 1. Screenshot of LASSIE. LASSIE provides a wizard-based user-friendly GUI with plenty of documentation. To cater for difficulties in defining course boundaries and user activity related to courses, LASSIE allows users to define these boundaries themselves. For example, a user may define course activity to be all activity in a group and/or all activity by a group of persons within a time period, and/or to filter according to specified tags, or relationships with others such as the teacher. LASSIE provides powerful and customizable reporting to empower users to make informed and intelligent interpretations of the data. Definitions of course, user activity, and learning objects can be specified for a single analysis or globally re-used. 3. RESULTS AND DISCUSSIONS. Analytics of social systems is data rich, noisy, influenced by many external factors and is therefore often difficult to interpret. For example, loss of activity or significant reduction in activity of a user for an extended period could be due to changes in personal life, tragedies, changing priorities, or due to a technical hiccup in the system. LASSIE provides analytics in the form of raw, numerical, and visual data for users, entities, entity groups, analytics of the entire site and correlation between any of these. Interpretation of data is left to humans. The system allows detailed and summarized tracking of user activity on Landing and the usage of different content types over a period of time. For example, Figure 2 shows a timeline of the collective activity of a group of users who took a course. The two high peaks correspond to first and last week of the course. As expected, there is high level of activity at the start and end of the course with some variation in the middle. The activity during the course is higher than the activity before or after the course. Activity on Landing continues to be higher after the course than before the course but only with selected tools such as blogs. Not surprisingly, spikes in activity in this course correspond to assessments and teacher-specified activities. LASSIE continues to provide very useful and insightful statistics about the Landing, its content, and how it is used. Discussing all the statistics and their interpretations is not possible in this paper, but we provide a few examples to illustrate the kind of analyses that are enabled by the system. Analyses show that a small subset of students who were not active before a course become active users during the course and remain active users past the course end date. Students who do not participate in extracurricular activities outside their courses tend to become inactive upon termination of the course. Course designs that favour social engagement will usually favour those who like to engage socially and the time on task that results from this may be the cause of greater motivation or the result of it. Figure 2. Collective blog, file and page posting activity before after and during the course. Two spikes corresponds to first and last assignments of the course. Figure 3 shows correlation between membership increase and addition of new pages to Landing. It also shows a constant decrease in the average content posted per user. Around 18 – 21 percent of the users are active users. It is interesting to observe, however, that different tools encourage quite different patterns of engagement: wiki pages, for instance, are intended to be used for collaborative content development while blogs are used for discussions on different topics. An active user is difficult to define. Many users become highly active or completely inactive for different periods of time. Many users are active readers and viewers but hardly post anything on the site. For the purpose of this discussion, an active user is someone who posts at least one content entry per month. Significant numbers of users continue to have a one-way relationship with Landing. They read and view regularly but do not participate with comments, votes, postings, or messages. Figure 3. Most of the pages are posted by a smaller group of users. As the number of members increases, this becomes more apparent. Decimal average was multiplied by 1000 to show the line in the graph. Users have tool and content preferences. Most users tend to use one or two tools (e.g. bookmarks, pages) most often. They tend to participate more in courses where their preferred tools part of the course structure. Group activity spikes are generally reflected in activity of users and spikes in activities of users are generally reflected in one or more groups. 78% of Landing content was created by 12% of the users. 91% of them are either directly linked or one link apart. A user following someone or in a shared group is a direct link. If a user’s direct link is linked to another user that he is not linked to, then the user is one link distance apart. Generally, users who follow many other users tend to be more active than people who follow fewer users. Many different networks can be constructed from social systems. Network of people following, network by groups, and content-posting network are some examples. Since most networks are based on some kind of activity, active users tend to be part of larger networks. Many networks such as networks of followers and followed users are scale-free networks. Degree distribution of scale-free networks follows the Power Law. If a node is removed from the network at random, there would be no significant change in connectivity, however, if the a hub node is removed, the loss of connectivity would be significant. Since the Landing project started, only one relatively inactive member has closed the account and three were banned, making it difficult to observe Power Law properties. However, LASSIE allows us to step back in time, remove a node and observe the effects. The findings confirmed the assumption that the networks are scale-free and that their degree distribution is governed by Power Law. Median and least active users in a course tend to become inactive after course completion. Active users tend to remain active in Landing after course completion date. Users participating in extra-curricular activities on the Landing during a course tend to continue participating in those activities past the course completion date. LASSIE provides valuable information about users over a timeline. Currently, the meaning behind the details in a timeline could be anyone’s guess, however, as the correlation capabilities of LASSIE improves and predictive modeling capabilities are added, it would be possible to identify in course and extracurricular behavior which leads to success in a course. Furthermore, this information would be highly valuable in counseling learners and designing courses that would facilitate learner success. The ultimate goal of a learning analytics software system is to be capable of clearly identifying activities and behaviour, which would lead to greater levels of success in a course. Such a software tool has yet to be invented for traditional or digital classrooms. Currently LASSIE provides invaluable information and tools to help us interpret the data. Many interesting correlations have been uncovered, however, caution has been taken not to jump to conclusions without further investigation, which requires adding additional capabilities to the software tool. Offering a course on a social system is not sufficient to improve learner success. The course structure needs to take the course material and user preferences into account. This cannot be accomplished without analytics software that can show how users are behaving in a social system and how they are responding to course structure and their online social environment. LASSIE provides this information but it is up to us to interpret the analytics data accurately. 4. CONCLUSION. In this paper, the LASSIE analytics software tool has been introduced and some examples provided of its use. It is a powerful, user-friendly and flexible tool designed to analyse learner and user activity in a social networking system. LASSIE is sufficiently flexible to allow course designers and instructors to explore beyond the course group area, allowing them to evaluate the performance of courses, course structures, as well as learner responses to new learning objects. It allows them to better understand their students and their activities. This information serves to help learning designers and instructors adapt, extend, and revise the course material and activities to achieve the pedagogical goals. As it develops, the tool becomes softer, more flexible, more capable of connecting disparate data. The softness and flexibility of the tool, like the Landing itself, allows many creative and unforeseeable uses. This means that greater care must be taken not to jump to conclusions without cross-checking correlations and their possible interpretations. Our next major challenge is to extend LASSIE to be able to work more effectively with the inverse set of those activities that are not course related yet which lead to learning. Analysis of a soft system, in which many of the technical and organisational processes are not embodied in software but in external systems and the minds of the users of that software requires a slightly different approach from that used for systems where goals, needs, methods and processes are clearer. Softer technologies need softer analytics LASSIE, like the eponymous sheepdog famed in Hollywood movies, has begun to herd and make sense of the information we need.