This work describes OU Social, an application that collects and analyses data from public Facebook groups set up by students to discuss particular Open University courses. This application exploits semantic technologies to monitor the behaviour of users over time as well as the topics that emerge from Facebook group discussions. The paper describes the architecture of OU Social and provides a brief overview of the analysis results obtained from 44 di erent Facebook groups examined over a 6 year period (2007-2013) The Open University (OU) is the largest university in the United Kingdom and the leading distance teaching institution in the world. The OU courses are directly available throughout Europe and, by means of partnership agreements with other institutions, in many other parts of the world. Given its on-line learning pro le, one of the key goals of the OU is the constant research and development of online learning and teaching solutions based on the analysis of usage data and on the feedback provided by the users about their experiences. To acquire this feedback the OU has created several websites and applications were students can discuss the di erent courses and share their learning experiences. This information, as well as OU's website usage data, is currently being collected and processed as part of the OU's usage data analytics process. However, there are other rich and rapidly growing sources of user feedback that are external to the OU, which could also be collected and investigated. With the emergence of social media, online learning is no longer restricted to particular in-house sites, but there is a clear tendency for students to share and discuss their learning material, methodologies and experiences on popular social networking sites, such as Facebook and Twitter. 1 This paper presents OU Social, a prototypical tool for collecting and analysing content from a large set of relevant Facebook public groups. These groups have been speci cally set up by Facebook users to bring together other students who 1 http://www.eric.ed.gov/PDFS/ED535130.pdf http://www.topuniversities.com/sites/qs.topuni/ les/Students-Online-UseageGlobal-Trends-Report-2013-nc.pdf
enrolled in particular OU courses or modules. The designed tool integrates two semantic analysis modules: (i) the behaviour analysis module, which categorises users into di erent behavioural roles (leaders, followers, etc.) by using a semanticrule based methodology and, (ii) the topic analysis module, which extracts and monitors the concepts that emerge from Facebook groups discussions (using a semantic annotation system). Data extracted from these modules is enriched via the OU's liked data portal (data.open.ac.uk) to provide a better overview of the courses under analysis.
The rest of the paper is structured as follows: Section 2 presents the architecture of the system and provides a brief overview of the semantic analysis modules. Section 3 presents some preliminary results obtained by this tool after analysing 44 di erent Facebook groups over a 6 year period (2007-2013). Section 4 concludes the paper and outlines future work. 2
This section describes the architecture of OU Social and provides an overview of its components. A video of this demo is available under http://people.kmi.open.ac.uk/miriam/OUSocial/OUSocialVideo.mov. The demo is not publicly accessible to avoid disclosing private student information.
Data Collection: Departing from the list of OU course codes, a crawler has been developed that extracts the name of all those Facebook public groups and pages containing a course code (e.g., B120) and the words "Open University" or "OU". For each of these groups information about users and posts is extracted (by using the Facebook Graph API2) and stored in an internal database.
Behaviour Analysis: The Behaviour Analysis component shows the type of people discussing a particular topic or concept of interest with regard to their online behaviour. It allows the OU's course managers to focus on a smaller, more manageable, set of students (read their contributions, monitor their opinion, etc.). The analysis module not only identi es those students that are mostly active (e.g., leaders, contributors) but also those who are generally inactive and may need additional learning support (e.g., lurkers, followers). Speci cally, this 2 https://developers.facebook.com/docs/reference/api/ analysis distinguish among eight types of user roles: Lurker, Follower, Daily User Contributor, Broadcaster, Leader, Celebrity and Super User. This analysis module makes use of the OUBO (Open University Behaviour Analysis Ontology) and the SIOC (Semantically Interlinked Online Communities ontology) ontologies to model the behaviour of users in the di erent Facebook groups. To infer the di erent roles that a user adopts over time the module applies semantic rules encoded using SPIN (e.g, if popularity=high and contribution=high then role=leader). For more details of this model, the ontologies and the role extraction process, the reader is referred to the following publication [1].
Topic Analysis: Questions, answers, discussions, learning material, etc. are distributed and shared via social networking sites. Detecting what are the topics that emerge from these discussions can help to identify emerging issues with respect to certain elements of interest to the OU. To obtain the topics for each post the analysis component makes use of TextRazor 3, a natural language processing tool based on knowledge bases such as Wikipedia, DBPedia and Freebase. TextRazor identi es key entities and topics in a piece of text, returning a mapping between each posts and a list of URIs.
Data Enrichment To complement the results of the analyses with concrete information about the courses discussed in the Facebook groups, course information is extracted via SPARQL queries from the OU's linked data site (data.open.ac.uk). E.g., the OU maintains a taxonomy of course categories (science, chemistry, ...) that complements the results obtained by the topic analysis. 3
An initial analysis have been conducted for 44 di erent Facebook groups over a
period of six years, from 2007 till 2013, including a total of 136,704 posts and
19,094 users. The demo aims to show conference attendees how di erent analyses
can be performed using this tool and how the use of semantic technologies can
help course managers and university sta to productively exploit social media
to obtain relevant feedback. Figures 1a and 1b display two examples of the OU
Social analyses. Figure 1a displays a tag cloud visualising the relevant topics
across all Facebook groups. Among these top topics we can nd People, Works,
Network Protocols, Behavioural Sciences and Educational Technology. This
visualisation can also be obtained for each group individually. Additionally, the
tool also allows the visualisation of the evolution of topics over time. Wikipedia
links are provided for each of the displayed topics thanks to the information
provided by TextRazor. Regarding the behaviour analysis, Figure 1b displays
an example of the role composition for the community built around the M263
Facebook group. We can see the evolution of the di erent roles over time. At the
beginning, during the creation of the group, it was mostly composed by inactive
users (lurkers and followers) but between the end of 201
This paper presents OU Social, a semantic social media analysis platform developed to collect and analyse students' feedback about OU courses expressed in Facebook public groups. The prototype is based on two semantic analysis modules that identify the behaviour of users and the emergent topics over time. Information about the courses is also integrated into the system by exploiting the OU's linked data portal. As seen by the video, extensions can be added to this prototype to facilitate the daily work of course managers. We are currently looking at tools like Google Trends (http://www.google.co.uk/trends/), Meltwater Buzz (buzz.meltwater.com), etc. to select visualisations that can better display the result of the analyses. Despite the existence of many other social media analysis tools in the market, OU Social is speci cally designed to ful l the needs of OU course managers. Apart from the interface extensions we are currently working on the integration of students historical data. This will enable correlating their behaviour and topic interests with their performance in di erent courses, providing the bases for more sophisticated analyses.