Indicator system is defined as a system that informs a user on a status, on past activities or on events that have occurred in a context; and helps the user to orientate, orgaise or navigate in that context without recommending specific actions.
So, it is not:
- a feedback system (analyse user interactions to inform learners on thier performance on a task and to guide the learners though it) or
- a recommender system (analyses interactions in order to recommend suitable follow-up activities),
- instead it provides information about past actions or the current state of the learning process.
- Moreover, smart indicator systems adapt their approach of information aggregation and indication according to a learner's situation and context.
The paper draws heavily on the notion of social navigation, interaction history and footprints, and offers a good review of this literature (ToRead).
The paper offers an architecture of smart indicators, where different layers are defined to support user modeling (first two) and helping the system to adapt to better decision making process (last two). Four layers:
- sensor layer
- semantic layer
- control layer, where a strategy defines the conditions according to learner's context
- indicator layer, presents aggregated information to the learner.
This approach of smart indicators adapts the strategies on the control layer (as opposed to semantic layer) to meet the changing needs of a learner.
SENSOR AND SEMANTIC LAYER
The paper further presents the information aggregates of sensor and semantic layers. The idea is to classify and organise the user's engagement (interaction foot prints) with the system, e.g. contributions, tagging activities. In the sensor layer, there is a division between "learner interaction" and "contextual sensors", e.g. location tracker, tagging activities (in my case this is considered direct) and contributions of peer-learners.
I am doing the same with my research data, and I call it the "user engagement" following the Yahoo!'s idea on STAR-metadata (kind of attentional and explicit metadata about users actions).
I tried to apply the classes of Chlan's prototype to my research data (learning repositories) that I collect using our CAM framework. Our focus being somewhat different, it did not really work out that well. The attempt below, though:
Direct: accessing resources through browsing, tag cloud, search result list, other user's favourites (implicit interest)
- user views metadata
- user views tags
- user views resource ("entry selection sensor")
- (timestamp on everything)
Direct: higher level interaction with a resource (explicit interest)
- user adds a resource to favourites and tags it ("entry contribution sensor", "tag selection sensor", "tagging sensor" or"tag tracing sensor", hard to say in my case)
- user rates the resource ("entry contribution sensor")
- user comments on the resource ("entry contribution sensor")
- shares resource with network ("entry contribution sensor")
- (timestamp on everything)
Contextual sensors could be (here I'm blending them with user information):
- context of a project within which the user access resources
- the information about the country and school from where the user is from
The semantic layer users the information from Sensor layer and transforms it into meaningful information by using an "activity aggregator". This calculates the activity for a given period of time for an individual learner or the whole community according to different ratings that each activity has (beginners have different way of counting activity from power-users).
In this prototype the control layer defines how the indicators adapt to the learner behaviour. There are two elemental strategies:
- motivate learners to participate to the community activity
- raise awareness on the personal interest profile and stimulate reflection on the learning process
This layer embeds the indicators into the user interface of the community system. The prototype is being tested by a group of PhD students now.
Glahn, Christian, Specht, Marcus, Koper, Rob (2007) Smart Indicators on Learning Interactions