This paper describes the Learning Resources A ective Recommender, a widget that recommends resources to the learner based on her current a ective state and her learning objectives. The widget is meant to be used in a Personalized Learning Environment in combination with widgets to search for resources. The architecture that supports the widget follows a client-server pattern, with the widget as the client and a recommendation service on the server side. The paper includes the description of both client and server and a discussion about the possibilities of this approach.
There is empirical evidence in the psychological literature that a ective states
affect cognitive processes like memorizing and decision making. This has provoked
researches in the eld of Technology-Enhanced Learning (TEL) to investigate
the e ects of including a ective states into the design of technological tools for
learning. Among the tools that take into account a ective states we can nd
intelligent tutoring systems (ITS) [
However, there is a type of tools in the TEL eld of TEL that still hasn't
shown evidence of including a ective states into its design process: recommender
systems. As it has been reported by Manouselis et al. [
Our proposal to explore the inclusion of a ective states in recommender systems consists of a widget to be used in a Personalized Learning Environment (PLE). The a ective states are explicitly indicated by the learner, thus the detection of a ective states is out of the objectives of the widget. The widget architecture follows a client-server approach, where the widget plays the role of the client and a web-based recommender service plays the role of the server. Besides the a ective state, the widget also considers the learning objectives of the learner in the recommendation process.
The rest of the article is structured as follows. Section 2 describes the widget, its architecture and both of its components. Section 3 discusses the evaluation of the generated recommendations and future work. 2
The architecture supporting the Learning Resources A ective Recommender (LRAR) follows a client-server approach, where the widget and a recommender web service the role play the role of the client and server respectively. 2.1
The recommendation process follows the method known as collective intelligence or collaborative ltering, and we have used the machine learning engine Apache Mahout for its implementation. Collaborative ltering can be item-based or userbased. When based on item similarity, a user is pointed to items that are related to the items already accessed by the user; a common example of this approach is the Amazon recommendation system. On the other hand, user-based collaborative ltering nds neighbors, users with similar patterns of resources accesses, and then points the user to resources relevant to her neighborhood. Henceforth, given our educational context we will refer to the users as learners and to the items as resources.
Our recommendation service is follows the user-based approach. The key element in the process is the similarity function that identi es the neighbors of a given learner. In our implementation, the similarity of two learners is proportional to the amount of resources accessed by the learners when indicating the same a ective state. The proportion of common learning objectives also a ects proportionally the similarity metric.
The recommendation service also handles secondary but needed functionalities, such as the creation of a learner pro le, the change of mood informed by the learner and the addition of a new learning resource. For this, it was needed to model four entities and their respective relationships: a ective states, learners, learning objectives and resources. The creation, edition and elimination of these entities is available through a RESTful web API, using the JSON format. 2.2
The widget has been developed on top of the ROLE Project [
Three di erent sections have been included in the widget: Resources, Pro le, and Settings. Resources is the main tab and has to functions; rst, it allows the learner to state her a ective state from a static list provided by the recommendation service. Second, it provides a list of learning resources ordered by relevance for the learner in her current state.
The Pro le tab presents a time-line of the a ective states reported by the learner. Its objective is to provide the learner with a visualization of her emotional changes during the learning activity being performed. The log of a ective states is also provided by the learning resource service.
Finally, the Settings tab allows the learner to set her learning objectives. These might be changed during the learning activity, which also triggers a change of the learning resources that are recommended. Fig. 1 presents a screen capture of the widget in action, with emphasis on the resources recommended to a frustrated learner.
The Learning Resources A ective Recommender is our proposal to analyze the e ect of including a ective information into the logic of a TEL recommender system. This approach allows learners to obtain recommendations based not only on their learning interests but also on their changes of a ective states during a learning activity.
Further work for LRAR is the evaluation of the recommendations generated, specially when compared with a system unaware of a ective information. Our plan to perform this evaluation is through the comparison of metrics used in information retrieval, like precision and recall.
Another line of future work consists of the use of sensors as a complement to the a ective state self-reported by the learner. We have special interest in sensors for galvanic skin response and for the recognition of face gestures. The inclusion of these sensors might provide a ective states with higher accuracy and thus help to improve the recommendation process.
Work partially funded by the EEE project, \Plan Nacional de I+D+I TIN201128308-C03-01", the \eMadrid: Investigacion y desarrollo de tecnolog as para el e-learning en la Comunidad de Madrid" project (S2009/TIC-1650), and \Consejo Social - Universidad Carlos III de Madrid".