AFEL Adaptive Recommender

The resource recommendation services aim to provide suggestions of learning resources to users in Didactalia. These recommendations are calculated based on the identified and extracted features indicative of learning activities such as past user interactions, resource popularity and resource content. Three use cases have been identified that are of interest for Didactalia:

  1. The recommendation of popular resources in the system (i.e., non-personalized).
  2. The recommendation of resources with whom like-minded users have interacted (i.e, personalized).
  3. The recommendation of (alternative) similar resources for a specific resource (i.e., more-like-this).

The first use case was realized using a MostPopular algorithm. This approach recommends items which are weighted and ranked by the number of interactions (see Figure 1). Therefore, the Most Popular approach is non personalized and thus, every user would receive the same recommendation. The approach can be refined on the use case where certain interactions are weighted differently. For example, if a person rated a resource compared to just looking at it.


Figure 1. Popular resources in Didactalia.

For tackling the second use case, a Collaborative Filtering (CF) algorithm was chosen. Collaborative methods analyze the interactions between users and items and recommend those items to a given user that similar users have interacted with in the past. In order to boost novelty, this is often extended to recommend items that are at the same time new to the user.

More specifically, in collaborative methods two users are treated as similar if they have liked the same items in the past. This in turn allows us to assume that these two users will also like the same (or similar) items in the future. Figure 2 shows how these recommendations can be accessed in Didactalia.


Figure 2: Access to personalized recommendations from Didactalia user’s menu.

To realize the third and final use case, a Content-based Filtering approach was used (Figure 3). Content-based methods try to recommend items with similar content features as given in the profile of a given user. In this respect, the user profile could consist of (i) demographic information, such as the age or the interests of the user, (ii) the content of the items the user has liked in the past, or (iii) a combination of both. In cases where the user profile and the item profile share the same features (e.g., shared categories or social tags), the recommender can directly try to match users and items based on the profile information. In contrast, if the user profile is given by items the user has liked in the past, the recommender tries to find similar items to those and combines the individual results into a combined list of recommendations.

Figure 3: Example of AFEL recommendations in Didactalia: similar resources.

For future work, we plan to enhance our recommendation services by also taking into account the complexity of the learning resources and matching this with the competence of the learners (i.e., adaptive recommendations).