Notes on Context-Aware Recommender Systems

I am reading a really interesting article on Context-Aware Recommender Systems for Learning. Sharing some of my notes here for future reference.

Context-Aware Recommender Systems for Learning Verbert, Manouselis, Ochoa, Wolpers, Drachsler, Bosnic and Duval 2012

shutterstock_130947476The incorporation of contextual information about the user in the recommendation process has attracted major interest. Such contextualization is researched as a paradigm for building intelligent systems that can better predict and anticipate the needs of users and act more efficiently in response to their behavior.

Recommendation algorithms, such as content-based filtering, collaborative filtering, knowledge-based filtering, and their hybridizations are widely discussed in the literature.

TEL (Technology Enhanced Learning) Recommender Systems

Most systems suggest learning resources or people who can help with a learning activity. Course recommenders typically provide advice to learners on courses to enroll in. Several social navigation systems rely on recommendation techniques to suggest resource sequences.

Intelligent Tutor systems use information about the learner to suggest personalized hints while the learner is solving a problem.

Context in TEL

Dey et al define context as “any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and application themselves.

Schilit divided context into 3 categories:

  • Computing context – such as network connectivity, communication costs, communication bandwidth, and nearby resources such as printers, displays, etc
  • User context – such as the profile of the user, location, people nearby, and social situation
  • Physical context – such as lighting, noise level, traffic conditions

Chen and Kotz added a fourth category – time.

Schmidt et al added a task category and define the following dimensions: the user, the social environment of the user, tasks, location, infrastructure, physical conditions, and time.

Zimmerman et al list individuality, activity, location, time, and relations as fundamental context categories. Individuality is divided into 4 elements: natural entity, human entity, artificial entity, and group entity.

Berri et al distinguish between technical and learner contexts.

Technical contexts – deals with the technical aspects of mobile devices and their operational environment including the capacity and bandwidth of the wireless network, input and output.

Learner Contexts – aims and objectives of the learner, prerequisites, background, current level of understanding and subject domain.

Environment constitutes computing, time and physical context

Derntl and Hummel define

  • world context that constitutes location and date and time,
  • physical context (persons, books, journals, learning equipment),
  • digital context (epapers, collaborations, elearning services)
  • device context (hardware, software, connectivity)
  • learner information context (personal information such as name, expertise, interests, and task specific information)

Li et al define 5 context dimensions:

  • who (user)
  • what (object)
  • how (activities)
  • where (location)
  • when (time)

Context-Aware Recommender Systems – traditionally, collaborative, content-based, knowledge-based, and hybrid recommender systems deal with 2 types of entities; users and items.


Recommendation via context-driven querying and search uses contextual information to query or search a certain repository of resources and presents the best matching resource.

Recommendations via contextual preference elicitation and estimation attempts to model and learn contextual user preferences. Built on knowledge of partial contextual user preferences and deal with data records of the form <user, item, context, rating)

  • Contextual prefiltering  contextual information is used to filter the data set before applying a traditional recommendation algorithm
  • Contextual postfiltering recommendations are generated on the entire data set.
  • Contextual modeling uses contextual information directly in the recommendation function as an explicit predictor of a rating for an item.

A Context Framework – a simple classification of context information that is relevant to context-aware applications.

Computing – characteristics can be classified in 3 areas:

  • Network – static and dynamic properties of the network
  • Hardware
  • Software

The acquisition and use of computing context is necessary to support intelligent interfaces that can select suitable learning resources for the device that is used.

Location contexts include:

  • Proximity of objects within a space
  • Communicative ability
  • Orientation

Time context includes date and time information.

Physical Conditions context describes the environmental conditions where the system or user is situated and commonly measures for heat, light, and sound.

Activity context reflects the tasks, objectives or actions of the user. Capturing user actions which are comprised of events within the application, and session and time related data.

Resource context captures the relevant characteristics of physical or virtual resources.