Recommendations

The key idea behind TV recommendations is filtering interesting programmes and bringing them to the user’s attention, which is becoming increasingly relevant to all TV watchers as channels proliferate.

Why is this of interest to NoTube?

NoTube expects to produce personalised content recommendations, using data models of users and contexts.

People often have several clusters of personal data on the Web, such as their profiles on social networks and ratings on YouTube and IMDB. Analogously to the personal data, there are many isolated clusters of broadcast data on the Web, such as broadcast data on electronic programme guides (EPGs) and background knowledge on Wikipedia.

NoTube asserts that the conjunction of all these bits and pieces of data can provide in- valuable information about someone’s interests and the TV broadcasts that should be recommended to them. Re-using existing online user data potentially solves the notorious ‘cold start’ problem, i.e. that a recommendation system cannot provide useful recommendations until a substantial amount of statistical user activity data has been gathered.

The NoTube vision is to realise a recommendation system that uses all these bits and pieces of data in conjunction. Semantic Web standards are important building blocks, since they enable the global identification mechanism of URIs and the means to define relations between data anywhere on the Web.

What NoTube has done so far in this area

Personalised recommendations based on a weighted interests User Profile

For these types of recommendations the user must have created a machine-processible User Profile from their existing social activity data using the NoTube Beancounter services. Beancounter services generate a User Profile of weighted interests from enriched user activity data.

This Profile can be used as an input to a NoTube Recommender service, which uses SPARQL queries over RDF data to pick up interesting matches between the user’s interests profile and enriched programme data.

Programme-to-programme recommendations based on interesting links

We are interested in serendipitous browsing of large video collections, so we are looking at ways to help users navigate through such collections by following interesting connections between programmes.

For example, for our BBC Archive Browser demo, the connections were created using a BBC-specific subject classification system, traditionally used by professional cataloguers for subject indexing/classification as part of the cataloguing procedure in the BBC’s internal TV and radio programme catalogue. The results of our user evaluation of this demo suggested to us that re-using existing programme metadata to navigate large video collections (including archives, on-demand content and EPGs) could help with the ‘cold start’ problem, whereby a recommendation system cannot provide useful recommendations until a substantial amount of user activity data has been gathered.

Recommendations influenced by social factors

We are exploring TV recommendations based on social influencers such as what my friends are watching, or via people who watch the same programmes as me, or the programmes people are talking about most.

Providing explanations

Presenting the pathways through graphs allows people to see the connection that led to a recommendation being made. An explanation based on an interesting story can potentially make a recommendation more satisfying to the user.

Future plans in NoTube

We are planning to compute smarter recommendations coming from an extensive browsing of semantic paths between enriched user data and enriched media items provided by broadcasters, in order to look for unexpected connections between a user and TV programmes he/she may like. We aim at leveraging the full power of Linked Data Cloud to find more and more links among interesting things and user activities.

Find out more

Watch this video about NoTube’s hybrid approach to recommendations: