Jack suffers from a particular case of information overloading about TV. He is fine with tons of movies, ads, documentaries, talk-shows hitting him every evening. He just can’t phisically stand the huge mass of news (here a ‘news’ stands for a single fact described and commented by the news-man) that at 8 p.m. fall over him without any sense and particular reason. He rather prefers not to watch any newscast. This situation is no more bearable, so he decides to try the Notube Personalized News app. He register himself on it: he’s Jack, english mothertongue with good proficency in italian (thanks to the italian mother), and he prefers to get news extracted from BBC and RAI newscasts. At the beginning those are the only information he puts into the app. A first news list appears, and he choses to watch 2 particular news. From these 2 news, more news are suggested by the app, similar under some aspects to the original ones, and so on (*R1). In this way he explores all different directions those initial 2 news can take him. He gets also full justifications about followed paths from the watched news to the recommended ones (*E1), and he’s quite satisfied about that. Then he begins to get saturated by this kind of news and let the app know him better. He allows the app to access his personal information about activities done in some social networks he uses most. I.e. Facebook and Twitter (*P). From pages liked in Facebook and hashtags used in twits, a profile in terms of weighted interests is computed by the app. Each interest is a subject Jack is interested in, like African History or Japan with a weight expressing the percentage over the whole bunch of interests, like 12% or 0.1% (*P). From now on, suggested news fall under these interests, in a mix guided by the weights (*R2). And what Jack gets, dynamically follows his changing interests, cause everything new Jack does in Facebook and Twitter, as well as the news chosen among those suggested, adds up to and slightly modifies the resulting profile. For a reasonably long time Jack has nothing to complain about the app behaviour. He doesn’t care about football? He doesn’t get anything about football. He’s compulsively attracted to left-wing party choices? He gets everything about that. After a good while, though, each news simply confirms his interests, making them stronger and stronger in his profile, while nothing positively surprises him. Sometimes an isolated news different from all the others comes out, but the corrensponding interests in the profile will never get a weight high enough to compete with the other ‘mainstream’ ones. Besides, he find the suggested news as a bit too cold and generic, having nothing to do with the objects of his activities in particular, simply adapting to his computed abstract interests. He knows that he’s demanding more and more of this app, but it’s easy to accustom oneself to better services. Then he finds a quite cryptic feature in the app, named ‘Tell me how to look for news for you’. So he discovers that he can tell the app exactly which paths to follow, starting from the objects of his activities, in order to find new interesting news, tailored for him (*R3). Aided by the app, he gives as input two paths: from a person involved in his activities to a person involved in some news, and from a location involved in his activities to a location involved in some news. Two simple paths to start with. After that, he begins to notice some news in the stream of the recommended news that have some strong connections with his activities. I.e.: an important football match is going to be played in his own little town (found by the second path, he guesses), or an actress he loves will be a participant of an evening show on BBC (by the first pattern he gave). Wow, Jack couldn’t ask for more. Now news generally follow his interests, besides some news that clear the barrier of his interests thanks to some specific reasons linking Jack to the news.
*R1: first recommender – from item to item without any user profiling
*E1: first enricher – adding paths between items
*P: Privacy issue: OAuth protocol
*R2: second recommender – from profile and item to item
*R3: thrid recommender: from activity object to item