How Google Now selects fresh content
"Interactions with search engines are often reminiscent of the movie “50 First Dates”, in that the search engine appears to have forgotten all past interactions." +Ramanathan Guha
, et al.
Show personalized content recommendations for the user’s long term interests
, using Fresh Aggregates:
"For each entity (or query) that occurs in a sufficiently large number of contexts, we get the set of users who have that entity (query) in any of their contexts. We then aggregate what those users have read recently. Thus for a given entity, we have aggregate statistics on what fresh content would be interesting for users with that entity." Context Aggregates
is similar, although details are quite different. The goal is to assist users with longer tasks
, e.g., trip planning or buying a camera, by leveraging the experience of users with similar tasks:
"For each URL, entity or query that occurs in a sufficiently large number of contexts, we compute aggregate statistics over the set of contexts in which that URL (or entity or query) appears. For example, given an URL, the aggregates tell us what other URLs, entities and queries users looked at later in that context. We also compute meta-data, such as the fraction of clicks in those contexts that are on news or videos."
Read the full article User Modeling for a Personal Assistant:http://www.rsrikant.com/papers/wsdm2015.pdf