Multiple User Interest Profiles

A method performed by one or more data processing apparatus, the method comprising:

• accessing a single user's long-term web browsing history including disparate content-based web pages that the user has visited over a predetermined period of time, the predetermined period of time including multiple different user sessions that occur over more than a week;
• selectively grouping the disparate content-based web pages into subsets of web pages, the subsets of web pages being related to various different topics;
• extracting keywords related to the various different topics from the subsets of web pages;
• creating, for the user, a plurality of user interest profiles, the plurality of user interest profiles being organized into different categories, each user interest profile of the plurality of user interest profiles corresponding to a particular topic and containing keywords extracted from a subset of web pages related to that particular topic;
• determining a context of a given web page to be presented by a device associated with the user including (i) determining a given topic of the given web page and (ii) determining a day of week when the request is received and the given web page is being displayed;
• identifying, in response to the request, fewer than all of the plurality of user interest profiles of the user that are related to some of the various different topics that are relevant to the determined given topic of the given web page and created for the determined day of the week when the request is received and the given web page is being displayed;
• and
• either (i) updating the identified fewer than all of the plurality of user interest profiles of the user based on keywords collected from the given web page presented by the device, or (ii) creating one or more new user interest profiles if no existing profiles were identified.

A system comprising:

• a processor configured to execute computer program instructions;
• and
• a computer storage medium encoded with computer program instructions that, when executed by the processor, cause the system to perform operations comprising:
• accessing a single user's long-term web browsing history including disparate content-based web pages that the user has visited over a predetermined period of time, the predetermined period of time including multiple different user sessions that occur over more than a week;
• selectively grouping the disparate content-based web pages into subsets of web pages, the subsets of web pages being related to various different topics;
• extracting keywords related to the various different topics from the subsets of web pages;
• creating, for the user, a plurality of user interest profiles, the plurality of user interest profiles being organized into different categories, each user interest profile of the plurality of user interest profiles corresponding to a particular topic and containing keywords extracted from a subset of web pages related to that particular topic;
• after creating the plurality of user interest profiles for the user:
• receiving a request for an item of content to be presented to the user in connection with a given web page that is being displayed by a device associated with the user;
• determining a context of the given web page including (i) determining a given topic of the given web page and (ii) determining a day of week when the request is received and the given web page is being displayed;
• identifying, in response to the request, fewer than all of the plurality of user interest profiles of the user that are related to some of the various different topics that are relevant to the determined given topic of the given web page and created for the determined day of the week when the request is received and the given web page is being displayed;
• choosing the content item to be presented to the user based at least in part on a set of keywords contained in the fewer than all of the plurality of user interest profiles of the user, wherein the set of keywords include fewer than all of the keywords extracted from all of the subsets of web pages;
• and
• causing the chosen content item to be delivered for presentation to the user.