Selecting keywords using co-visitation information

A non-transitory computer storage medium encoded with a computer program, the computer program comprising instructions that when executed by a data processing apparatus cause the data processing apparatus to perform operations comprising:

• identifying a particular online resource that includes non-text content;
• obtaining co-visitation data for the particular online resource, the co-visitation data specifying one or more co-requested online resources for the particular online resource, each of the one or more co-requested online resources being requested by a user device within a threshold period of a request for the particular online resource by the user device;
• identifying keywords for each of the one or more co-requested online resources, the identified keywords for each co-requested online resource in the one or more co-requested online resources including keywords that were selected for the co-requested online resource based on text content of the co-requested online resource;
• determining, for each of one or more requests of a co-requested online resource that was received from the user device within the threshold period subsequent to a request by the user device for the particular online resource, a delay between the request by the user device for the particular online resource and the request by the user device for the co-requested online resource;
• assigning a relative weight to each of the one or more requests for the co-requested online resource based, at least in part, on the delay;
• determining a keyword score for each of one or more of the identified keywords, the keyword score for each identified keyword being determined based, at least in part, on a number of the one or more co-requested online resources with which the identified keyword is associated and the relative weight that was assigned to each of the one or more requests for the co-requested online resource;
• and
• selecting, based at least in part, on the keyword scores for the identified keywords, one or more of the identified keywords as keywords for the particular online resource.

A system comprising:

• a data store storing co-visitation data for a plurality of online resources, the co-visitation data specifying, for each of the plurality of online resources, one or more co-requested online resources for the online resource, each of the one or more co-requested online resources for the online resource being requested by a user device within a threshold period of the request for the online resource by the user device;
• and
• a data processing apparatus configured to interact with the data store, the data processing apparatus being further configured to perform operations comprising:
• identifying a particular online resource that includes non-text content;
• obtaining, from the data store, co-visitation data for the particular online resource, the co-visitation data specifying one or more co-requested online resources for the particular online resource, each of the one or more co-requested online resources being requested by a user device within a threshold period of a request for the particular online resource by the user device;
• identifying keywords for each of the one or more co-requested online resources, the identified keywords for each co-requested online resource in the one or more co-requested online resources including keywords that were selected for the co-requested online resource based on text content of the co-requested online resource;
• determining, for each of one or more requests of a co-requested online resource that was received from the user device within the threshold period subsequent to a request by the user device for the particular online resource, a delay between the request by the user device for the particular online resource and the request by the user device for the co-requested online resource;
• assigning a relative weight to each of the one or more requests for the co-requested online resource based, at least in part, on the delay;
• determining a keyword score for each of one or more of the identified keywords, the keyword score for each identified keyword being determined based, at least in part, on a number of the one or more co-requested online resources with which the identified keyword is associated and the relative weight that was assigned to each of the one or more requests for the co-requested online resource;
• and
• selecting, based at least in part, on the keyword scores for the identified keywords, one or more of the identified keywords as keywords for the particular online resource.