Personalized search result rankings

A method, comprising:

• receiving segment information containing information about entities related to a segment;
• receiving an activity history of a user;
• generating tags based at least on the activity history and the segment information, the tags corresponding to the entities in the segment information;
• calculating confidence values associated with the tags based at least on the activity history and the segment information using a first machine-learning model, the confidence values measuring the user's preferences for the associated tags that correspond to the entities in the segment information;
• storing the tags and the associated confidence values in user preference information;
• receiving a search query from the user;
• obtaining search results based at least on the search query, the search results having a non-personalized ranking that is based at least on global popularities of the search results;
• determining that the search query is directed to the segment based at least on the search query or the search results;
• calculating confidence scores based at least on frequencies of terms relating to the corresponding entities, the confidence scores reflecting degrees to which the search results relate to the corresponding entities;
• inferring a user preference for a particular search result based on a particular confidence score that indicates how much the particular search result relates to a particular entity and based on a particular confidence value that indicates how much the user prefers the particular entity;
• modifying the non-personalized ranking of the search results by adjusting a ranking of the particular search result based at least on the user preference to generate a personalized ranking of the search results reflecting the adjusted ranking of the particular search result, a degree to which the ranking of the particular search result is adjusted being determined based on the particular confidence score and the particular confidence value;
• and
• sending the personalized ranking of the search results to the user.

A computer-readable storage medium comprising instructions which, when executed by one or more hardware processors, cause the one or more hardware processors to perform a process comprising:

• storing segment information associated with a segment, the segment information including information about entities relating to the segment, the entities having entity identifiers;
• receiving an activity history of a user;
• generating user preference information associated with the user based at least on the activity history and the segment information, the user preference information including tags having entity identifiers that match the entity identifiers in the segment information, the tags being associated with confidence values that measure degrees to which the user prefers the corresponding entities in the segment information, the confidence values being calculated using a machine-learning model based at least on the activity history and the segment information;
• receiving a search query from the user;
• obtaining search results responsive to the search query, the search results having a non-personalized ranking that is based at least on global popularities of the search results;
• classifying the search query as being directed to the segment based at least on the search query or the search results;
• analyzing a web page associated with a particular search result to identify a particular entity in the segment information that relates to the particular search result;
• calculating a particular confidence score based at least on frequencies of terms relating to the particular entity, the particular confidence score reflecting a degree to which the particular search result relates to the particular entity;
• finding a particular tag and a particular confidence value in the user preference information that corresponds to the particular entity based at least on a particular entity identifier associated with the particular entity;
• modifying the non-personalized ranking of the search results by adjusting a ranking of the particular search result to generate a personalized ranking of the search results reflecting the adjusted ranking of the particular search result, a degree to which the ranking of the particular search result is adjusted being determined based on the particular confidence score and the particular confidence value;
• and
• sending the personalized ranking of the search results to the user.