Ranking and presenting search engine results based on category-specific ranking models
AssigneeHome Depot Product Authority, LLC
InventorsRongkai Zhao, Rajdeep Mondal, Ravi Sambhu, Naveen Krishna
Life of this patent
Estimating the term…
Abstract
Methods of operating a search engine may include calculating multi-modal document vector models for each of a plurality of electronic documents, training category-specific, search query-specific ranking models with respective machine learning algorithms based on those document vector models, and applying each of those models to further instances of the same search query to rank the documents responsive to that search query.
Claims20
A method of operating a search engine for a plurality of electronically-readable documents, each document associated with a respective category selected from a plurality of categories, the method comprising:
- receiving a search query from a user;
- executing a search on the plurality of documents based on the search query to generate a set of responsive documents, the set of responsive documents comprising a first subset of one or more documents associated with a first category and a second subset of one or more documents associated with a second category;
- ranking the responsive documents within the set, wherein ranking the responsive documents comprises:
- applying a first ranking model to the set of responsive documents to create a first ordered sub-list, the first ranking model associated with the first category;
- and
- applying a second ranking model to the set of responsive documents to create a second ordered sub-list, the second ranking model associated with the second category;
- creating an ordered list of documents according to the first ordered sub-list and the second ordered sub-list, wherein an initial subpart of the ordered list comprising at least a highest-ranked document from the first ordered sub-list and at least a highest-ranked document from the second ordered sub-list;
- and
- returning the ordered list to the user responsive to the search query.
A method of operating a search engine for a plurality of electronically-readable documents, the method comprising:
- obtaining a set of user search queries to a search engine, wherein each user search query in the set of user search queries is the same as or similar to each other user search query in the set of user search queries;
- obtaining a respective list of documents returned by the search engine responsive to each user search query in the set of user search queries;
- obtaining a set of user selections of one or more of the documents in each respective list so as to associate respective user selections of documents with respective user search queries;
- determining that a first subset of the documents within the set of documents are associated with a first category;
- determining that a second subset of the documents within the set of documents are associated with a second category;
- training a first document ranking model for the first category based on the respective user selections of documents associated with respective user searches;
- training a second document ranking model for the second category based on the respective user selections of documents associated with respective user searches;
- and
- storing the first and second document ranking models for use in ranking results of further user searches with the search engine that are similar to the set of user searches.
A method of operating an electronic document search engine, the method comprising:
- calculating a multi-modal vector model of each of a plurality of electronic documents;
- training two or more ranking models for each of a plurality of training search queries based on the multi-modal vector models and based on user behavior responsive to a plurality of search results, each ranking model specific to a category;
- receiving a further search query from a user;
- determining that the further search query is the same as or similar to a particular one of the training search queries;
- receiving, from a search engine, a list of some of the documents that are responsive to the further search query;
- applying the two or more ranking models associated with the particular one of the training search queries to the list to create a ranked list of the responsive documents;
- and
- returning the ranked list of the responsive documents to the user.
Dependencies are inferred from the claim text.