Entity- and string-based search using a dynamic knowledge graph
AssigneeMicrosoft Technology Licensing, LLC
InventorsHamed Firooz, Lin Guo
Life of this patent
Estimating the term…
Abstract
Techniques for performing a database search using a rewritten and annotated query are disclosed herein. In example embodiments, a profile lexicon is generated from a set of raw user profiles. A click-through lexicon is generated from a raw query log. A machine-learning model is trained for entity prediction using selected data. Query tagger data is generated using the profile lexicon, the click-through lexicon, and the machine-learning model. A raw query is received. The raw query is rewritten as an annotated query based on the generated query tagger data. A search of a database is performed using the annotated query. Results of the search are returned in response to the receiving of the raw query for presentation in a user interface.
Claims20
A system comprising:
- one or more computer processors;
- one or more computer memories including computer executable instruction that when executed by the one or more computer processors cause the one or more computer processors to perform operations comprising:
- generating a profile lexicon from a set of raw user profiles;
- generating a click-through lexicon from a raw query log;
- training a machine-learning model for entity prediction using selected data;
- generating query tagger data using the profile lexicon, the click-through lexicon, and the machine-learning model;
- receiving a raw query;
- rewriting the raw query as an annotated query, the rewriting based on the generated query tagger data;
- performing a search of a database using the annotated query;
- returning results of the search in response to the receiving of the raw query for presentation in a user interface.
A method comprising:
- generating a profile lexicon from a set of raw user profiles;
- generating a click-through lexicon from a raw query log;
- training a machine-learning model for entity prediction using selected data;
- generating query tagger data using the profile lexicon, the click-through lexicon, and the machine-learning model;
- receiving a raw query;
- rewriting the raw query as an annotated query, the rewriting based on the generated query tagger data;
- performing a search of a database using the annotated query;
- returning results of the search in response to the receiving of the raw query for presentation in a user interface.
A non-transitory machine-readable storage medium embodying instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
- generating a click-through lexicon from a raw query log;
- training a machine-learning model for entity prediction using selected data;
- generating query tagger data using the profile lexicon, the click-through lexicon, and the machine-learning model;
- receiving a raw query;
- rewriting the raw query as an annotated query, the rewriting based on the generated query tagger data;
- performing a search of a database using the annotated query;
- returning results of the search in response to the receiving of the raw query for presentation in a user interface.
Dependencies are inferred from the claim text.