Continuous evaluation and adjustment of search engine results
Inventors
Lachmi, Karin • Levitt, Daniel • Pardo, Ehud • Grunspan, Dan
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11281678-B2
Publication Date
2022-03-22
Expiration Date
Abstract
Taught is a search engine for science tools which dynamically evaluates search rank of said science tools through Natural Language Processing and machine learning. The search engine accepts into a corpus of public and private materials, which references individual science tools. Each item of the corpus is evaluated both as to how much that given item should be trusted, and what that item says about individual science tools. Each science tool is evaluated based on what the whole corpus of input data contains concerning those science tools, taking into account how valuable the source of the data is in order to render an overall score and search rank. The search engine generates a judgement of each individual science tool, which is dynamically updated as new information becomes available in the corpus of input data.
Core Innovation
The invention provides a science-tool search engine that ingests a growing corpus of published and private materials and continuously evaluates the science tools described in the corpus. It semantically identifies objects and source values contained within the corpus of published documents, where each source value is a name of an organization and the objects and source values each include one or more words identified within a published document in the corpus.
The system ties each instance of a first object throughout the corpus to a source value by identifying a first instance of the object in a first published document, identifying a respective source value in text surrounding that instance, and analyzing the textual semantics of the surrounding text and the source value to determine that the source value identifies a source for the object. From these associations, it generates object-source pairs and derives science-tool score factors and overall numeric ratings used for search ranking.
Search requests for a first object are received and a plurality of search results are returned, where the source of the object-source pair varies and the search rank for the first object with a given source is based on a number of times the given source is tied or associated to the first object throughout the corpus. The system assigns trust/value weights to corpus items and combines weighted score factors into overall numeric ratings that drive dynamic search ranking over time, including evaluation and adjustment based on user interaction and analytic signals.
Claims Coverage
Two independent claims are present, covering a method and a system for developing a search engine rank for object-source pairs in a corpus of published documents. The core coverage centers on semantically identifying objects and organization source values, tying object instances to sources based on surrounding-text semantics, generating object-source pairs, and ranking results where the rank is based on the frequency of association across the corpus.
Semantically identifying objects and source values in a corpus
An evaluation module semantically identifies objects and source values contained within a corpus of published documents, where each source value is a name of an organization and where the objects and source values each include one or more words identified within a published document.
Tying object instances to a source using surrounding-text semantics
The evaluation module ties each instance of a first object throughout the corpus to a source value based on identifying a first instance of the first object in a first published document, identifying a respective source value in text surrounding the first instance, and analyzing the textual semantics of the surrounding text and the respective source value to determine the source for the first object.
Generating object-source pairs and ranking by association frequency across the corpus
The search engine receives a search request for the first object and returns a plurality of search results including the object-source pairs, wherein the source varies across results, and a search rank for the first object with a given source is based on a number of times the given source is associated to the first object throughout the corpus of published documents.
Across the independent claims, the patent covers semantically extracting object instances and organization source values from published documents, associating each object instance to a source by analyzing semantics in surrounding text, and ranking search results for object-source pairs primarily by the frequency of the source associated with the object throughout the corpus.
Stated Advantages
Provides dynamic search ranking of object-source pairs based on how often a given source is tied to an object across the corpus of published documents.
Assigns trust/value weights to corpus items and uses overall numeric ratings to drive search ranking and continuous evaluation and adjustment.
Documented Applications
Science-tool search within a corpus of published documents, returning search results for a requested object with different source values (organization names) and providing ranking based on corpus association frequency.
Interested in licensing this patent?