US 2022/0067557 A1Published application

Explaining Semantic Search

AssigneeGaliano Medical Solutions Inc.

InventorsRoger D. Donaldson, Gregory K. Lancaster

Life of this patent

Estimating the term…

Loading patent…

Abstract

The invention uses document retrieval to explain to a human user the properties of a query object that are revealed by a machine learning procedure, lending interpretability to the procedure. A query object is compared to reference objects by transforming the query object and reference objects into representative tokens. Reference objects with many tokens in common with the query object are returned as relevant result objects by a document retrieval system. The token representation furthermore admits comparison between features of the query object and matched features of the reference object or between the query object and groups of reference objects having common features, thus emphasising characteristics of the query and reference objects of semantic importance to the user based on the intention of their search. Embodiments include retrieval of 2-dimensional or 3-dimensional images, audio clips, and text.

Claims19

Dependent claims

A computer-implemented method for providing human-interpretable explanation to output results from a machine learning decision process, said method comprising the following steps performed by a data processing apparatus:

using a tokenizing function, computing query tokens representing a query object (105) being subjected to said machine learning decision process;
accessing a database in which there are stored representative tokens derived, using said tokenizing function, from an assembled corpus of reference objects that are of a same type as said query object, and are therefore useful to provide explanation of semantic features of said query object;
querying said database for identification from among the corpus of reference objects any result objects (110) that each share at least one common token with the query object, the identified result objects thereby forming a subset of the corpus of reference objects:
assessing similarity between the query object and each identified result object in said subset of the corpus of reference objects based on said at least one common token;
grouping the identified result objects of said subset of the corpus of reference objects into at least one result set (305);
sorting the identified result objects within each result set in decreasing order of similarity to the query object;
from among the at least one sorted result set, selecting a collection of reference objects of high-ranking similarity to the query object;
and
causing presentation, to a user, of said collection of identified reference objects alongside the query object such that the semantic features of the query object are easily comparable by said user to the semantic features of each of the identified reference objects in said collection, thereby explaining important semantic features of the query object that contributed to the output results from the machine-learning decision process.

6 more elements

A system for providing human-interpretable explanation to output results from a machine learning decision process, the system comprising a data processing apparatus comprising one or more processors connected or communicable with a data store and a presentation device, and one or more non-transitory computer readable media having stored thereon statements and instructions executable by said one or more processors to perform the following steps:

using a tokenizing function, computing query tokens representing a query object (105) being subjected to said machine learning decision process;
accessing said data store, in which there are stored representative tokens derived, using said tokenizing function, from an assembled corpus of reference objects that are of a same type as said query object, and are therefore useful to provide explanation of semantic features of said query object;
querying said data store for identification from among the reference objects any result objects (110) that each share at least one common token with the query object, the identified result objects thereby forming a subset of the corpus of reference objects;
assessing similarity between the query object and each identified result object in said subset of the corpus of reference objects based on said at least one common token;
grouping the identified result objects of said subset of the corpus of reference objects into at least one result set (305);
sorting the identified result objects within each result set in decreasing order of similarity to the query object;
from among the at least one sorted result set, selecting a collection of reference objects of high-ranking similarity to the query object;
and
causing presentation, on said presentation device, of said collection of identified reference objects alongside the query object such that the semantic features of the query object are easily comparable by said user to the semantic features of each of the identified reference objects in said collection, thereby explaining important semantic features of the query object that contributed to the output results from the machine-learning decision process.

6 more elements

Dependencies are inferred from the claim text.