Systems and methods for automated modeling of quality for products and services based on contextual relevance

A quality assessment system comprising:

• a computer with one or more processors that implement a neural network configured to:
• receive a plurality of descriptive characteristics that describe attributes of a particular product or service and related products or services of competitors from source sites of the particular product or service and the competitors and from reviews of the particular product or service;
• model a relevance of each particular descriptive characteristic from the plurality of descriptive characteristics based on a recurrence of the particular descriptive characteristic in the reviews of the particular product or service and a number of the source sites that mention the particular descriptive characteristic;
• and
• generate a contextual relevance model that differentiates a first set of descriptive characteristics that are relevant to the particular product or service from a second set of descriptive characteristics that are irrelevant to the particular product or service based on a modeled relevance of each descriptive characteristic;
• and one or more processors configured to:
• filter a plurality of reviews directed to the particular product or service based on the contextual relevance model, wherein filtering the plurality of reviews comprises retaining a first set of the plurality of reviews that reference one or more of the first set of descriptive characteristics, and discarding a second set of the plurality of reviews that do not reference one or more of the first set of descriptive characteristics;
• compute a distance separating descriptive characteristics of the first set of reviews in a vector space based on semantic similarity between the descriptive characteristics of the first set of reviews;
• generate a plurality of theme clusters by grouping a first subset of the first set of reviews to a first theme cluster of the plurality of theme clusters based on the descriptive characteristics of the first subset of reviews being within a threshold distance of one another in the vector space, and grouping a second subset of the first set of reviews to a second theme cluster of the plurality of theme clusters based on the descriptive characteristics of the second subset of reviews being within the threshold distance of one another in the vector space;
• and
• present an interface comprising a first visualization for the first theme cluster, and a second visualization for the second theme cluster, wherein the first visualization graphically presents a sentiment expressed in the first subset of reviews and a first theme label that is defined from commonality amongst the descriptive characteristics of the first subset of reviews grouped to the first theme cluster, and wherein the second visualization graphically presents a sentiment expressed in a second subset of reviews and a second theme label that is defined from commonality amongst the descriptive characteristics of the second subset of reviews grouped to the second theme cluster.

A non-transitory computer-readable medium, storing a plurality of processor-executable instructions to:

• retrieve a plurality of descriptive characteristics that describe attributes of a particular product or service and related products or services of competitors from source sites of the particular product or service and the competitors and from reviews of the particular product or service;
• generate a contextual relevance model with a neural network of a quality assessment system, wherein generating the contextual relevance model comprises modeling a relevance of each particular descriptive characteristic from the plurality of descriptive characteristics based on a recurrence of the particular descriptive characteristic in the reviews of the particular product or service and a number of the source sites that mention the particular descriptive characteristic, and wherein the neural network differentiates a first set of descriptive characteristics that are relevant to the particular product or service from a second set of descriptive characteristics that are irrelevant to the particular product or service based on a modeled relevance of each descriptive characteristic;
• filter a plurality of reviews directed to the particular product or service based on the contextual relevance model, wherein filtering the plurality of reviews comprises retaining a first set of the plurality of reviews that reference one or more of the first set of descriptive characteristics, and discarding a second set of the plurality of reviews that do not reference one or more of the first set of descriptive characteristics;
• compute a distance separating descriptive characteristics of the first set of reviews in a vector space based on semantic similarity between the descriptive characteristics of the first set of reviews;
• generate a plurality of theme clusters with an encoder of the quality assessment system, wherein generating the plurality of theme clusters comprises grouping a first subset of the first set of reviews to a first theme cluster of the plurality of theme clusters based on the descriptive characteristics of the first subset of reviews being within a threshold distance of one another in the vector space, and grouping a second subset of the first set of reviews to a second theme cluster of the plurality of theme clusters based on the descriptive characteristics of the second subset of reviews being within the threshold distance of one another in the vector space;
• and
• present, by the quality assessment system, an interface comprising a first visualization for the first theme cluster, and a second visualization for the second theme cluster, wherein the first visualization graphically presents a sentiment expressed in the first subset of reviews and a first theme label that is defined from commonality amongst the descriptive characteristics of the first subset of reviews grouped to the first theme cluster, and wherein the second visualization graphically presents a sentiment expressed in a second subset of reviews and a second theme label that is defined from commonality amongst the descriptive characteristics of the second subset of reviews grouped to the second theme cluster.