Artificial intelligence system using deep neural networks for pairwise character-level text analysis and recommendations

A method, comprising:

• performing, by one or more computing devices:
• determining, using a first machine learning model trained to perform character-level analysis, a respective feature set corresponding to individual text collections of a first set of text collections pertaining to a first item according to a semantic analysis of the individual text collections, wherein the semantic analysis determines respective features of the individual text collections that indicate that the first item was obtained for a specific application, and wherein the first set comprises a first text collection;
• obtaining, using at least a portion of a second machine learning model, a first score of the first item with respect to one or more semantic criteria for the specific application, wherein input provided to train the second machine learning model is based at least in part on the respective feature sets, and wherein the second machine learning model comprises a pairwise preference learning model;
• generating, based at least in part on the first score, a recommendation associated with the first item;
• and
• providing the recommendation and one or more recommendation explanations, wherein a first recommendation explanation of the one or more recommendation explanations comprises a representation of at least a first portion of the first text collection, wherein the first portion is indicated in the respective feature set determined for the first text collection, and wherein the representation comprises an indication of an explanatory contribution of a particular character relative to one or more other characters.

A system, comprising:

• one or more computing devices of an artificial intelligence service for text analytics;
• wherein the one or more computing devices are configured to:
• obtain, corresponding to individual ones of a plurality of items, a respective set of one or more text collections, wherein a first set of text collections corresponding to a first item comprises a plurality of reviews;
• determine, using a first machine learning model trained to generate character-level predictions of text sequences, a respective feature set corresponding to individual text collections of the first set according to a semantic analysis of the individual text collections, wherein the semantic analysis determines respective features of the individual text collections that indicate that the first item was obtained for gifting;
• obtain, using at least a portion of a second machine learning model, a first gifting-related score of the first item, wherein the second machine learning model comprises a pairwise preference learning model, and wherein input provided to train the second machine learning model is based at least in part on (a) the respective feature sets and (b) one or more gifting-related interaction records;
• generate, based at least in part on the first score, a gift-giving recommendation with respect to the first item;
• and
• provide, via one or more presentation devices, the gift-giving recommendation and one or more recommendation explanations, wherein a first recommendation explanation of the one or more recommendation explanations comprises a representation of at least a portion of a first review of the plurality of reviews, wherein the first portion is indicated in the respective feature set determined for the first review, and wherein the representation comprises an indication of an explanatory contribution of a particular character relative to one or more other characters.