Combined wide and deep machine learning models for automated database element processing systems, methods and apparatuses

Inventors

Song, Bing • Balbien, Jeffrey Michael • LU, HAO • Yang, Phillip • Soon-Shiong, Patrick

Assignees

LOS ANGELES TIMES COMMUNICATIONS LLC • Nantmedia Holdings LLC • Nantworks LLC • Immunitybio Inc

Abstract

A method of automated database element processing includes training a wide machine learning model with historical feature vector inputs to generate a wide ranked element output. The method includes training a deep machine learning model with the historical feature vector inputs to generate a deep ranked element output. The method includes generating a set of inputs specific to an individual entity, obtaining a set of current article database elements, and creating a feature vector input according to the set of inputs and the set of current article database elements. The method includes processing the feature vector input with the wide machine learning model to generate a wide ranked element list, processing the feature vector input with the deep machine learning model to generate a deep ranked element list, and merging database elements of the wide and deep ranked element lists to generate a ranked element recommendation output.

Core Innovation

The invention describes an automated recommender system that processes current news article database elements to produce roadblock article recommendations for a target article. The system generates a set of inputs specific to a subscriber, derived from article access records associated with the subscriber, and obtains a set of candidate roadblock articles to create a feature vector input according to the subscriber-specific inputs and the candidate roadblock articles.

The feature vector input is processed using at least one trained machine learning model to generate a wide and deep merged ranked list of roadblock articles. The description specifies using historical article elements and entity access records as training inputs, including unsupervised learning and self-supervised training.

The system selects a roadblock article from the wide and deep merged ranked list, generates a roadblock link for the selected roadblock article, and inserts the roadblock link within a target article. The ranked lists are merged using freshness/click/section/byline scoring, deduplication thresholds, and optional feature-level fusion using an autoencoder to produce dense wide features merged with deep features.

Claims Coverage

The provided independent claims are clm-00001, clm-00021, and clm-00022. Each independent claim includes the same core inventive flow with a wide-and-deep merged ranked list of roadblock articles, subscriber-specific inputs derived from article access records, selecting a roadblock article, generating a roadblock link, and inserting the link within a target article. Across the independent claims, the inventive features are consistent, differing only in whether the subject matter is a system, a method, or non-transitory computer-readable media.

The independent claims focus on producing a subscriber-specific feature vector from article access records, scoring candidate roadblock articles using a wide-and-deep merged ranked list from at least one trained machine learning model, and then selecting a roadblock article to generate and insert a roadblock link within a target article.

Stated Advantages

Documented Applications