US 12,647,492 B1Granted

Automatically moderating content of media programs using multi-tiered machine learning solutions

AssigneeAmazon Technologies, Inc.

InventorsJuan Martin Borgnino, Sanjeev Kumar, Shenshen Liang, Ayman Mahfouz, Robert Eicher Simmering, Harshal Dilip Wanjari, Muhammad Yahia

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

As a media program is aired to listeners, a control system monitors audio data transmitted to the listeners and interactions received from the listeners to determine whether the media program has violated or may violate one or more rules. The audio data is processed to identify words expressed therein and features of the audio data. Additionally, features of users (e.g., a creator or any listeners or guests) may be calculated based on any information or data available regarding such users. An embedding is formed with data representing the words, the audio features and the user features, and provided to a model trained to determine whether a media program is at risk of violating any rules. One or more actions are selected and executed or recommended based on a score generated by the model representing a level of risk that a rule has been, is being or will be violated.

Claims20

Dependent claims

A first computer system, comprising:

at least one computer processor;
and
at least one data store including one or more sets of instructions stored thereon that, when executed by the at least one computer processor, cause the first computer system to perform actions comprising:
receiving media content from at least one of:
a second computer system associated with a creator of a media program;
or
a third computer system associated with a media source;
causing the media content to be transmitted to at least a fourth computer system associated with a listener in accordance with a first media program;
processing the media content to identify a first portion of the media content corresponding to words and a second portion of the media content not corresponding to the words;
recognizing at least one of the words within the first portion of the media content using at least a first machine learning model;
generating an audio feature based at least in part on the second portion of the media content using at least the first machine learning model;
determining information regarding the first media program, wherein the information regarding the first media program comprises at least one of:
a text-based description of the first media program;
a rating of the first media program;
an acoustic feature of the first media program;
an identifier of at least one of the plurality of listeners to the first media program;
or
an image associated with the first media program;
generating an embedding based at least in part on:
the at least one of the words;
the audio feature;
and
the information regarding the first media program;
providing the embedding to at least a second machine learning model as an input;
receiving an output from the second machine learning model in response to the input;
calculating a score representative of a risk that the first media program has violated at least one rule based at least in part on the output;
and
terminating, based at least in part on the score and prior to completion of playback of the first media program, further transmission, by at least the first computing device, of the media content from the second computer system associated with the creator.

25 more elements

A computer-implemented method comprising:

under control of at least a first computer system configured with executable instructions,
receiving, by at least the first computer system from a second computer system, first data representing first media content of a first media program, wherein the second computer system is associated with a creator of the first media program;
transmitting, by at least the first computer system, second data to at least a third computer system, wherein the second data comprises at least some of the first data, and wherein the third computer system is associated with one of a plurality of listeners to the first media program;
receiving, by at least the first computer system from the third computer system, information representing at least one interaction with the third computer system by the one of the plurality of listeners during a playing of at least the first media content by the third computer system;
identifying, by at least the first computer system, a first portion of the second data representing a set of words spoken by at least one participant during the first media program;
determining, by at least the first computer system, a first feature based at least in part on a second portion of the second data;
determining, by at least the first computer system, a second feature based at least in part on the information representing the at least one interaction;
providing, by at least the first computer system, information regarding at least the set of words, the first feature and the second feature to a first machine learning model as a first input;
identifying, by at least the first computer system, a first output generated by the first machine learning model in response to the first input;
determining, by at least the first computer system, a first level of risk that at least one rule associated with the first media program has been violated based at least in part on the first output;
and
terminating, based at least in part on the first level of risk and prior to completion of playback of the first media program, further transmission, by at least the first computer system, of the first data representing the first media content of the first media program to the third computer system.

9 more elements

A method comprising:

under control of one or more computing systems configured with executable instructions,
determining, by at least a first computer system, information regarding a first media program, wherein the information regarding the first media program comprises:
a text-based description of the first media program;
a rating of the first media program;
an acoustic feature of the first media program;
an identifier of at least one of the plurality of listeners to the first media program;
or
an image associated with the first media program;
determining, by at least the first computer system, information regarding a creator associated with the first media program, wherein the information regarding the creator comprises:
a history of violations of at least one rule associated with the creator;
or
feedback received from the at least one of the plurality of listeners to one of the first media program or a second media program associated with the creator,
providing, by at least the first computer system, information regarding the first media program and the information regarding the creator to a first machine learning model as a first input;
identifying, by at least the first computer system, a first output generated by the first machine learning model in response to the first input;
calculating, by at least the first computer system, a first score representing a likelihood that at least one rule associated with the first media program will be violated based at least in part on the first output;
transmitting, by at least the first computer system, first data to at least a second computer system associated with one of a plurality of listeners to the first media program, wherein the first data comprises media content in accordance with the first media program;
identifying, by at least the first computer system, a first portion of the first data representing a set of words spoken by at least one participant during the first media program;
determining, by at least the first computer system, a first feature based at least in part on a second portion of the first data;
providing, by at least the first computer system, information regarding at least the set of words, the first feature and the first score to the first machine learning model as a second input;
identifying, by at least the first computer system, a second output generated by the first machine learning model in response to the second input;
calculating, by at least the first computer system, a second score representing a likelihood that at least one rule associated with the first media program has been violated based at least in part on the second output;
and
terminating, based at least in part on the second score and prior to completion of playback of the first media program, further distribution by at least the first computing device of the first media program associated with the creator.

20 more elements

Dependencies are inferred from the claim text.