Actionable Item Detection in Conversations via Neural Model Adaptation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems face challenges in detecting actionable items in conversations between humans without disrupting the flow of communication and often lack sufficient training data specific to the target environment, leading to suboptimal performance.

Innovation Solution

A computer-implemented technique using a machine-trained model, such as a deep-structured convolutional neural network, is employed to recognize actionable items in conversations by converting utterance information into recognized speech and adapting a source-environment model to a target-environment model, leveraging more abundant but less reliable training data from a source environment.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a machine-trained model is used to detect actionable items in conversations, then detection accuracy is improved, but the model requires sufficient training data specific to the target environment which is often scarce

Engineering Contradiction:
Improvedetection accuracyVSAvoidtraining data availability
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent introduces a data translation component that acts as an intermediary between source environment data and target environment detection. This component translates actionable items from the source environment (where abundant training data exists) into the target environment context, enabling the machine-trained model to achieve high detection accuracy without requiring scarce target environment-specific training data.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent creates a virtual copy of the source environment model adapted for the target environment. By copying the successful source environment model architecture and training it with translated data, the system replicates the high detection accuracy of the source model while adapting it to the target environment's specific context and terminology.

Inventive Principle:
Principle #26Copying

2Difficulty of detecting and measuring

If existing systems detect actionable items in conversations, then detection capability is provided, but the flow of communication among participants is disrupted

Engineering Contradiction:
Improvedetection capabilityVSAvoidcommunication flow
Core Design Contradiction:
Difficulty of detecting and measuringVSEase of operation

Solution Approach 1:

The patent extracts the detection function from the communication flow itself and places it in a separate background processing layer. The actionable item detection operates independently of the conversation flow, analyzing audio streams in the background without interrupting or affecting the natural interaction between participants, thus maintaining ease of operation while providing detection capability.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system introduces an intermediary processing layer that sits between the audio input and the communication flow. This intermediary component translates and analyzes actionable items in the background, mediating between the raw audio stream and the communication process without disrupting the natural flow of conversation.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If training data from source environment is used for target environment, then model performance is improved through adaptation, but the training data from source environment is less reliable for target environment

Engineering Contradiction:
Improvemodel performanceVSAvoiddata reliability
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent implements a feedback mechanism in the data translation component that continuously monitors the quality of translated actionable items. The system uses feedback from the translation process to refine and adjust the translation strategies, gradually improving the reliability of source environment data when applied to the target environment through iterative optimization.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent applies parameter changes by adapting the model's detection parameters and translation thresholds based on the specific characteristics of the target environment. The system dynamically adjusts sensitivity, specificity, and translation parameters to optimize the reliability of source environment data application to the target environment, transforming the less reliable source data into reliable target environment insights.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3353776B1Detecting actionable items in a conversation among participants
Publication Date: 2019.07.10 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3353776B1 patent drawingFigure 1
  • EP3353776B1 patent drawingFigure 2
  • EP3353776B1 patent drawingFigure 3A~3D

AI summary

A computer-implemented technique is described herein for detecting actionable items in speech. In one manner of operation, the technique entails: receiving utterance information that expresses at least one utterance made by one participant of a conversation to at least one other participant of the conversation; converting the utterance information into recognized speech information; using a machine-trained model to recognize at least one actionable item associated with the recognized speech information; and performing at least one computer-implemented action associated the actionable item(s).The machine-trained model may correspond to a deep-structured convolutional neural network. In some implementations, the technique produces the machine-trained model using a source environment corpus that is not optimally suited for a target environment in which the model is intended to be applied. The technique further provides various adaptation techniques for adapting a source-environment model so that it better suits the target environment.