Smart Reply Diversity via Synonym Hierarchy Graph

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Generating smart replies for users can drain electronic resources due to the computational load and limited display space on devices, particularly mobile devices, where excessive redundant replies waste valuable screen space and resources.

Innovation Solution

The system uses a hierarchical graph data structure and diversity rules to select and display a limited set of smart replies, ensuring diversity by limiting the number of replies with common parent nodes, thereby optimizing resource usage and screen space on devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the system generates a large number of candidate replies to provide diverse smart reply options, then the diversity and relevance of replies improve, but the computational load and electronic resource consumption increase

Engineering Contradiction:
Improvediversity of smart repliesVSAvoidcomputational load
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The system pre-generates a hierarchical graph of synonyms and related terms offline, organizing them by semantic categories and relationships. This preliminary structuring allows the system to quickly retrieve and combine pre-processed linguistic elements during runtime, avoiding the need to generate all possible replies from scratch for each user interaction, thus reducing computational load while maintaining reply diversity.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The reply generation process is divided into multiple stages: first selecting a core concept from the input message, then retrieving related synonyms and phrases from different hierarchical levels, and finally combining these elements according to diversity rules. This segmentation allows the system to manage computational resources efficiently by processing only relevant portions of the linguistic space at each stage rather than evaluating all possible replies simultaneously.

Inventive Principle:
Principle #1Segmentation

2Adaptability or versatility

If the system displays many smart reply options to provide users with more choices, then the user experience improves, but the display screen space is wasted due to redundancy

Engineering Contradiction:
Improvenumber of reply optionsVSAvoiddisplay screen space
Core Design Contradiction:
Adaptability or versatilityVSArea of stationary object

Solution Approach 1:

The system applies different selection criteria to different positions in the reply list. High-probability, most relevant replies are placed in prominent positions with larger visual weight, while lower-probability alternatives are placed in less prominent positions. This local differentiation allows the system to provide diverse options without uniformly occupying screen space, as less important replies require less visual real estate.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system generates a larger set of candidate replies than are ultimately displayed, then applies filtering and selection rules to reduce this set to a manageable number that fits within available screen space. By generating more candidates than needed and then selectively displaying only the most relevant ones, the system ensures adequate diversity while respecting display constraints.

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If the system evaluates numerous parameters for each user to provide personalized replies, then the relevance and personalization improve, but the computational expense and training data requirements increase

Engineering Contradiction:
Improvepersonalization accuracyVSAvoidmodel parameters
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system employs a small set of universal linguistic rules and hierarchical graph structures that can be applied across all users and contexts. Instead of maintaining separate complex models for each user, the same core linguistic knowledge base and combination rules serve all users, providing consistent quality while reducing the need for extensive user-specific training data and model parameters.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentUS11334714B2Generating diverse smart replies using synonym hierarchy
Publication Date: 2022.05.17 MICROSOFT TECHNOLOGY LICENSING LLC
  • US11334714B2 patent drawing
  • US11334714B2 patent drawing
  • US11334714B2 patent drawing

AI summary

Techniques for generating diverse smart replies using a synonym hierarchy are disclosed herein. A computer system may detect that a first set of one or more messages having first content has been transmitted from a first computing device of a first user to a second computing device of a second user, determine a plurality of candidate replies based on the first content of the first set of one or more messages, and then select a plurality of smart replies from the plurality of candidate replies using a hierarchical graph data structure and at least one diversity rule. The selecting of the plurality of smart replies comprises omitting at least one of the plurality of candidate replies from selection based on the at least one diversity rule, which limits a number of the plurality of smart replies that have a common parent node.