Linguistic Preference Model for Audience-Tuned Text Summarization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional summarization techniques generate summaries that are not human-like and struggle to tune summaries to different target audience vocabularies without altering the original meaning, and they cannot generate multiple summaries tailored to various audience vocabularies effectively.

Innovation Solution

A word generation model combined with a linguistic preference model is used to generate targeted summaries, where the linguistic preference model is trained on target audience data to learn word preferences, and feedback is provided to adjust subsequent word generation, ensuring the summary is tuned to the target audience vocabulary during the generation process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional extractive summarization techniques are used to generate summaries, then key phrases can be extracted from input textual content, but the generated summaries are not human-like and fail to adapt to different target audience vocabularies

Engineering Contradiction:

Improveadaptability to target audience vocabularyVSAvoidmaintainment of original meaning

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The linguistic preference model is trained in advance on target audience training data to learn word preferences before the actual summarization task. This preliminary training enables the model to adapt to different target audience vocabularies without altering the original meaning during the summarization process itself.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system incorporates feedback mechanisms where the linguistic preference model continuously learns from target audience training data, adjusting word preferences based on audience-specific linguistic patterns. This feedback loop enables the model to generate summaries that are both human-like and faithful to the original text while adapting to different audiences.

Inventive Principle:

Principle #23Feedback

2Ease of operation

If post-processing tuning is applied to a generated summary to adapt it to a target audience, then the summary can be adjusted to audience preferences, but the meaning of the original text is changed

Engineering Contradiction:

Improvereadability for target audienceVSAvoidoriginal meaning

Core Design Contradiction:

Ease of operationVSLoss of information

Solution Approach 1:

Instead of applying tuning after summary generation, the system performs the adaptation work in advance by training the linguistic preference model on target audience data. During summarization, the pre-trained model directly generates audience-appropriate summaries without requiring post-processing adjustments that could alter the original meaning.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The linguistic preference model acts as an intermediary between the source text and the target audience, translating the original meaning into audience-appropriate language during the generation process itself, rather than as a post-processing step that risks distorting the original meaning.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If a single summarization model is used to generate summaries for multiple different target audiences, then the system is simpler to maintain, but it cannot generate multiple summaries tuned to different audience vocabularies

Engineering Contradiction:

Improvecapability to generate multiple audience-specific summariesVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The linguistic preference model is designed with multi-functionality to handle multiple target audiences. By training on diverse target audience training data and learning audience-specific word preferences, the single model can generate summaries tailored to different audiences without requiring separate models for each audience type.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS10534854B2Generating a targeted summary of textual content tuned to a target audience vocabulary

Publication Date: 2020.01.14 ADOBE INC

AI summary

A targeted summary of textual content tuned to a target audience vocabulary is generated in a digital medium environment. A word generation model obtains textual content, and generates a targeted summary of the textual content. During the generation of the targeted summary, the words of the targeted summary generated by the word generation model are tuned to the target audience vocabulary using a linguistic preference model. The linguistic preference model is trained, using machine learning on target audience training data corresponding to a corpus of text of the target audience vocabulary, to learn word preferences of the target audience vocabulary between similar words (e.g., synonyms). After each word is generated using the word generation model and the linguistic preference model, feedback regarding the generated word is provided back to the word generation model. The feedback is utilized by the word generation model to generate subsequent words of the summary.