Reference-Text Content Selection for Accurate Summary Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing text generation techniques based on neural networks do not effectively incorporate information as text when generating summaries, relying instead on importance scores and length embeddings.

Innovation Solution

A text generation apparatus utilizing neural networks for content selection and generation units that process input text and reference text to generate summaries, incorporating information as text through a content selection unit based on Bidirectional Encoder Representations from Transformers (BERT) and a Transformer-based pointer generator model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If information is provided as importance scores or length embeddings, then the neural network can process it numerically, but the information cannot be given as text

Engineering Contradiction:

Improveinformation representation flexibilityVSAvoidinput processing complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces a reference text as an intermediary between the input text and the generation unit. This reference text serves as a mediator that conveys information (such as length constraints or content guidelines) in a natural language format that the generation unit can directly process, eliminating the need for complex numerical encoding and decoding processes

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the mechanical system of numerical encoding/decoding with a language-based processing system. Instead of converting information into numerical embeddings and then interpreting them, the system directly processes natural language reference texts through the neural network, simplifying the information flow

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Loss of information

If traditional neural network models are used for text generation, then the model can process input text, but it cannot effectively incorporate additional information as text

Engineering Contradiction:

Improveinformation retentionVSAvoidmodel architecture complexity

Core Design Contradiction:

Loss of informationVSDevice complexity

Solution Approach 1:

The patent merges the input text and reference text processing into a unified neural network architecture. The content selection unit and generation unit share parameters and process both texts simultaneously, allowing information from the reference text to be integrated naturally with the input text without requiring separate processing pipelines

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a multi-functional neural network that can process both input text and reference text through the same architecture. The shared parameters enable the model to handle multiple types of textual information uniformly, whether it's the main input text or the auxiliary reference text containing additional constraints or information

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12430510B2Text generation apparatus, text generation learning apparatus, text generation method, text generation learning method and program

Publication Date: 2025.09.30 NT T INC

AI summary

A text generation apparatus includes a content selection unit that acquires a reference text based on an input text and information different from the input text and a generation unit that generates a text based on the input text and the reference text, wherein the content selection unit and the generation unit are neural networks based on learned parameters, so that information to be considered when generating a text can be added as text.