LLM Keyword Boosting and Suppression for Multilingual Text Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing large language models (LLM) trained in foreign languages generate unwanted or undesired keywords due to user's inability to modify training data, leading to suboptimal natural language generation in specific language contexts.

Innovation Solution

Implement a boosting keyword set and a suppressing keyword set to control sentence generation using an artificial neural network model, where the boosting keywords are promoted and suppressing keywords are suppressed based on language-specific word distributions in public and proprietary data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a foreign language trained LLM model is used as base model, then the model can generate sentences in multiple languages, but foreign language keywords are generated in Korean language contexts and unwanted keywords cannot be controlled

Engineering Contradiction:

Improvemulti-language generation capabilityVSAvoidkeyword generation accuracy

Core Design Contradiction:

Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent applies local quality by creating language-specific keyword sets (Korean boosting keywords, English suppressing keywords) that are applied locally to control the generation process. The boosting keyword set and suppressing keyword set are tailored specifically for Korean language contexts, allowing the model to maintain multi-language capability while achieving precise control over keyword generation in specific language contexts.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent changes parameters by introducing probability distribution control parameters that adjust the likelihood of generating specific keywords. By modifying the generation probability parameters based on boosting and suppressing keyword sets, the system can control which keywords appear in the generated sentences without retraining the entire model, thus maintaining versatility while improving precision.

Inventive Principle:

Principle #35Parameter changes

2Device complexity

If the base LLM model training data set cannot be modified, then the model structure remains simple, but unnecessary or undesired keywords are generated

Engineering Contradiction:

Improvemodel structure simplicityVSAvoidkeyword generation accuracy

Core Design Contradiction:

Device complexityVSManufacturing precision

Solution Approach 1:

The patent applies preliminary action by pre-defining boosting keyword sets and suppressing keyword sets before the sentence generation process. These keyword sets are prepared in advance based on language-specific characteristics and user requirements, allowing the model to generate accurate keywords without modifying the underlying training data or model structure.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary mechanism (keyword control module with boosting and suppressing keyword sets) that sits between the base LLM model and the final output. This intermediary layer controls keyword generation by adjusting probabilities without requiring changes to the base model's training data, thus maintaining model simplicity while improving keyword accuracy.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Manufacturing precision

If keyword generation is controlled using boosting and suppressing keyword sets, then desired keywords are generated more frequently, but the system complexity increases

Engineering Contradiction:

Improvekeyword generation accuracyVSAvoidprocessing system complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies partial action by focusing keyword control on specific critical keywords rather than attempting to control all words in the generated sentence. The boosting and suppressing keyword sets target only the most important keywords that need control, reducing the overall processing complexity while still achieving significant improvement in keyword generation accuracy.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20250384224A1Apparatus and method of processing natural language using boosting keyword set and suppressing keyword set

Publication Date: 2025.12.18 SIONIC AI INC

AI summary

A natural language processing method performed in an electronic device including at least one processor and at least one memory storing commands to be executed by the at least one processor, the method including acquiring a boosting keyword set including at least one boosting keyword that is an object of generation boost when generating a sentence using an artificial neural network model, acquiring a suppressing keyword set including at least one suppressing keyword that is an object of generation suppression when generating a sentence using the artificial neural network model, and generating sentences through the artificial neural network model based on the boosting keyword set and the suppressing keyword set.