Weighted Text Generation for Logical Consistency and Completeness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models (LLMs) struggle to generate texts with deeper meaning and logical consistency, particularly in longer and more intellectually demanding contexts, often omitting significant information and failing to recognize complex relationships.
Innovation Solution
A method and device that utilize input data with identification and weighting information, processed by machine learning models to generate weighted text and speech output, ensuring that only relevant information is included based on user-defined or algorithmically determined priorities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If LLMs generate text based on statistical methods and probabilities, then they can produce grammatically correct and contextually appropriate sentences, but they lack deeper logical consistency and understanding of complex relationships
Solution Approach 1:
The patent introduces an intermediary processing layer between the LLM and the final output. This layer includes components such as a critique model, a knowledge graph generator, and a consistency checker that mediate the transformation of raw LLM output into logically consistent text. The intermediary processes the probabilistic output through multiple validation and refinement stages, adding logical structure without requiring complete redesign of the underlying LLM architecture.
Solution Approach 2:
The text generation process is segmented into multiple independent modules: an LLM for initial text generation, a critique model for evaluating logical consistency, a knowledge graph generator for structuring relationships, and a refinement module for final output. Each module performs a specific function, allowing the system to achieve high logical consistency through coordinated operation of specialized components rather than relying on a single complex model.
2Loss of information
If LLMs process unstructured information to generate coherent text, then they can create meaningful text passages, but they omit significant information and fail to recognize complex relationships
Solution Approach 1:
The system performs preliminary actions by extracting and structuring key information into a knowledge graph before the main text generation process. This pre-processing step identifies complex relationships and significant information elements, organizing them in a structured format that the LLM can reference during generation. This preliminary structuring ensures that important information is preserved and relationships are recognized without requiring the LLM to process all raw information sequentially, thus maintaining efficiency.
Solution Approach 2:
The patent implements a feedback mechanism where the generated text is evaluated by a critique model that checks for information completeness and logical consistency. The critique model identifies omitted significant information and feeds this feedback back to a refinement module, which then generates corrected or supplemented text. This iterative feedback loop ensures comprehensive information coverage while maintaining reasonable generation speed through targeted rather than exhaustive processing.
3Duration of action of moving object
If LLMs use complex statistical methods to predict next words, then they can produce contextually appropriate sentences, but they struggle with longer and more intellectually demanding texts
Solution Approach 1:
The patent adds another dimension to the text generation process by introducing a knowledge graph structure that exists alongside the sequential text generation. This knowledge graph provides a parallel organizational framework that captures long-range dependencies and complex relationships across the entire text, independent of the sequential word-by-word generation process. This dimensional addition allows the system to maintain contextual understanding in longer texts by referencing the structured knowledge graph rather than relying solely on sequential probability predictions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention describes a method for automatically generating a weighted text output and/or speech output based on input data; the method comprising: - providing (S1) input data containing identification information for objects included in the input data; - providing (S2) weighting information for at least one of the objects and/or for at least one of the identification information; and - processing (S3) the input data and/or the objects and/or the identification information and the weighting information by at least one machine learning model to generate the text output and/or speech output weighted on the basis of the weighting information.