Scholarlink

In this paper, we present two techniques for use in context-aware systems: Semantic Decomposition, which sequentially decomposes input prompts into a structured and hierarchal information schema in which systems can parse and process easily, and Selective Context Filtering, which enables systems to systematically filter out specific irrelevant sections of contextual information that is fed through a system’s NLP-based pipeline. Our objectives are to develop and integrate these techniques for use in dynamic LLM-to-system interfaces, improve contextually cohesive LLM responses, and streamline automated workflows. Experiments are conducted with our synthetic datasets SynAsst, SynPrompt along with the dataset OASST to evaluate our methods. Additionally, we developed a novel scoring technique for measuring the consistency of an LLM’s output named Exponential Consistency Index (ECI). Results indicate that using Semantic Decomposition and Selective Context Filtering yields high ECI scores, indicating a high degree of consistency within outputs. These findings highlight the techniques’ effectiveness in enhancing LLM adaptability, facilitating more responsive and coherent integration into Context-Aware pipelines.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semantic Decomposition and Selective Context Filtering: Text Processing Techniques for Context-Aware NLP-Based Systems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Karl John Villardar,
  • Chris Jordan Aliac

摘要

In this paper, we present two techniques for use in context-aware systems: Semantic Decomposition, which sequentially decomposes input prompts into a structured and hierarchal information schema in which systems can parse and process easily, and Selective Context Filtering, which enables systems to systematically filter out specific irrelevant sections of contextual information that is fed through a system’s NLP-based pipeline. Our objectives are to develop and integrate these techniques for use in dynamic LLM-to-system interfaces, improve contextually cohesive LLM responses, and streamline automated workflows. Experiments are conducted with our synthetic datasets SynAsst, SynPrompt along with the dataset OASST to evaluate our methods. Additionally, we developed a novel scoring technique for measuring the consistency of an LLM’s output named Exponential Consistency Index (ECI). Results indicate that using Semantic Decomposition and Selective Context Filtering yields high ECI scores, indicating a high degree of consistency within outputs. These findings highlight the techniques’ effectiveness in enhancing LLM adaptability, facilitating more responsive and coherent integration into Context-Aware pipelines.