Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Prompting is not Enough: Exploring Knowledge Integration and Controllable Generation

Tingjia Shen, Hao Wang, Chuan Qin, Ruijun Sun, Yang Song, Defu Lian, Hengshu Zhu, Enhong Chen

TL;DR

GenKI tackles open-domain question answering by separating knowledge integration from output formatting. It consists of a three-stage pipeline: dense passage retrieval, knowledge integration via a joint autoregressive and instruction-tuning objective using retrieved knowledge, and controllable generation with a reward-based ensemble to match target formats. Across TriviaQA, MSMARCO, and CMRC-2018, GenKI achieves state-of-the-art or strong baselines performance, and ablations reveal a linear relationship between retrieval quality and knowledge recall up to model-size bottlenecks. The framework demonstrates robustness when the full knowledge base is unavailable and adapts to multiple answer formats, highlighting its practical impact for reliable, format-consistent OpenQA with LLMs.

Abstract

Open-domain question answering (OpenQA) represents a cornerstone in natural language processing (NLP), primarily focused on extracting answers from unstructured textual data. With the rapid advancements in Large Language Models (LLMs), LLM-based OpenQA methods have reaped the benefits of emergent understanding and answering capabilities enabled by massive parameters compared to traditional methods. However, most of these methods encounter two critical challenges: how to integrate knowledge into LLMs effectively and how to adaptively generate results with specific answer formats for various task situations. To address these challenges, we propose a novel framework named GenKI, which aims to improve the OpenQA performance by exploring Knowledge Integration and controllable Generation on LLMs simultaneously. Specifically, we first train a dense passage retrieval model to retrieve associated knowledge from a given knowledge base. Subsequently, we introduce a novel knowledge integration model that incorporates the retrieval knowledge into instructions during fine-tuning to intensify the model. Furthermore, to enable controllable generation in LLMs, we leverage a certain fine-tuned LLM and an ensemble based on text consistency incorporating all coherence, fluency, and answer format assurance. Finally, extensive experiments conducted on the TriviaQA, MSMARCO, and CMRC2018 datasets, featuring diverse answer formats, have demonstrated the effectiveness of GenKI with comparison of state-of-the-art baselines. Moreover, ablation studies have disclosed a linear relationship between the frequency of retrieved knowledge and the model's ability to recall knowledge accurately against the ground truth. Our code of GenKI is available at https://github.com/USTC-StarTeam/GenKI

Prompting is not Enough: Exploring Knowledge Integration and Controllable Generation

TL;DR

GenKI tackles open-domain question answering by separating knowledge integration from output formatting. It consists of a three-stage pipeline: dense passage retrieval, knowledge integration via a joint autoregressive and instruction-tuning objective using retrieved knowledge, and controllable generation with a reward-based ensemble to match target formats. Across TriviaQA, MSMARCO, and CMRC-2018, GenKI achieves state-of-the-art or strong baselines performance, and ablations reveal a linear relationship between retrieval quality and knowledge recall up to model-size bottlenecks. The framework demonstrates robustness when the full knowledge base is unavailable and adapts to multiple answer formats, highlighting its practical impact for reliable, format-consistent OpenQA with LLMs.

Abstract

Open-domain question answering (OpenQA) represents a cornerstone in natural language processing (NLP), primarily focused on extracting answers from unstructured textual data. With the rapid advancements in Large Language Models (LLMs), LLM-based OpenQA methods have reaped the benefits of emergent understanding and answering capabilities enabled by massive parameters compared to traditional methods. However, most of these methods encounter two critical challenges: how to integrate knowledge into LLMs effectively and how to adaptively generate results with specific answer formats for various task situations. To address these challenges, we propose a novel framework named GenKI, which aims to improve the OpenQA performance by exploring Knowledge Integration and controllable Generation on LLMs simultaneously. Specifically, we first train a dense passage retrieval model to retrieve associated knowledge from a given knowledge base. Subsequently, we introduce a novel knowledge integration model that incorporates the retrieval knowledge into instructions during fine-tuning to intensify the model. Furthermore, to enable controllable generation in LLMs, we leverage a certain fine-tuned LLM and an ensemble based on text consistency incorporating all coherence, fluency, and answer format assurance. Finally, extensive experiments conducted on the TriviaQA, MSMARCO, and CMRC2018 datasets, featuring diverse answer formats, have demonstrated the effectiveness of GenKI with comparison of state-of-the-art baselines. Moreover, ablation studies have disclosed a linear relationship between the frequency of retrieved knowledge and the model's ability to recall knowledge accurately against the ground truth. Our code of GenKI is available at https://github.com/USTC-StarTeam/GenKI

Paper Structure

This paper contains 31 sections, 11 equations, 5 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Open-domain Question Answering
  4. LLMs-based Open-domain Question Answering
  5. RAG-LLMs on Open-domain Question Answering
  6. Preliminary
  7. OpenQA Problem Definition
  8. Instruct tuning on LLMs
  9. Methodology
  10. Knowledge retriever
  11. Knowledge integration
  12. Controllable generation
  13. Experimental Settings
  14. Datasets
  15. Baselines
  16. ...and 16 more sections

Figures (5)

  • Figure 1: An illustration of hallucination encountered when employing LLMs in OpenQA task, along with the variations in answer formats across different OpenQA datasets.
  • Figure 2: Overall framework of GenKI. The left part presents several examples of our prompts used in the knowledge integration and controllable generation module. The right part outlines the process of tuning LLMs to learn knowledge and answer questions in specific task situations.
  • Figure 3: Real cases of outputs after Knowledge integration and controllable generation structure.
  • Figure 4: The linear fitting relationship between quality of retrieved result and knowledge integration effect
  • Figure 5: Comparison of (Above) GenKI, GenKI without knowledge base and best baseline, (Below) GenKI, GenKI in Out-Of-Domain scenario and baselines.