Leveraging LLMs for KPIs Retrieval from Hybrid Long-Document: A Comprehensive Framework and Dataset

“In order to extract information about the keywords from multiple selected segments, several popular strategies are available. Besides the Refine Strategy employed in our method, another commonly used approach is the Map-Reduce Strategy. As illustrated in Figure 4, the Map-Reduce Strategy aims to generate a summary from multiple document segments, consisting of two stages: Map and Reduce. In the Map stage, the LLM generates a segment summary for each document segment individually. While in the Reduce stage, the LLM combines all the segment summaries to form a single document summary.”