Text-Preserving Lossy Text Compression: A Study of Strategic Deletion and LLM Reconstruction
This study explores lossy text compression by strategically deleting parts of a source text and using a large language model (LLM) to reconstruct it. The goal is to create useful textual artifacts that remain recognizable, unlike abstractive summarization. It tests different deletion strategies—from simple character removal to complex semantic analysis—to find practical methods for context- and bandwidth-limited LLM pipelines.