The authors identify a class of workloads called TADA (Text Analysis in Data Analytics), in which processing text columns is core to answering a user query that may interleave LLM-based semantic steps with exact SQL or Python operations.
Semantic atoms enable reusable text analysis in agentic data workflows
The authors show that exposing row-level tags and category tables allows exact reuse by later symbolic operations, reducing LLM calls by an order of magnitude.
Big Tech
Jinxiang Xie · Zihao Li · Rui Ding · Bartuer Zhou · Shi Han · Dongmei Zhang
Nanjing University · Tsinghua University · Microsoft Research
Research Digest··3 min read
The paper introduces TADA-Agent, a system that uses query-driven semantic atoms, Tagging and Categorization, to materialize intermediate text-processing results as row-aligned columns or category tables, which subsequent SQL or Python steps can reuse without additional LLM calls.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.