Entity Dense LLM Summaries for Actionable Insights
In the Fall 2024, we performed an exploratory project with Rystad Energy to find prompting techniques and strategies that improve the entity density of Large Language Model (LLM) summaries.
Using DocETL, a system for LLM-powered data processing from the Epic Data lab at UC Berkeley, we explored a wide variety of prompting techniques and strategies on news articles. The entity dense summaries were then used to improve the analytical capabilities of LLMs, by increasing the quality of actionable insights from the news articles, tailored to Rystad Energy’s use case.
Our delivery consisted of the code used to reproduce our best performing pipeline in a Python module, as well as a report on our findings, with details on our test results, speed and economic viability, as well as suggestions for automated evaluations.
-
Rystad Energy is a leading global energy research and business intelligence firm, renowned for its vast databases and in-depth analysis across the oil, gas, and renewable energy sectors. Since its beginning 20 years ago, Rystad has delivered consulting and analytics services to a wide array of entities and is present all over the world, with offices in Oslo, New York, London, Singapore, Rio De Janeiro, Beijing and Sydney.