Automatic Recommendation Technique for Data Driven Product Improvement with Large Language Models

Abstract

Large language models (LLM) have an outstanding ability to accurately generate documents and code following the intent of brief instructions input by humans. In the future society, it is expected that ideas conceived by humans will be immediately materialized through LLMs, thereby increasing the importance of continuous and rapid generation of ideas by humans. However, continuously producing product improvement ideas from various perspectives, such as business strategy and usability, is a challenging task for humans. In this study, we aim to develop a technique that automatically recommends improvement ideas for existing products, mainly software products, based on data resources. The proposed technique repeatedly analyzes data and creates improvement ideas through LLMs with prompt templates defined based on the requirements process model. Although the experimental results with ChatGPT did not demonstrate the effectiveness of the proposed technique, they highlighted several new challenges for future study.