# Large Language Models for User Interest Journeys

**Authors:** Konstantina Christakopoulou, Alberto Lalama, Cj Adams, Iris Qu, Yifat Amir, Samer Chucri, Pierce Vollucci, Fabio Soldo, Dina Bseiso, Sarah Scodel, Lucas Dixon, Ed H. Chi, Minmin Chen  
**arXiv:** [2305.15498](https://arxiv.org/abs/2305.15498)  
**Date:** 2023-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.15498v1/

## Key points

- Large language models (LLMs) have impressive capabilities in natural language understanding and generation.
- LLMs have untapped potential for improving personalized user experiences on recommendation platforms.
- The authors aim to leverage LLMs to provide deeper user understanding and more interpretable recommendations.
- Current recommender systems capture users' interests through historical activities, but these representations are difficult to examine or interpret.
- LLMs can reason through user activities and describe their interests in nuanced and interesting ways, similar to humans.
- "Interest journeys" are persistent user interests that would benefit from personalized descriptions.
- The authors introduce a framework that performs personalized extraction of interest journeys using LLM techniques like few-shot prompting, prompt-tuning, and fine-tuning.
- Extensive experiments on a large-scale industrial platform demonstrate the potential of LLMs in providing deeper, more interpretable, and controllable user understanding.
- LLM-powered user understanding can lead to new journey-aware, assistive, and frictionless conversation experiences on recommendation platforms.
- Collaborative filtering-based approaches do not meet users' needs for higher-level semantic preferences.

## Summary

Large language models (LLMs) have shown impressive capabilities in natural language understanding and generation. However, their potential for improving personalized user experiences on recommendation platforms is largely untapped. In this paper submitted to ACM, the authors aim to address this gap by leveraging LLMs to provide deeper user understanding and more interpretable recommendations. Currently, recommender systems capture users' interests through their historical activities on the platform. However, these generated user representations are difficult to examine or interpret. On the other hand, if we were to ask people about their interests in life, they would talk about hobbies or relaxation routines. The authors argue that LLMs can reason through user activities and describe their interests in nuanced and interesting ways, similar to how a human would. The authors define "interest journeys" as persistent and overarching user interests that would benefit most from personalized descriptions. They introduce a framework that first performs personalized extraction of interest journeys and then summarizes them using LLMs techniques like few-shot prompting, prompt-tuning, and fine-tuning. Through extensive experiments on a large-scale industrial platform, the authors demonstrate the potential of LLMs in providing deeper, more interpretable, and controllable user understanding. The authors believe that LLM-powered user understanding can lead to entirely new user experiences on recommendation platforms that are journey-aware, assistive, and enable frictionless conversation. They argue that collaborative filtering-based approaches do not meet users' needs for higher-level semantic preferences.