# Enhancing User Personalization in Conversational Recommenders

**Authors:** Allen Lin, Ziwei Zhu, Jianling Wang, James Caverlee  
**arXiv:** [2302.06656](https://arxiv.org/abs/2302.06656)  
**Date:** 2023-02-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2302.06656v1/

## Key points

- Conversational recommenders are popular for personalizing recommendation experiences
- Existing approaches to conversational recommendation have limitations
- The proposed framework introduces a greedy NDCG attribute selector and a user representation refiner
- The framework outperforms state-of-the-art conversational recommenders in terms of performance and efficiency
- It provides a more personalized experience for users under the proposed setting
- Traditional recommender systems face challenges in capturing precise user preferences and providing justifications
- The proposed approach enhances user personalization by prioritizing attributes and refining user representations

## Summary

In the field of recommender systems, conversational recommenders have gained popularity as a powerful tool for personalizing a user's recommendation experience. These systems allow users to engage in back-and-forth dialogues to quickly find the most suitable items. However, many existing approaches to conversational recommendation only partially explore the user preference space and make limiting assumptions about how user feedback should be incorporated. This often leads to long dialogues and poor recommendation performance. To address these limitations, this paper proposes a novel conversational recommendation framework with two unique features. Firstly, it introduces a greedy NDCG attribute selector that prioritizes attributes representing the actual preference space of the user. This enhances user personalization during the interactive preference elicitation process. Secondly, it incorporates a user representation refiner that effectively combines the user preferences collected from the interactive process to obtain a more personalized understanding of the user. The proposed framework is evaluated through extensive experiments on four commonly used datasets. The results demonstrate that it outperforms all state-of-the-art conversational recommenders in terms of both recommendation performance and conversation efficiency. Additionally, it provides a more personalized experience for users under the proposed multi-groundtruth multi-round conversational recommendation setting. In related work, traditional recommender systems face challenges in capturing precise user preferences and providing human-interpretable justifications for recommendations. Existing works have addressed these challenges by leveraging auxiliary data or employing conversational recommenders that interact with users in real-time to elicit detailed preferences. Overall, this paper presents an innovative approach to enhance user personalization in conversational recommenders by prioritizing relevant attributes and refining user representations. The findings highlight its superiority over existing methods and its ability to provide a more personalized recommendation experience for users.