# Toward Active Learning in Cross-domain Recommender Systems

## TL;DR

This paper investigates active learning strategies in cross-domain recommender systems, highlighting how auxiliary domain preferences can significantly enhance recommendation quality for new users.

## Contribution

It introduces a novel evaluation framework for active learning in cross-domain RSs and demonstrates the impact of auxiliary domain data on strategy performance.

## Key findings

- Auxiliary domain preferences improve active learning effectiveness.
- Cross-domain scenario differs significantly from single-domain in active learning.
- Evaluation framework tailored for cross-domain active learning.

## Abstract

One of the main challenges in Recommender Systems (RSs) is the New User problem which happens when the system has to generate personalised recommendations for a new user whom the system has no information about. Active Learning tries to solve this problem by acquiring user preference data with the maximum quality, and with the minimum acquisition cost. Although there are variety of works in active learning for RSs research area, almost all of them have focused only on the single-domain recommendation scenario. However, several real-world RSs operate in the cross-domain scenario, where the system generates recommendations in the target domain by exploiting user preferences in both the target and auxiliary domains. In such a scenario, the performance of active learning strategies can be significantly influenced and typical active learning strategies may fail to perform properly. In this paper, we address this limitation, by evaluating active learning strategies in a novel evaluation framework, explicitly suited for the cross-domain recommendation scenario. We show that having access to the preferences of the users in the auxiliary domain may have a huge impact on the performance of active learning strategies w.r.t. the classical, single-domain scenario.