Loading [MathJax]/extensions/MathMenu.js
User-Specific Adaptive Fine-Tuning for Cross-Domain Recommendations | IEEE Journals & Magazine | IEEE Xplore

User-Specific Adaptive Fine-Tuning for Cross-Domain Recommendations


Abstract:

Making accurate recommendations for cold-start users has been a longstanding and critical challenge for recommender systems (RS). Cross-domain recommendations (CDR) offer...Show More

Abstract:

Making accurate recommendations for cold-start users has been a longstanding and critical challenge for recommender systems (RS). Cross-domain recommendations (CDR) offer a solution to tackle such a cold-start problem when there is no sufficient data for the users who have rarely used the system. An effective approach in CDR is to leverage the knowledge (e.g., user representations) learned from a related but different domain and transfer it to the target domain. Fine-tuning works as an effective transfer learning technique for this objective, which adapts the parameters of a pre-trained model from the source domain to the target domain. However, current methods are mainly based on the global fine-tuning strategy: the decision of which layers of the pre-trained model to freeze or fine-tune is taken for all users in the target domain. In this paper, we argue that users in RS are personalized and should have their own fine-tuning policies for better preference transfer learning. As such, we propose a novel User-specific Adaptive Fine-tuning method (UAF), selecting which layers of the pre-trained network to fine-tune, on a per user basis. Specifically, we devise a policy network with three alternative strategies to automatically decide which layers to be fine-tuned and which layers to have their parameters frozen for each user. Extensive experiments show that the proposed UAF exhibits significantly better and more robust performance for user cold-start recommendation.
Published in: IEEE Transactions on Knowledge and Data Engineering ( Volume: 35, Issue: 3, 01 March 2023)
Page(s): 3239 - 3252
Date of Publication: 14 October 2021

ISSN Information:

Funding Agency:


1 Introduction

The past decade has seen a remarkable progress in deep learning (DL) and their applications in recommender systems (RS). A variety of neural network models [1], [2], [3], [4], [5] with larger and deeper architectures are proposed to model user interaction behaviors from online systems. Among them, sequential recommendation models, such as the GRU4Rec [1], NextItNet [2] and SASRec [4] have become especially popular since they in general require neither much feature engineering nor explicit user embeddings when making recommendations. Despite the success, deep neural network models tend to fail in practice when their training data (i.e., user interactions) is insufficient. Such scenarios widely exist in practical RS, when a large number of new users enroll in but have fewer interactions. Fortunately, the interaction behaviors of cold-start users are likely to be accessible from other online systems. For example, a user in Amazon who has few purchase records might have hundreds of clicking interactions in YouTube. Such observed interaction feedback by YouTube could be a clue to infer her preference and make recommendations in Amazon.

In this paper, we assume data of both domains is available and put aside privacy concerns.

To this end, cross-domain recommendations (CDR) that transfer knowledge from a related source domain, have been proposed and become a popular way to tackle the recommendation problem of cold users.

References

References is not available for this document.