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 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.

Dual Attentive Sequential Learning for Cross-Domain Click-Through Rate Prediction

“…CDR is proposed to transfer knowledge across domains for solving the cold-start and data sparsity issues. Recent CDR approaches can be broadly divided into joint learning [12,15,16,18,19,21,37] methods and pre-training & fine-tuning [1,31] methods. In section 1, we have analyzed key characteristics of existing CDR methods.…”

“…Click-through rate (CTR) prediction [5,22,28,35,36] which estimates the probability of a user to click on a candidate item, is a crucial task for real-world recommender systems (RSs), search engines, and online advertising systems. In Taobao 1 , one of the largest e-commerce platforms in the world, there are multiple recommendation domains where each domain has its own CTR prediction model. Different domains have their specific themes (such as collections of good product, new items, specials, live streaming, etc.…”

“…Previous efforts on cross-domain CTR prediction can be broadly categorized into joint learning methods [16,18,21] and pre-training & fine-tuning methods [1]. The former ones jointly optimize the target domain model and the source domain model.…”

“…However, such methods are often inefficient to deploy online at large portals like Taobao due to the very large-scale source domain training data. Besides, multi-task learning could induce negative impacts to the target domain task because of the conflicts between source domain objectives and target domain objectives [1,30]. Alternatively, pre-training & fine-tuning methods replace target [1].…”

“…Besides, multi-task learning could induce negative impacts to the target domain task because of the conflicts between source domain objectives and target domain objectives [1,30]. Alternatively, pre-training & fine-tuning methods replace target [1].…”