Towards Open-world Cross-Domain Sequential Recommendation: A Model-Agnostic Contrastive Denoising Approach

TL;DR

This paper presents a meta-modelling architecture suitable for cross-domain sequential recommendation (CDSR) with a focus on improving the consistency and effectiveness of models in open-world CDSR scenarios.

Abstract

Cross-domain sequential recommendation (CDSR) aims to address the data sparsity problems that exist in traditional sequential recommendation (SR) systems. The existing approaches aim to design a specific cross-domain unit that can transfer and propagate information across multiple domains by relying on overlapping users with abundant behaviors. However, in real-world recommender systems, CDSR scenarios usually consist of a majority of long-tailed users with sparse behaviors and cold-start users who only exist in one domain. This leads to a drop in the performance of existing CDSR methods in the real-world industry platform. Therefore, improving the consistency and effectiveness of models in open-world CDSR scenarios is crucial for constructing CDSR models (\textit{1st} CH). Recently, some SR approaches have utilized auxiliary behaviors to complement the information for long-tailed users. However, these multi-behavior SR methods cannot deliver promising performance in CDSR, as they overlook the semantic gap between target and auxiliary behaviors, as well as user interest deviation across domains (\textit{2nd} CH).

Figures (4)

• Figure 1: Previous methods focus on constructing their structure based on overlapping users with rich behaviours (a) under a closed-world environment. In this study, we aim to design the model for an open-world environment that accounts for the majority of long-tailed users (b) and cold-start users (c) with sparse historical behaviours.
• Figure 2: Overview of our MACD approach. Unlike previous CDSR methods, our MACD is a general and model-agnostic approach that can be integrated with most off-the-shelf SDSR methods. Our MACD fully leverages auxiliary sequences to explore the potential interests in an open-world CDSR scenario. The denoising interest-aware network (DIN) not only explores explicit interests within the domain but also transfers implicit interest information across domains. With the abundant purified auxiliary sequence information, the representations of long-tailed users can be enhanced. Furthermore, through a well-designed contrastive information regularizer in the DIN and the fusion gate unit, our MACD minimizes the semantic gap and interest deviation between the target and auxiliary behaviors.
• Figure 3: (a)-(d) show the effect of length of sequence on model performance.
• Figure 4: (a)-(d) show the effect of harmonic factor $\lambda$ on model performance..