IU4Rec: Interest Unit-Based Product Organization and Recommendation for E-Commerce Platform

TL;DR

The paper addresses the limitation of item-centric recommendations on C2C platforms by introducing IU4Rec, an Interest Unit based two-stage framework that clusters products into Interest Units (IUs) using attribute-driven and semantic clustering (GSID) methods, and redesigns the user interface to support two-stage recommendations. It proposes the IU-Boosted Network, which leverages IU-level features, hierarchical IU click sequences, and an IU-focused attention mechanism to improve CTR prediction. Empirical results on Alibaba Xianyu show offline and online improvements in AUC, GAUC, CTR, and GMV, with an online deployment validating practical impact. Formally, the core task is to estimate the click-through probability $p_{ctr}=p(y=1|x)$ under a production setting, and IU4Rec achieves superior performance by persisting user interests at the IU level even when individual items are sold out.

Abstract

Most recommendation systems typically follow a product-based paradigm utilizing user-product interactions to identify the most engaging items for users. However, this product-based paradigm has notable drawbacks for Xianyu~\footnote{Xianyu is China's largest online C2C e-commerce platform where a large portion of the product are post by individual sellers}. Most of the product on Xianyu posted from individual sellers often have limited stock available for distribution, and once the product is sold, it's no longer available for distribution. This result in most items distributed product on Xianyu having relatively few interactions, affecting the effectiveness of traditional recommendation depending on accumulating user-item interactions. To address these issues, we introduce \textbf{IU4Rec}, an \textbf{I}nterest \textbf{U}nit-based two-stage \textbf{Rec}ommendation system framework. We first group products into clusters based on attributes such as category, image, and semantics. These IUs are then integrated into the Recommendation system, delivering both product and technological innovations. IU4Rec begins by grouping products into clusters based on attributes such as category, image, and semantics, forming Interest Units (IUs). Then we redesign the recommendation process into two stages. In the first stage, the focus is on recommend these Interest Units, capturing broad-level interests. In the second stage, it guides users to find the best option among similar products within the selected Interest Unit. User-IU interactions are incorporated into our ranking models, offering the advantage of more persistent IU behaviors compared to item-specific interactions. Experimental results on the production dataset and online A/B testing demonstrate the effectiveness and superiority of our proposed IU-centric recommendation approach.

Figures (5)

• Figure 1: An overview of the IU4Rec framework, which consists of (a) the construction of interest unit: products are organized into Interest Unit using DSI-based clustering methods on top of the product attributes and textual information Tay2022TransformerMA, and (b) the recommendation of interest unit and the corresponding products. More specifically, the latter component operates in two stages -- stage one recommends products utilizing both item and interest unit attributes, and stage two focuses on identifying products in the given interest unit.
• Figure 2: An illustration of Embedding and MLP structure with sequence modeling for the deep CTR prediction model
• Figure 3: One example of the foundational understanding system generated by the semantic clustering method.
• Figure 4: The redesigned product format. Left is stage one style and right image is stage two style with explanationo on the middle
• Figure 5: An overview of proposed IU-Boosted Network, which consists of three components: (1) the interest unit-level feature for each product, (2) the user's hierarchical IU click sequence to determine their interest unit preference, and (3) the attention mechanism introduced for handling multiple items within the interest unit.