CoActionGraphRec: Sequential Multi-Interest Recommendations Using Co-Action Graphs

TL;DR

CoActionGraphRec (CAGR) is proposed, a text based two-tower deep learning model (Item Tower and User Tower) utilizing co-action graph layers utilizing co-action graph layers to enhance user and item representations on eBay.

Abstract

There are unique challenges to developing item recommender systems for e-commerce platforms like eBay due to sparse data and diverse user interests. While rich user-item interactions are important, eBay's data sparsity exceeds other e-commerce sites by an order of magnitude. To address this challenge, we propose CoActionGraphRec (CAGR), a text based two-tower deep learning model (Item Tower and User Tower) utilizing co-action graph layers. In order to enhance user and item representations, a graph-based solution tailored to eBay's environment is utilized. For the Item Tower, we represent each item using its co-action items to capture collaborative signals in a co-action graph that is fully leveraged by the graph neural network component. For the User Tower, we build a fully connected graph of each user's behavior sequence, with edges encoding pairwise relationships. Furthermore, an explicit interaction module learns representations capturing behavior interactions. Extensive offline and online A/B test experiments demonstrate the effectiveness of our proposed approach and results show improved performance over state-of-the-art methods on key metrics.

Figures (5)

• Figure 1: User behavior sequence example. The user initially wanted to buy a pair of pants. After comparing multiple items, he viewed one of the pants at two time points: t1 and t4 (i.e. "Repeated View"), and then decided to purchase the pair of pants. Then he thought of pairing the pants with a pair of shoes, so he continued browsing and selected a suitable pair of shoes to place an order (i.e. "Complementary Purchase").
• Figure 2: Model Architecture Overview.
• Figure 3: Example of Item Tower and User Tower graph construction. a) The global item graph has 5 items with edges representing the co-actions. We extract the subgraph of item 1 from the global item graph. Here $\{$node = eBay item, edge = co-action$\}$. b) User sequence with 4 steps. Each step corresponds to a node in the user graph. Each node represents a user action on an eBay item at time $t_i$, with $t_1 < t_2 < t_3 < t_4$. The sequence is converted to a directed graph of 4 nodes. The directed relationship is determined by temporal order, since a user action can only be influenced by past actions, not future actions. Here $\{$node = user action on an eBay item, edge = temporal direction (edge weight learned during model training)$\}$.
• Figure 4: Module details. Figure (a) shows the steps of Co-action Aggregation Module, (b) depict the process of edge embedding generation, (c) shows the steps of Explicit Interaction Module
• Figure 5: Example of recommendations for an individual user, (a) shows the user’s historical interacted items, (b)(c)(d) shows the three interests we have captured for this user, with (b) representing the first interest on Athletic Shoes, (c) representing the second interest on Men's Bags, (d) representing the third interest on Pants.