Abstract:In this era of information explosion, recommendation systems play a key role in helping users to uncover content of interest among massive amounts of information. Pursuing a breadth of recall while maintaining accuracy is a core challenge for current recommendation systems. In this paper, we propose a new recommendation algorithm model, the interactive higher-order dual tower (IHDT), which improves current models by adding interactivity and higher-order feature learning between the dual tower neural networks. A heterogeneous graph is constructed containing different types of nodes, such as users, items, and attributes, extracting richer feature representations through meta-paths. To achieve feature interaction, an interactive learning mechanism is introduced to inject relevant features between the user and project towers. Additionally, this method utilizes graph convolutional networks for higher-order feature learning, pooling the node embeddings of the twin towers to obtain enhanced end-user and item representations. IHDT was evaluated on the MovieLens dataset and outperformed multiple baseline methods. Ablation experiments verified the contribution of interactive learning and high-order GCN components.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve The paper aims to address the challenges faced by recommendation systems in the era of information explosion, specifically the pursuit of broad recall while maintaining accuracy. Although the current dual-tower neural network models in recommendation systems are efficient in large-scale candidate record screening, they fail to effectively model the interactions between features due to the independent training of user and item towers, resulting in poor recommendation accuracy. To solve this problem, the authors propose a new recommendation algorithm model—the Interactive High-Order Dual-Tower (IHDT) model. ### Main Contributions and Methods 1. **Model Design**: - Proposed an IHDT model based on heterogeneous graph construction, extracting richer feature representations through meta-paths. - Introduced an interaction learning mechanism, injecting relevant features between the user and item towers. - Utilized graph convolutional networks for high-order feature learning, aggregating node embeddings to obtain enhanced final representations. 2. **Experimental Validation**: - Evaluated on the MovieLens dataset, showing superior performance across multiple benchmark methods. - Validated the effectiveness of the interaction learning and high-order graph convolutional network components through ablation experiments. ### Analysis of Experimental Results - **Performance Comparison**: The IHDT algorithm outperforms benchmark algorithms on all metrics, particularly excelling on sparse datasets. - **Impact of Recommendation List Length**: As the length of the recommendation list increases, recall and NDCG metrics show an upward trend, while precision metrics smoothly decline. - **Parameter Sensitivity Study**: Investigated the impact of key parameters such as the number of model layers, representation vector dimensions, learning rate, and L2 regularization coefficient on model performance. In summary, the paper proposes a novel recommendation system model, IHDT, aimed at improving the accuracy and diversity of recommendation systems, especially suitable for large-scale recommendation scenarios.

Graph neural network recommendation algorithm based on improved dual tower model

Multi-Behavior Enhanced Heterogeneous Graph Convolutional Networks Recommendation Algorithm based on Feature-Interaction

A multi-intent-aware recommendation algorithm based on interactive graph convolutional networks

SGNNRec: A Scalable Double-Layer Attention-Based Graph Neural Network Recommendation Model

Hi-GNN: hierarchical interactive graph neural networks for auxiliary information-enhanced recommendation

Dual Variational Graph Reconstruction Learning for Social Recommendation

Enhancing interactive graph representation learning for review-based item recommendation

ContextGNN: Beyond Two-Tower Recommendation Systems

Knowledge Graph Double Interaction Graph Neural Network for Recommendation Algorithm

A Study of Recommendation Methods Based on Graph Hybrid Neural Networks and Deep Crossing

A Collaborative Filtering Recommendation Algorithm Based on Community Detection and Graph Neural Network

A Graph-Neural-Network-Based Social Network Recommendation Algorithm Using High-Order Neighbor Information

Dual-Channel Multiplex Graph Neural Networks for Recommendation

A Graph Neural Network Social Recommendation Algorithm Integrating the Multi-Head Attention Mechanism

Reviews Meet Graphs: Enhancing User and Item Representations for Recommendation with Hierarchical Attentive Graph Neural Network.

Knowledge-Enhanced Hierarchical Graph Transformer Network for Multi-Behavior Recommendation

Learning from Hierarchical Structure of Knowledge Graph for Recommendation

Incorporating A Triple Graph Neural Network with Multiple Implicit Feedback for Social Recommendation

A Recommendation Approach Based on Heterogeneous Network and Dynamic Knowledge Graph

Hybrid Graph Neural Network Recommendation Based on Multi-Behavior Interaction and Time Sequence Awareness

Graph Heterogeneous Multi-Relational Recommendation