All-in-One: Heterogeneous Interaction Modeling for Cold-Start Rating Prediction

TL;DR

All-in-One: Heterogeneous Interaction Rating Network (HIRE) introduces a data-driven framework to model heterogeneous interactions for cold-start rating prediction. It centers on the Heterogeneous Interaction Module (HIM), which uses stacked multi-head self-attention layers to jointly model user, item, and attribute interactions within sampled prediction contexts. By constructing context tensors from neighborhood sampling and training with a mean-squared error objective, HIRE achieves state-of-the-art performance across three cold-start settings on three real-world datasets, while offering interpretable interaction weights. The approach avoids reliance on external side information or manually defined heterogeneous patterns, unlocking robust recommendations for cold-start users and items with practical implications for real-world recommender systems.

Abstract

Cold-start rating prediction is a fundamental problem in recommender systems that has been extensively studied. Many methods have been proposed that exploit explicit relations among existing data, such as collaborative filtering, social recommendations and heterogeneous information network, to alleviate the data insufficiency issue for cold-start users and items. However, the explicit relations constructed based on data between different roles may be unreliable and irrelevant, which limits the performance ceiling of the specific recommendation task. Motivated by this, in this paper, we propose a flexible framework dubbed heterogeneous interaction rating network (HIRE). HIRE dose not solely rely on the pre-defined interaction pattern or the manually constructed heterogeneous information network. Instead, we devise a Heterogeneous Interaction Module (HIM) to jointly model the heterogeneous interactions and directly infer the important interactions via the observed data. In the experiments, we evaluate our model under three cold-start settings on three real-world datasets. The experimental results show that HIRE outperforms other baselines by a large margin. Furthermore, we visualize the inferred interactions of HIRE to confirm the contribution of our model.

Figures (9)

• Figure 1: An illustrative example: to predict the first girl's preference for music, relation between the girl and a boy seems to be irrelevant. The erroneous classification for cake (food) and movie (entertainment) is unreliable for rating prediction. HIRE directly models heterogeneous interactions in a data-driven way.
• Figure 2: 3 cold-start scenarios for rating prediction: Entities in dotted boxes are the cold-start entities.
• Figure 3: The architecture of HIRE:For ratings of cold user/items to be predicted, HIRE first samples a prediction context. Then, the model constructs the initial embedding for the prediction context and learns the heterogeneous interactions via $L$ HIMs. Finally, the output embedding of the last HIM is transformed into the predicted rating matrix.
• Figure 4: Message Passing in complete graphs via learned interactions
• Figure 5: An example of prediction context construction

Theorems & Definitions (2)

• Example 1
• Example 2