Abstract
Cross-platform recommendation aims to improve recommendation accuracy through associating information from different platforms. Existing cross-platform recommendation approaches assume all cross-platform information to be consistent with each other and can be aligned. However, there remain two unsolved challenges: i) there exist inconsistencies in cross-platform association due to platform-specific disparity, and ii) data from distinct platforms may have different semantic granularities. In this paper, we propose a cross-platform association model for cross-platform video recommendation, i.e., Disparity-preserved Deep Cross-platform Association (DCA), taking platform-specific disparity and granularity difference into consideration. The proposed DCA model employs a partially-connected multi-modal autoencoder, which is capable of explicitly capturing platform-specific information, as well as utilizing nonlinear mapping functions to handle granularity differences. We then present a cross-platform video recommendation approach based on the proposed DCA model. Extensive experiments for our cross-platform recommendation framework on real-world dataset demonstrate that the proposed DCA model significantly outperform existing cross-platform recommendation methods in terms of various evaluation metrics.
Introduction
Cross-platform recommendation (CPR) is a powerful tool for mitigating "data sparsity"—the challenge where a system lacks sufficient interaction data for new users. By leveraging a user’s activity on one platform (e.g., Twitter), a system can better predict their preferences on another (e.g., YouTube). However, traditional CPR methods often rely on a "consistency assumption" that ignores two critical factors: Platform-Specific Disparity and Granularity Differences. Users often exhibit different behaviors across platforms, and the semantic detail of content descriptions varies significantly between services.
The DCA Model
The Disparity-preserved Deep Cross-platform Association (DCA) model addresses these challenges using a modified multi-modal autoencoder architecture.
Partially-Connected Structure
To explicitly account for platform-specific disparities, DCA utilizes a partially-connected hidden layer. This layer is divided into three components: a Twitter-specific part ($h^T$), a YouTube-specific part ($h^Y$), and a common part ($h^C$). By restricting connections, the model ensures that unique platform characteristics are preserved in their respective subspaces, preventing them from introducing noise into the common features used for cross-platform association.
Nonlinear Mapping for Granularity
Semantic granularity differences—such as the gap between specific Twitter hashtags and broad YouTube categories—are inherently nonlinear. DCA employs nonlinear activation functions (e.g., Sigmoid) within its deep learning framework to capture these complex relationships, which simple linear models often miss.
Empirical Evidence of Disparity
A core contribution of this research is the empirical demonstration that user interests are not perfectly consistent across platforms. The authors utilized t-SNE visualizations to show that users who appear similar on Twitter often diverge significantly in their YouTube consumption patterns.
Experimental Results
The DCA model was evaluated using the CASIA-crossOSN dataset, which contains linked accounts between Twitter and YouTube. It was compared against several baselines, including Linear Regression (LR), Latent Attribute-based association (LA), and standard Multi-Layer Perceptrons (MLP).
- Association Accuracy: DCA achieved a significant reduction in prediction error, outperforming the LA method with a 9.3% lower Mean Absolute Error (MAE) when predicting YouTube interests from Twitter data.
- Recommendation Performance: In the final video recommendation task, DCA reached an F-score@10 of 0.296, a 6.1% improvement over the LA-based method and a 5.3% improvement over standard MLP approaches.
- Ablation Study: Experiments with a standard (fully-connected) multi-modal autoencoder showed inferior performance compared to the partially-connected DCA, confirming that preserving platform-specific disparity is crucial for accurate association.
Conclusion
The DCA model demonstrates that preserving platform-specific disparities, rather than forcing total alignment, leads to more robust cross-platform recommendations. By combining a partially-connected architecture with nonlinear mapping, DCA effectively bridges the gap between diverse social media platforms. Future work may involve integrating modern transformer-based feature extraction and expanding the model to handle more than two platforms simultaneously.
Summary updated