Modeling User Intent Beyond Trigger: Incorporating Uncertainty for Trigger-Induced Recommendation

Abstract: To cater to users' desire for an immersive browsing experience, numerous e-commerce platforms provide various recommendation scenarios, with a focus on Trigger-Induced Recommendation (TIR) tasks. However, the majority of current TIR methods heavily rely on the trigger item to understand user intent, lacking a higher-level exploration and exploitation of user intent (e.g., popular items and complementary items), which may result in an overly convergent understanding of users' short-term intent and can be detrimental to users' long-term purchasing experiences. Moreover, users' short-term intent shows uncertainty and is affected by various factors such as browsing context and historical behaviors, which poses challenges to user intent modeling. To address these challenges, we propose a novel model called Deep Uncertainty Intent Network (DUIN), comprising three essential modules: i) Explicit Intent Exploit Module extracting explicit user intent using the contrastive learning paradigm; ii) Latent Intent Explore Module exploring latent user intent by leveraging the multi-view relationships between items; iii) Intent Uncertainty Measurement Module offering a distributional estimation and capturing the uncertainty associated with user intent. Experiments on three real-world datasets demonstrate the superior performance of DUIN compared to existing baselines. Notably, DUIN has been deployed across all TIR scenarios in our e-commerce platform, with online A/B testing results conclusively validating its superiority.

• The paper introduces DUIN, a model that integrates contrastive learning and Gaussian uncertainty measures to capture both explicit and latent user intent.
• The model outperforms conventional methods with notable gains in AUC and improved CTR on industrial datasets like Alibaba.com using its three-module architecture.
• The framework provides a scalable, adaptive solution for dynamic user interactions, with potential applications beyond e-commerce into broader recommendation tasks.

Evaluating Deep Uncertainty Intent Network for Trigger-Induced Recommendation

The paper introduces the Deep Uncertainty Intent Network (DUIN), a model designed to enhance the efficiency and accuracy of Trigger-Induced Recommendation (TIR) systems. This work is primarily targeted at addressing the constraints and challenges of current recommendation methodologies, which often rely heavily on trigger items—those initial items clicked by users—leading to limited exploration of user intent. The proposed model, DUIN, aims to encapsulate both explicit and latent aspects of user intent while incorporating uncertainty through innovative modeling techniques.

Core Contributions and Methodology

DUIN is structured around three critical modules:

1. Explicit Intent Exploit Module: This module focuses on extracting explicit user intent using a contrastive learning framework. Here, items within a user's behavior sequence that share attributes with the trigger item help model the user's explicit intent. The module applies a contrastive learning paradigm to generate rich, discriminative intent representations, overcoming the challenge of representation degeneration often seen in neural models.
2. Latent Intent Explore Module: To navigate the complexities of latent user intent, DUIN leverages a multi-view relationship model, capturing the probabilistic transitions between items. The model constructs latent intent relevance scores which aid in uncovering not only similar but also popular and complementary items relative to both the trigger and target items.
3. Intent Uncertainty Measurement Module: Unlike traditional models that consider user intent as static, the DUIN captures the fluctuating nature of intent by modeling it as a Gaussian distribution. This distributional approach allows the model to account for uncertainties and changes in user intent, providing a more nuanced recommendation.

The integration of these modules within DUIN allows it to outperform previous approaches. Extensive experiments showed that DUIN offers superior performance on real-world datasets (Alibaba.com, Alimama, and ContentWise), demonstrating notable gains in AUC and Relative Improvement metrics.

Experimental Evaluation

DUIN was evaluated using three distinct datasets which offer insights into its adaptability and robustness across domains. The experiments underscore the model's efficacy against various benchmarks, including conventional click-through rate (CTR) prediction models and existing TIR-centric solutions such as DIHN, DIAN, and DEI2N. Notably, DUIN's performance gains indicate its sophisticated capability in discerning nuanced user preferences and affiliations.

Furthermore, the scalability and real-time applicability of DUIN were affirmed through successful deployment in industrial settings, specifically within major TIR scenarios at Alibaba.com. Here, DUIN demonstrated enhancements in CTR and conversion rates, emphasizing its practical effectiveness in a commercial environment.

Implications and Future Directions

The introduction of uncertainty into intent modeling marks a promising advancement in recommendation systems. The Gaussian distribution approach represents a methodological leap in addressing the variability of user interactions post-trigger engagement. DUIN's framework can potentially be extended or adapted for broader applications in sequential recommendation settings beyond e-commerce, such as content streaming services where user preferences are dynamic.

Future developments could focus on refining the multi-view relationship model to incorporate more sophisticated and diverse interaction data, further enhancing latent intent exploration. Additionally, the approach used in DUIN for intent uncertainty may intersect with reinforcement learning algorithms to dynamically adjust recommendations as user patterns evolve.

In conclusion, the development of the Deep Uncertainty Intent Network represents a significant contribution to the field of recommendation systems, innovating beyond traditional paradigms by modeling user intent with deliberate attention to both explicit and latent dimensions, coupled with a robust uncertainty framework.