Beyond Item Dissimilarities: Diversifying by Intent in Recommender Systems (2405.12327v2)

Abstract: Recommender systems that overly focus on short-term engagement prevents users from exploring diverse interests. To tackle this challenge, numerous diversification algorithms have been proposed. These algorithms typically rely on measures of item similarity, aiming to maximize the dissimilarity across items in the final set of recommendations. In this work, we demonstrate the benefits of going beyond item-level similarities by utilizing higher-level user understanding--specifically, user intents that persist across multiple interactions or recommendation sessions--in diversification. Our approach is motivated by the observation that user behaviors on online platforms are largely driven by their underlying intents. Therefore, final recommendations should ensure that a diverse set of intents is accurately represented. While user intent has primarily been studied in the context of search, it is less clear how to incorporate real-time dynamic intent predictions in recommender systems. To address this gap, we develop a probabilistic intent-based whole-page diversification framework for the final stage of a recommender system. Starting with a prior belief of user intents, the proposed framework sequentially selects items for each position based on these beliefs and subsequently updates posterior beliefs about the intents. This approach ensures that different user intents are represented on a page, towards optimizing long-term user experience. We experiment with the intent diversification framework on YouTube. Live experiments on a diverse set of intents show that our framework increases Daily Active Users and overall user enjoyment, validating its effectiveness in facilitating long-term planning. Specifically, it enables users to consistently discover and engage with diverse content that aligns with their underlying intents over time, leading to an improved long-term user experience.

Diversifying by Intent in Recommender Systems

Introduction

Recommender systems form the backbone of many online platforms, helping users discover content that aligns with their interests. Traditionally, these systems have been designed to maximize short-term metrics such as clicks and likes. However, there’s a growing understanding that this approach can inadvertently harm the long-term user experience. Think about it—clickbait articles or overly repetitive recommendations can turn users off after a while.

This paper presents a fascinating approach to tackle this challenge by incorporating user intents into the recommendation process. By understanding and predicting a user’s underlying intents that persist across multiple visits, the system can diversify recommendations in a way that enhances long-term satisfaction. Let's dive in and see how they achieved this.

Core Concept: Intent-Based Diversification

The key idea of the paper is to use user intents to inform the final recommendation stage. User intents are the underlying motivations behind why users behave the way they do on a platform—such as seeking entertainment, learning something new, or exploring unfamiliar content.

1. Intent Prediction: The system first predicts the user's intent using a machine learning model that analyzes signals such as past interactions, time of day, and session length.
2. Diversification Framework: Armed with these intent predictions, the system employs a probabilistic model to diversify the recommended items. This model makes sure that items covering different intents appear on the same recommendation page.

The authors proposed a clever algorithm for diversification, which calculates the probability that each item will satisfy the user's predicted intents. By sequentially picking items that maximize this probability, they ensure a diverse and relevant set of recommendations. This approach helps in achieving a balance between showing highly relevant items while also introducing new and varied content that aligns with different user intents.

Strong Numerical Results

The framework was tested on a massive content recommendation platform that serves billions of users daily. Here are some key findings from their live A/B tests:

• User Enjoyment: There was a significant increase of 0.09% in overall user enjoyment.
• Daily Active Users (DAU): DAU improved by 0.05%, suggesting that users were more likely to return to the platform regularly.
• Diversity and Exploration: Users engaged more with novel content providers, indicating a higher quality of exploration.

While these numbers may seem modest, they translate to substantial improvements given the scale of the user base involved.

Implications and Future Directions

The approach described in the paper has several practical implications:

• Balanced User Experience: By balancing short-term satisfaction with long-term engagement, the framework addresses two often competing goals.
• Dynamic Adaptation: The system adapts to the user's changing intents, ensuring a personalized and evolving recommendation experience.

Theoretically, this work opens the door to more comprehensive user models that consider long-term behaviors and outcomes. Future research could explore the integration of more intricate intent taxonomies or the application of this approach to other types of recommender systems, such as e-commerce or social media.

Conclusion

This paper makes a compelling case for incorporating user intents into the recommender system to optimize for long-term user satisfaction. Through a well-thought-out probabilistic diversification algorithm, it manages to improve key metrics such as user enjoyment and DAU, while also encouraging users to explore new content.

As AI continues to advance, frameworks like these will likely become a staple in the design of smarter, more user-centric recommendation systems. The next steps could involve refining the intent prediction models or expanding the types of intents to cover more user motivations.

Overall, this work is a significant step towards making our digital experiences richer and more fulfilling in the long run.