Methodologies for Improving Modern Industrial Recommender Systems

Shusen Wang

2023-07-21

Abstract:Recommender system (RS) is an established technology with successful applications in social media, e-commerce, entertainment, and more. RSs are indeed key to the success of many popular APPs, such as YouTube, Tik Tok, Xiaohongshu, Bilibili, and others. This paper explores the methodology for improving modern industrial RSs. It is written for experienced RS engineers who are diligently working to improve their key performance indicators, such as retention and duration. The experiences shared in this paper have been tested in some real industrial RSs and are likely to be generalized to other RSs as well. Most contents in this paper are industry experience without publicly available references.

Information Retrieval,Machine Learning

What problem does this paper attempt to address?

The paper primarily explores how to improve modern industrial recommendation systems (RS) with the aim of enhancing key performance indicators such as user retention and activity. Below is a summary of the main issues the paper attempts to address:

### Issues the Paper Attempts to Address

1. **Optimization of Evaluation Metrics**:
- How to improve key evaluation metrics through enhancements in the recommendation system, including traffic (e.g., daily active users, monthly active users), user retention (e.g., LT7, LT30), user dwell time, number of impressions, and number of clicks.
- Discuss how to balance different metrics, such as making decisions when one metric improves while others decline.

2. **Enhancement of the Retrieval Process**:
- How to improve the performance of the recommendation system by enhancing the candidate generation (retrieval) stage, specifically by improving the two-tower model, enhancing the item-to-item (I2I) model, adding new retrieval models, and introducing new item pools.
- Explore how these improvements can increase user satisfaction and retention.

3. **Upgrading Deep Learning Models**:
- How to improve recommendation accuracy by enhancing deep learning architectures (such as ranking models), including using more advanced neural network structures, feature crossing methods, and multi-task learning methods.
- Discuss the application of online learning techniques and their potential challenges to achieve faster model updates and capture users' latest interests.

4. **Increasing Diversity**:
- How to improve user experience by increasing the diversity of recommended content in the retrieval, pre-ranking, and ranking stages, using methods such as Maximal Marginal Relevance (MMR) and Determinantal Point Process (DPP).

5. **Attention to Special User Groups**:
- How to provide specialized recommendation strategies for new and inactive users, including creating high-quality item pools, adjusting the number of ads, giving special treatment to new items, and adopting specific ranking models.

6. **Utilization of User Engagement Activities**:
- How to further optimize the performance of the recommendation system by encouraging user behaviors such as following, sharing, and commenting, especially for social media recommendation systems.

Through the above methods, the paper aims to provide a series of practical techniques and strategies to help engineers improve existing recommendation systems, thereby better meeting user needs and enhancing overall business performance.