Comic Book Recommendation System for Mobile Webtoon Users

This project focused on developing a recommendation system for a newly launched mobile webtoon service. The goal was to enhance user engagement and retention by providing personalized comic book recommendations, even with limited initial user interaction data.

Project Overview

• Duration: 3+ months
• Role: One of two data scientists sharing project responsibilities

Problem Statement and Project Planning

In the competitive landscape of mobile webtoon services, providing personalized and engaging recommendations from the outset is crucial for user retention and service growth. The challenge was to create a robust recommendation system capable of delivering relevant suggestions even for new users with minimal interaction history.

Key aspects of the problem definition and planning phase included:

1. Stakeholder Collaboration: Extensive discussions with product managers and UX designers to understand user behavior and service requirements.

2. Requirement Gathering: Identifying key features for the recommendation system, including the need to address the cold start problem.

3. Metric Definition: Defining success metrics such as user engagement rates, retention rates, and recommendation click-through rates.

4. Challenge Identification: Recognizing key challenges such as the cold start problem, scalability requirements, and the need for real-time adaptation to user preferences.

Data Analysis and Feature Engineering

My primary responsibilities in this phase included:

1. Data Collection: Designed systems to collect and analyze user interaction data, including reading history, favorites, and ratings.

2. Metadata Analysis: Examined webtoon metadata such as genre, author, and publication date to support content-based filtering.

3. Feature Engineering: Developed relevant features for the recommendation system:
   • Created user profile features based on reading history and preferences
   • Engineered content features from webtoon metadata
   • Developed temporal features to capture trends and seasonality in user behavior

4. Data Quality Assurance: Implemented processes to ensure data integrity and handle missing or inconsistent data.

5. Cross-Functional Collaboration: Maintained ongoing communication with various stakeholders:
   • Collaborated with the product team to align recommendation strategies with overall service goals
   • Partnered with the data infrastructure team to ensure efficient data collection and processing
   • Engaged with UX designers to optimize the presentation of recommendations

Model Development

The modeling approach was tailored to the specific requirements of a new service with limited initial data:

1. Hybrid Recommendation System:
   • Implemented a hybrid model using the LightFM library, combining collaborative and content-based filtering.
   • Leveraged this approach to mitigate the cold start problem by utilizing both user interaction data and comic book metadata.
2. Cold Start Strategies:
   • Developed popularity-based recommendations for new users.
   • Implemented initial user preference surveys to gather immediate preference data.
3. Real-Time Adaptation:
   • Designed the system to dynamically adapt to real-time user behavior, ensuring recommendations remained relevant as more data became available.
4. Model Optimization:
   • Conducted extensive experimentation to balance recommendation relevance and diversity.
   • Implemented techniques to avoid over-recommending popular items at the expense of personalization.
5. Scalability Considerations:
   • Designed the system architecture to efficiently handle growing user bases and content libraries.
6. Evaluation Metrics:
   • Developed custom evaluation metrics that considered both accuracy and user engagement factors.

System Implementation and Monitoring

Key aspects of the implementation and monitoring phase included:

1. A/B Testing: Implemented A/B testing frameworks to evaluate different recommendation strategies.

2. Performance Monitoring: Developed dashboards to track key performance indicators such as user engagement and retention rates.

3. Iterative Improvement: Regularly updated the model based on new data and changing user preferences.

Results and Impact

• Successfully launched the recommendation system, contributing to high initial user engagement rates.
• Achieved significant improvements in user retention compared to baseline non-personalized recommendations.
• Developed a scalable framework capable of adapting to growing user bases and content libraries.
• Created a system that effectively mitigated the cold start problem, providing relevant recommendations even for new users.

This project demonstrated my ability to develop and implement complex recommendation systems that directly impact user engagement and retention. It showcased my skills in machine learning, data analysis, and aligning technical solutions with business objectives in a rapidly evolving digital product environment.

Back to Works