Manga Explore System: Accurate, Lightweight, and Scalable Recommendations (v1.3.4)

Problem: Generic suggestions, slow loading and lack of reliability

Many platforms struggle to provide recommendation lists that are truly relevant to individual readers. Suggestions are often generic, lack clear foundation, or take too much time to load, disrupting the user experience.

Manga Explore System solves these limitations by systematically collecting and processing behavioral data from the manga_reading_history table, tracking detailed chapter reading records, access history, and frequently revisited genre groups.

The result is that users receive recommendations that match their taste, display instantly, and fit the context of the manga they’re currently reading.

Process: From raw data to quality recommendation lists

The system starts with Item-based Collaborative Filtering – the highest priority and most stable method. Instead of relying on user preferences, this technique analyzes direct relationships between manga series: “People who read manga A also read manga B”. Using Jaccard Similarity combined with Confidence Score, the system calculates similarity based on shared reader count and generates highly reliable recommendations.

Scoring formula: Combined Score = (Jaccard × 0.3) + (Confidence × 0.7), where Confidence = shared readers / total readers of source manga.

In parallel, for logged-in users with reading history, the system builds detailed reading profiles from the manga_reading_history table:

• List of manga read and chapters completed for each series.
• Reading intensity classification (light, medium, heavy) based on chapter count.
• Preferred genre groups, scored based on weighted chapters read in each genre.

User-based Collaborative Filtering then identifies reader groups with similar behavior, measuring similarity via Cosine Similarity to filter out noise. Manga that similar groups have read but the current user hasn’t will be added to recommendations, scored using the formula similarity × log(chapters_read + 1) to reduce excessive influence from series with massive chapter counts.

The Genre-based from Reading History branch analyzes genre preferences from actual reading history, combined with genres of the current manga. Scores are calculated based on four factors: 40% from favorite genres in history, 30% from genres matching current manga, 20% from popularity (view count), and 10% from follower count.

Results from three branches are merged in multi-signal scoring with weights: Item-based (40%), User-based (30%), Genre-based (30%) to create the final list, prioritizing manga that are both appealing and attractive.

When personal data is insufficient or users aren’t logged in, the system automatically switches to intelligent fallback using only Item-based Collaborative Filtering, ensuring the Explore tab is never empty.

User experience: Smooth, clear and easy to share

The Explore tab is placed in the manga detail page (single-manga.php) but doesn’t affect initial loading speed thanks to lazy fetch mechanism. Content is only loaded when users open the tab, via endpoint /wp-json/initmanga/v1/explore/{manga_id}.

Notably, hash-based URL (#tab-explore) allows direct linking to the recommendation tab, convenient for sharing or bookmarking. HTML results are rendered server-side using the init_manga_render_explore_html() function, automatically detecting layout (Cover Split or Details) and using the item-details template part to ensure interface consistency with the rest of the page.

Performance and optimization: Light on loading, powerful in processing

The system is optimized to use resources only when truly needed. Main techniques include:

• Enhanced batch queries: using OBJECT_K in wpdb->get_results() to create arrays with key = manga_id, completely solving the N+1 Query Problem.
• Item Similarity Cache: caching similarity calculation results between manga series for 6 hours, as this data changes infrequently.
• Reading Profile Cache: storing user reading profiles for 1 week, only rebuilding when new reading activity occurs.
• Multi-level Caching: caching user lists who read manga for 6 hours, caching final recommendation results for 24 hours.
• Logarithmic scoring: using logarithmic functions to limit bias from series with excessive chapter counts.
• Optimized WP_Query: using no_found_rows, meta_value_num and post__in to reduce query costs.

Specifically, the init_manga_get_multiple_user_profiles() function uses batch processing to retrieve reading profiles of multiple users simultaneously, significantly reducing the number of database queries.

Security and privacy: Safe and transparent

Data from the manga_reading_history table is only used for recommendation purposes and never reveals the identity of other users. The REST API endpoint /wp-json/initmanga/v1/explore/{manga_id} returns minimal HTML/JSON, complies with WordPress access permissions, and only processes published manga.

Thresholds are carefully set: minimum Cosine Similarity of 0.15, minimum Combined Score of 0.1 for Item-based filtering, and a limit of 200 users for collaborative filtering to avoid overload.

Algorithm structure, weights, and caching mechanisms are clearly defined in the source code, creating favorable conditions for testing, maintenance, and expansion.

Conclusion

Manga Explore System v1.3.4 brings a new standard for manga recommendation features in WordPress environments.

By combining stable Item-based Collaborative Filtering with personalized User-based Collaborative Filtering, the system ensures high recommendation quality even with sparse data. Performance optimization strategies with batch processing and multi-level caching help readers quickly access works that match their preferences, while ensuring smooth experience and future scalability.