Design and Iteration of a Hotness Ranking Algorithm for an Automotive Forum

1. Introduction

As one of the earliest core products of Auto Home, the forum has been a leading Chinese automotive community for 18 years, but changing user habits and the rise of mobile internet now demand a new direction, including modernized product forms, operations, and more efficient content discovery.

The traditional recommendation algorithm is unsuitable because forum sub‑sections have limited content; therefore a hotness algorithm focusing on real‑time popularity and interaction feedback is introduced to break the "filter bubble" and encourage diverse discussions.

2. Mathematical Foundations of the Hotness Algorithm

2.1 History of Hotness Algorithms Early implementations such as Delicious (2003), Hacker News, StackOverflow, and Reddit are referenced as precedents for the forum’s approach.

2.2 Quantifying "Hotness" Key variables include view count, strong interactions (replies), weak interactions (likes, shares), and a time‑decay factor, combined into a mathematical model.

2.3 Creation and Iteration of the Hotness Formula

Initial formula uses variables X1‑X6 and weight coefficients a‑d; parameters are data‑driven. Decay coefficients differ per forum to balance freshness and depth. Testing revealed cheating behaviors (bots, fake clicks), leading to anti‑cheat enhancements using statistical baselines and penalty factors.

Further refinements introduced the F1 score to mitigate manipulation via excessive replies, and sentiment analysis to detect hidden cheating where posts receive unusually high positive feedback.

Robustness is improved with fault‑tolerant design, allowing a post to compensate low scores in one dimension with high scores in others.

2.4 Summary of Formula Development The section recaps the iterative process, emphasizing continuous evolution to achieve fair and accurate hotness ranking.

3. Application and Effects of the Algorithm

3.1 New List Page – Hotness as Sorting Criterion AB tests show significant business metric improvements, though challenges remain for fresh content and low‑traffic sub‑forums.

3.2 Foundational Data for Recommendation Systems Hotness scores now feed recommendation pipelines, automating content selection and enhancing fairness.

3.3 Modernization – Replacing "Featured Posts" with "Hot Posts" The algorithm replaces manual curation, providing a more objective way to surface high‑quality discussions.

4. Conclusion

The paper reviews the hotness algorithm’s theory, development, deployment, and impact, acknowledges limitations such as timeliness, human oversight, and evolving cheating tactics, and looks forward to continuous refinement to serve the forum’s users better.