The Edge: The Recommendation Funnel

Before making a purchase, people increasingly turn to online platforms and communities to seek out recommendations, opinions, and lived experiences from others. But behind what users see on the surface lies a quieter set of choices — recommendation algorithms that shape how information is surfaced, shared, and ultimately consumed. Ziwei Cong, assistant professor of marketing, discusses new research on how different approaches to algorithmic curation can reshape user behavior, platform dynamics, and the way people connect with information online.

Your paper investigates how recommendation algorithms influence user social and topical interest. What are the main takeaways?

We used a large-scale quasi-experiment on Zhihu — the largest online knowledge sharing platform in China — to analyze its shift from content-base filtering to social filtering. We found that while this transition strengthened the social fabric of the platform, it simultaneously altered how users engage with specific topics. We fiind that switching from topic-based recommendation algorithms, where content is suggested based on the topics to which a user has subscribed, to social filtering, where content is recommended based on what users’ connections engage with, increases social tie formation by about 15% but reduces question subscriptions by roughly 20%. Compared to content-based filtering, social filtering also exposes general users more to content consumed by their followers, who tend to have stronger niche interests.

Why should experts consider this influence when deciding what algorithms to use?

In organizations, this is relevant for people who develop or manage a presence on public or proprietary platforms (e.g., Reddit, Weibo, Facebook, LinkedIn, and Yelp), including but not limited to developers, designers, and social media managers. It helps ensure that users contribute high-quality content and that people asking questions receive useful, relevant answers. Ultimately, it supports the goal of providing knowledge and meaningful responses that improve the overall user experience.

It is also important for policymakers, as online social bubbles have become a significant issue. Social media platforms increasingly place users in filtered environments where conversations and debates are shaped by algorithms that prioritize what we already like or agree with. As a result, people are often surrounded by content and viewpoints similar to their own, reinforcing existing beliefs rather than exposing them to diverse perspectives and communities.

What kind of recommendation system could balance personalization and social connections while supporting long-term platform growth and monetization?

Although we recommend future research examine social filtering in the context of densely connected undirected social networks or other User-Generated Content platforms, our study provided enough insights to believe platforms should adopt hybrid recommendation algorithms that assign different weights to different algorithms for individual users based on user characteristics. In the long run, this helped Zhihu to expand its business model while enabling established users to monetize their expertise and follower base by offering paid content.

This story was originally featured in the Georgetown Business Spring/Summer 2026 Magazine.