From SNS for "Likes" to SNS for "Understanding": Can Changing Algorithms Reduce Division?

When we open SNS, we feel as if we are choosing information ourselves. We read posts that catch our interest, watch entertaining videos, and hit "like" on opinions we resonate with. The timeline appears to be a mirror reflecting our interests.

However, in reality, that mirror is not transparent. What appears at the top, what becomes less visible, and which posts keep us engaged longer are all decided by the platform's algorithms. The information environment we think we are choosing is largely designed.

A new study by researchers from the University of Copenhagen, the Technical University of Dresden, and the Max Planck Institute for Human Development strongly reaffirms this seemingly obvious but often overlooked fact. The conclusion of the study is simple: by slightly changing the way SNS posts are arranged, the polarization of opinions and the accuracy of judgments about reality can be altered.

In other words, the division on SNS is not happening solely because "users want it that way." The design of the feed itself may be expanding or narrowing societal perceptions.

A Pleasant Feed Isn't Always the Right One

In today's major SNS platforms, engagement is often a crucial metric. Likes, shares, comments, reactions, and time spent. Posts that elicit strong user reactions are deemed more valuable and are more likely to be seen by a larger audience.

This mechanism is very rational from a business perspective. The more users react, stay, and return, the more valuable the platform becomes. It also aligns well with the advertising model. Posts that evoke strong emotions like anger, surprise, empathy, fear, or a sense of superiority are less likely to make users close the screen.

The problem is that information that provokes strong reactions does not necessarily lead to accurate understanding.

The study showed that engagement-focused feeds, especially personalized engagement rankings that prioritize posts favored by politically similar individuals, tend to polarize participants' beliefs and reduce the accuracy of their judgments.

Ironically, such feeds were often perceived as "insightful" by participants. Posts that align closely with one's own thoughts, are emotionally satisfying, and provoke reactions seem beneficial to the individual. However, they do not necessarily help in understanding the world more accurately.

In fact, the more comfortable and agreeable the information environment, the more it may reinforce one's own viewpoint and obscure opposing perspectives. This is where the danger of SNS lies. Users may think they are receiving "information that suits them," but in reality, they might be surrounded by "information that reinforces their biases."

How Was the Experiment Conducted?

To measure how SNS algorithms affect polarization and accuracy, the research team conducted a two-stage online experiment targeting residents of the United States.

In the first stage, 500 participants evaluated a total of 72 short, controversial posts on six themes related to politics and society. Participants chose to agree, disagree, or remain neutral on each post. This made it visible which posts liberal and conservative participants supported or rejected.

In the second stage, 1,000 new participants were introduced. Participants first indicated their thoughts on each theme, then viewed a short feed consisting of three posts, and finally stated their thoughts again.

Which posts were displayed depended on the conditions assigned to the participants. The study mainly compared the following ranking methods:

A method that arranges posts randomly.
A method that prioritizes posts with high engagement.
A personalized engagement method that prioritizes posts favored by one's political group.
A bridging method that prioritizes posts supported by both liberals and conservatives.
An intelligence method that prioritizes posts expected to enhance the accuracy of the group's judgment.

Researchers measured whether participants' thoughts converged or diverged and how accurate the group's judgments became. They also examined how participants felt about the posts, such as whether they found them polite, emotional, or insightful.

What makes this design interesting is that, despite being an SNS experiment, it does not rely on internal data from large platforms. The researchers did not conduct large-scale interventions on existing SNS but changed only the arrangement of posts in a relatively controlled environment. As a result, it was shown that differences in arrangement alone could affect belief formation.

What Is a "Bridging" Algorithm?

The focus of this study is on the bridging algorithm.

This approach prioritizes posts that receive a certain level of support from people with differing political stances, rather than posts that strongly resonate with only one side. For example, instead of posts that only liberals enthusiastically support or posts that only conservatives strongly react to, it prioritizes posts that both sides feel are "at least worth reading."

This mechanism is different from merely arranging "neutral-looking posts." On themes with social conflict, completely neutral opinions rarely exist. What matters is whether people with different stances can start a discussion from the same information.

The bridging method does not eliminate the opposition of opinions. Rather, it prioritizes information that allows mutual recognition of each other's existence, assuming that opposition exists. While SNS currently excels at "amplifying empathy within groups," the bridging method aims to form a "foundation that can be shared even among different stances."

The study showed that this bridging method could potentially enhance agreement between liberal and conservative participants in some cases. This serves as an important counterexample to the pessimistic view that SNS inevitably deepens division.

Another Key: "Collective Intelligence"

Another alternative is the intelligence ranking. This method prioritizes posts expected to enhance the accuracy of the entire group's judgment.

Here, accuracy does not simply mean "aligning with the majority opinion." On themes like social predictions or factual judgments, where correctness or validity can be somewhat verified, the issue is whether the group's judgment becomes closer to reality.

On SNS, the loudest, most emotional, and most easily spread voices often stand out. However, to make better collective judgments, the most prominent opinions are not always necessary. Sometimes, calm and unassuming but important information, verifiable claims that are not extreme, and explanations that do not provoke reflexive anger are needed.

The intelligence ranking is precisely the idea of bringing such information to the forefront. The study showed that this method tended to enhance the accuracy of factual judgments compared to random displays or engagement-focused displays.

This could significantly change the design philosophy of SNS. The question has been "What posts make users react more?" But another question is possible: "Which posts should be shown for the user group to understand the world more accurately?"

SNS Reactions: Expectations, Caution, and a Sense of Reality

As a news item immediately after its release, this study has not yet widely visualized large-scale SNS reactions to the article itself. It is difficult to confirm a noticeable number of comments on Phys.org, and general user reactions may spread in the future. Meanwhile, posts on LinkedIn by the researchers themselves and related field researchers highlight the practical significance of this study.

Particularly emphasized is the point that alternative SNS algorithms can be experimented with without partnering with giant tech companies. Even without direct access to the complex recommendation systems within platforms, consensus and accuracy-focused rankings can be verified using reaction data to posts and basic attribute information. This holds significant meaning for researchers and policymakers.

Additionally, the point that it can be implemented without using AI is likely to attract interest. In recent years, discussions on improving SNS have tended to focus on AI-based detection of harmful posts and fact-checking. However, what this study has shown is that even without advanced generative AI or complex classification models, merely changing the arrangement of posts can potentially alter social impacts.

On the other hand, when looking at general discussions about SNS, the reaction is not entirely optimistic.

The first question that arises is, "Who decides what is accurate?" The intelligence ranking is appealing, but determining what is considered correct and which themes can have objective answers is challenging. On topics involving political, ethical, and cultural value judgments, the word accuracy itself can become a point of contention.

Next, there is concern that "bridging" might prioritize only safe opinions. When choosing posts supported by both opposing camps, while extreme misinformation might be suppressed, there is also the possibility that important minority accusations or inconvenient challenges to the existing majority might be buried. Many opinions that have changed society were not initially accepted by both camps.

Furthermore, from the user's perspective, there might be a backlash against "having the timeline manipulated by someone again." Many are dissatisfied with engagement-focused algorithms, but that does not mean they will unconditionally accept feeds designed with different values. Among SNS users, there is a fatigue with algorithms themselves. There is a strong voice asking to stop recommended displays and return to chronological displays, allowing more personal choice.

What makes this study interesting is that it does not suppress such questions but rather provides a foundation for discussion. By showing that engagement optimization is not the only option, it raises the question, "What should be optimized?" to society.

SNS for Business or SNS for Democracy?

The biggest issue raised by this study is the question of what SNS should be designed for.

For companies, the most straightforward success metrics are user time spent and reaction numbers. The more people use it for longer and react frequently, the easier it is to succeed as a business. However, the desirable SNS metrics for society should not be limited to that.

Can people understand each other?
Is it difficult for misinformation to spread?
Can people with different stances talk about the same reality?
Does verifiable information reach instead of anger and insults?
Can minority voices be preserved while suppressing extreme agitation?

These are conditions that are not directly reflected in advertising revenue metrics but are essential for SNS to function as a space for public discussion.

Researchers also point out that platforms themselves may not actively adopt such alternative algorithms, as they might sacrifice engagement. A feed where users remain angry and stay longer might be better for business, even if a feed that encourages calm understanding and early departure is better for society.

This is where discussions on policy and regulation come in. If SNS plays a significant role as a public space, should its design be entirely left to corporate revenue goals? Should transparency be demanded? Should third-party audits be introduced? Should users be allowed to choose from multiple algorithms? Or should engagement optimization be restricted on themes with high social risk?

This study provides experimental evidence for such discussions.

However, It's Not a Panacea

Of course, it's premature to conclude that "changing algorithms will solve division" based on this study alone.

The experimental environment is much simpler than real SNS. The feeds participants view are short, and the themes are limited. In actual SNS, friendships, influencers, ads, bots, news media, video recommendations, comment sections, quoted posts, and external links are intricately intertwined. Real political attitudes are not formed by just a few posts.

Moreover, there are factors other than algorithms that contribute to polarization. Economic insecurity, regional differences, education, media environment, party politics, and cultural identity are deeply involved in the structure of society itself. While SNS may amplify these, they are not the sole cause.

Nevertheless, what makes this study important is that it shows "at least feed design can be changed." It is not a magic solution to immediately resolve societal division, but it is not a technical design that should be left unchecked either.

What Kind of Timeline Do We Want?

When considering the future of SNS, it is often discussed as a binary choice between "freedom of expression or regulation." However, what this study indicates is the design issue that lies before that.

Even if the same posts exist, the social impact changes depending on the order, to whom, and in what context they are displayed. It's not just about deleting or keeping posts, but what is highlighted that matters.

SNS has grown by maximizing reactions. As a result, we may feel like we are being shown "what we want to see," but in reality, we might have been shown "what we can't help but react to."

However, another SNS is possible. A feed that connects understanding rather than amplifying anger. A feed that finds a reality that can be shared with people of different stances, rather than just strengthening the correctness of one's own camp. A feed that brings forward posts that slightly improve judgment rather than those that go viral.

Whether such an SNS is truly attractive to many users is still unknown. But at least this study has shown that the current engagement supremacy is not the only path.

Delivering information that deepens "understanding" rather than information with many "likes."
If SNS can move in that direction, the timeline could become a place for society to re-confirm reality with each other, rather than just a place to kill time.

The issue is not just whether it is technically possible. It is about which values we and the platforms prioritize.

---

Source URL

Phys.org article: Refer to the research overview, experimental design, explanations of engagement, bridging, and intelligence algorithms, and researcher comments.
https://phys.org/news/2026-05-alternative-algorithms-users-accurate-polarized.html

University of Copenhagen news release: Announcement by the research institution. Refer to research objectives, main results, two-stage experiment with 500+1000 people, and explanations of each ranking condition.
https://news.ku.dk/all_news/2026/05/alternative-social-media-algorithms-can-help-users-form-more-accurate-and-less-polarized-beliefs/

EurekAlert! news release: Research news distribution operated by AAAS. Refer to DOI, research methods, subjects, publication date, and experimental overview.
https://www.eurekalert.org/news-releases/1128531

ACM Digital Library paper page: Refer to the research paper "Simple changes to content curation algorithms affect the beliefs people form in a collaborative filtering experiment" for DOI and publication information.
https://dl.acm.org/doi/10.1145/3772318.3790602

LinkedIn posts by researchers: Introduction of the research by authors and co-authors. Refer to points about experimenting with alternative algorithms without collaboration with giant platforms, the absence of AI, and evaluating consensus and accuracy.
https://www.linkedin.com/posts/stefanmherzog_ever-worried-how-social-media-affects-our-activity-7437795175004909568-gbyK