Natural breakpoints or "chapters" for long topics?

YouTube shows hotspots in videos when you scrub the video timeline:

As you can see, there are peaks and valleys of view count that I always wanted us to get to @sam @eviltrout.

It’s an interesting thought experiment to imagine porting our existing “summarize this topic” feature over to a video … “I ain’t got time to watch a 30 minute video, just show me the best parts” :thinking:

4 Likes

This applies to topics with a large number of hits and comments. A text that has just been published cannot be automatically split into logical parts.

The easiest thing to do with posts. Perhaps it would be worth considering the reactions of users. Then the least useful posts could be shaded or hidden completely, thus compressing the thread. The same rules might be applied to the paragraphs of the first message.

The most popular places from the first post are paragraphs whose quotations were used in the thread.

Then it would be worth considering the tags and picture captions used by the author of the topic, so as not to accidentally hide important things.

We need more active user interactions. Involve them by telling that the content of the topic is shaped by them. “Hey, your opinion is taken into account, look how great it is!”

1 Like

Well yes, this is why the “Summarize This Topic” button only appears when a topic has 50 or more replies… that’s the way it has always worked.

1 Like

Having discussed paginations and infinite scrolling with users on another forum, and since I suspect that what users really miss with infinite scrolling isn’t a pagination strictly speaking, but some sort of breakpoints, I wonder if Discourse’s dev team made progress on these breakpoints thoughts.

4 Likes

It’s striking how useful this “timeline popularity” feature is in practice. Heck, it’s great for podcasts, for long articles, for long chats… really anywhere. Show me what specific parts people are actually reading, listening, and looking at.

The Summarize This Topic is at the top of the topic. So, as you enter the topic at the top and scroll down, you have the option to see “just the good stuff”. The idea that you will mostly enter at the top of the topic is a bit of an assumption, a reasonably safe one I suppose, but…

I can’t help wondering if it’d make sense to advertise the “Summarize” feature (and offer to turn it on) when you (somehow) get deep linked into a longer topic, arriving at the bottom, or anywhere in the middle of a long conversation? :thinking:

4 Likes

I think there is a small issue with encouraging users to go or focus on breakpoints (if a breakpoint is a particularly liked post or a post that had a lot of replies, for example).

It can make people miss other interesting content. They’ll go for the “best” selected discussion content, and then there is a snowball effect. They’ll see and like the selected posts that are displayed to them, and other posts will be ignored because not displayed.

On your screenshot, most of the “most viewed parts” are linked to chapters, but on videos that have few or no chapters, the “most viewed part” exist as well outside chapters marks:

And I also use them thinking it will naturally lead me to what I want to see. Sure. It may in most cases. But how to know without watching the whole actual video (even with fast forwards)?

I can’t think about that without thinking about the Reddit upvote system, where the most upvoted posts are the most likely to be the most upvoted because they are by default on the top of the replies. Snowball effect. And I’ve seen a LOT of top comments on Reddit that contain erroneous information.
As a user, if you want to correct them, you write a message… That no one will ever read it because your message with 0 upvotes will lay at the bottom of XX or XXX comments. That happened to me too many times and I don’t bother to do it anymore.


I really like this topic. This is a very, very interesting subject.

3 Likes

This topic occurred to me when thinking about potential AI integrations with Discourse, so I gave it a quick try:

Can Discourse ship frequent Docker images that do not need to be bootstrapped?
Title: Sam’s Initial Response on Docker Images
Start: 1
Title: Discussion on Docker Compose and Container Design
Start: 18
Title: Discourse Team Explains Their Approach
Start: 23
Title: Community Members Propose Alternatives
Start: 49
Title: Continued Discussion on Container Design Tradeoffs
Start: 136
Title: Updates and Reflections in Recent Years
Start: 165

Why isn't Discourse more frequently recommended as a "community platform"?
Title: Initial Question and Context
Start: 4
Title: CDCK Team Responses and Clarifications
Start: 6
Title: Discourse Positioning and Marketing
Start: 27
Title: User Experiences and Feedback
Start: 125
Title: UI/UX Challenges and Suggestions
Start: 60
Title: Open Source and Self-Hosting Considerations
Start: 97
Title: Comparison to Other Platforms
Start: 155
Title: Future Direction and Development
Start: 151

Discord is taking aim at Discourse. How does Discourse remain unique and stand out from the crowd?
Title: Initial Comparison and Concerns
Start: 1
Title: Discourse’s Advantages
Start: 2
Title: Chat vs Forum Functionality
Start: 6
Title: SEO and Discoverability
Start: 8
Title: Mobile Experience and Notifications
Start: 40
Title: Open Source and Data Ownership
Start: 78
Title: Discord’s New Forum Features
Start: 101
Title: Integration Between Platforms
Start: 110
Title: Future Outlook for Both Platforms
Start: 133

Federation support for Discourse
Title: Initial Proposal and Benefits
Start: 1
Title: Early Challenges and Existing Examples
Start: 3
Title: ActivityPub Implementation Ideas
Start: 14
Title: Renewed Interest and Development Plans
Start: 76
Title: Plugin Specification and Development
Start: 87
Title: Plugin Release and Next Steps
Start: 120

The State of JavaScript on Android in 2015 is... poor
Title: Android JavaScript Performance Lags Far Behind iOS
Start: 1
Title: Examining the Reasons for Poor Android Performance
Start: 39
Title: Exploring Alternatives to Improve Android Experience
Start: 18
Title: Is This Really a Major Problem for Users?
Start: 84
Title: Progress on Android JavaScript Performance Over Time
Start: 246
Title: The Performance Gap Persists in Recent Years
Start: 261

It does need some prompt iteration to add some flavor to those titles, but the core of the idea appears to be workable.

8 Likes

I stumbled upon this thread while researching the same problem, and I was genuinely surprised to see that here had already articulated the “Hotness” idea back in 2014

It’s now 2026, and this feature still doesn’t exist. The pain point, however, very much does. I regularly encounter topics with 1000+ posts where finding the meaningful content feels like searching for a needle in a haystack. The “Summarize” button helps, but it only shows me what people said — not where the interesting moments are spatially located in the timeline.

I spent some time building a script to visualize this concept directly on the timeline. It’s rough, but it demonstrates my idea: horizontal bars extending from the timeline track, where the length represents the reaction count of each post. Clicking a bar jumps you straight to that post.

Through this experiment, I encountered a few practical considerations worth discussing.

Threshold and Filtering: Not every post with 2 reactions deserves a bar. I think it useful to set a minimum threshold, either as an absolute number (e.g., at least 10 reactions) or as a relative percentage (e.g., at least 30% of the highest-reacted post in the topic). Without filtering, the timeline becomes noisy. Additionally, generating a heatmap for topics with only 10 posts seems unnecessary, so a minimum topic size setting would make sense.

Overlapping Bars: When multiple high-engagement posts are clustered together, their bars overlap and become difficult to select. One solution I thought of: when hovering over a bar, the user could press the Up/Down arrow keys to cycle through overlapping bars at that position. The currently focused bar would highlight, and its tooltip would update accordingly.

The original discussion mentioned likes, replies, read time, bookmarks, and internal/external links. My script only uses reaction count because that data is readily accessible. But a weighted composite score might produce better results. I’m curious what the team has explored since 2014 regarding the internal hotness algorithm.

I’m genuinely looking forward to seeing this concept revisited. The timeline is already an excellent navigation tool — adding a hotness layer would make it even more powerful for those of us who dive into long-running topics.

2 Likes