Invision Community 5: Topic Summaries

[Matt]

Whether you have hours to browse a community or are short on time, scrolling through a very long topic can be more than a little frustrating when you want to follow the topic's core journey.

Recently, we discussed another feature designed to help support-based communities find helpful answers quickly, but what about social topics that do not have a simple question-and-answer format?

We've all come to a lengthy topic for the first time and found it a little intimidating to find the most relevant content among hundreds of posts, which don't always further the topic. These off-topic posts are important because they help social cohesion and build relationships between members at the time of posting. Still, those visiting later often want the truth of the topic.

Invision Community 5 brings a topic summary feature designed to make the most of your time.

The topic summary is generated by an algorithm that uses many touch points such as average read times, reactions, number of shares, external linking and more to determine how useful a post is via a numeric ranking.

The summary shows an estimated read time of the entire topic and an estimated read time using the summary, which gives your members a good idea of the time they'll save.

A shorter read time will make longer topics more accessible to a greater audience.

Adjusting the summary
We believe that algorithms should be used to support human decisions but not override them. Those with permission can add posts from the summary if they feel they are more relevant. Likewise, posts can be removed if you think they are irrelevant.

Interactions with Helpful Posts
Invision Community supports a broad range of communities, including support-based and social communities. We are improving our toolset to help both.

You can have helpful post-voting enabled as well as topic summaries enabled. When this is the case, the topic summary will show until the helpful post-voting meets a threshold. Once that threshold is met, the helpful post information will replace it.

Of course, not all communities and not every forum will have the support features enabled, meaning the topic summary will be the only way to reduce the topic complexity.

Less is more
Browsing the summary gives you a concise view of the topic's journey with no distractions, a vital strategy for growth. 

By allowing members to focus on the core journey, you reward the time they spend on your community and make it more accessible for those short on time.

We hope you've enjoyed this feature introduction and would love to hear your thoughts!

View full blog entry

[DawPi]

Nice! 😮

Available for both types (Cloud / Self-Hosted)?

[Matt]

6 minutes ago, DawPi said:

Available for both types (Cloud / Self-Hosted)?

We haven't made any firm decisions on features and package levels yet.

[Adlago]

To be honest, I have abstracted from your version 5. Why??? There are many factors - the entry of AI, especially in social networks, and probably also in forum systems, which is basically yours... There is something elusive, but irritating entry into the IT sphere... And unpredictable. .. I personally find it puzzling and wakes up my conservative thinking... Sorry, I can't be of any help to your version 5 for now...

Me too, actually, I don't expect anything in terms of speed...ha ha...🥰

[Cedric V]

Beautiful. This will be a very handy feature, well done. 

[Rikki]

I clicked into this very excited, thinking you'd be showing off an AI-generated 'executive summary' of topics. Genuinely curious: why not feed those posts with high impact signals into an LLM to summarize? Not long ago your approach would have been great, but now I think will be in danger of feeling pretty clunky, manual and old fashioned. This seemed like it'd be the perfect opportunity to use a modern tool to solve the problem.

[My Sharona]

 

[Matt]

59 minutes ago, Adlago said:

To be honest, I have abstracted from your version 5. Why??? There are many factors - the entry of AI, especially in social networks, and probably also in forum systems, which is basically yours... There is something elusive, but irritating entry into the IT sphere... And unpredictable. .. I personally find it puzzling and wakes up my conservative thinking... Sorry, I can't be of any help to your version 5 for now...

Me too, actually, I don't expect anything in terms of speed...ha ha...🥰

We are not using AI models at this time.

Client browser speed will be greatly improved in v5, we're already hitting high 90s in Pagespeed scores for mobile out of the box.

[Matt]

24 minutes ago, Rikki said:

I clicked into this very excited, thinking you'd be showing off an AI-generated 'executive summary' of topics. Genuinely curious: why not feed those posts with high impact signals into an LLM to summarize? Not long ago your approach would have been great, but now I think will be in danger of feeling pretty clunky, manual and old fashioned. This seemed like it'd be the perfect opportunity to use a modern tool to solve the problem.

The problem is that too many people want to throw AI at everything. I've seen loads of AI based 'executive summaries' on things like Helpscout, Amazon Reviews, etc that honestly do not offer much value.

We are happy with using signals and a mathematical model for this feature. @Matt Finger has been discussing building a LLM for many things, so it's an area we are interested in.

[Adlago]

12 minutes ago, Matt said:

 

Client browser speed will be greatly improved in v5, we're already hitting high 90s in Pagespeed scores for mobile out of the box.

This is achievable for me and now with 4.x version...I will be glad that you have achieved what you say...

 

[sonimik1]

I'm really excited about everything, but I think I'm getting tired of waiting for at least a beta. It is certain, but very certain that by this time and on this date they would already launch an alpha, but I was wrong. My idea is to try, as many here are desperately waiting, it's simple, we need to try.

[Randy Calvert]

They have said they hope to have a sneak peak by end of the year so they still have a month. You might need to wait a little longer.🙂

[David N.]

Thanks for sharing all those preview videos! I really enjoy the insight into these new features. I really like the direction it's taking so far. 🙂

I personally don't care for A.I. or LLMs, I feel like they are not part of what a community is: a group of human beings sharing a common interest, supporting and helping each other through social interactions. I'm not interested in talking to a machine or reading text that was generated by a machine. 

As for waiting, there's no point releasing a beta if it isn't ready to be released. I don't mind being patient and waiting a little longer - especially if it means ending up with a better, more reliable product. 

[MythonPonty]

 

How could this be seen from the user's perspective?
"Is it motivating to write more quality posts so that I appear in the summary", or can the user think "If I don't appear in the summary anyway, I won't write any more posts". 

I think the tool is nice... But users are sometimes strange in their decisions. How would you see that ?

[KT Walrus]

I hope you never do AI generated topic summaries. I’d rather use the ReaderView in Safari to read topics. If I want summaries of the topic, I’ll use one of the many AI tools available for MacOS to generate AI summaries of text in the browser. Browser makers are starting to build AI into their browsers to do these summaries. Check out Arc’s AI features. I’m pretty sure that Apple and Google will integrate more AI features in the coming year. As long as IC5 supports Reader View in Safari, I’m pretty sure that AI built into the OS will generate these summaries for the user. Use your developers time supporting ReaderView and properly printing Topics including all replies to a PDF for topic archiving on the user’s device. I bet next year’s MacOS release will build in AI features such as article summaries directly into Siri or ReaderView where they belong. 

This reminds me of websites that used to build in spell checking which never worked very well. What a waste of developer time. Some features really belong to the OS or browser so they are well implemented and generally available. 

[Charles]

AI and LLM and all that could be another good metric but community-based metrics are useful too. For example, someone may post a long-winded rant and I might reply "ok."

AI and such would see that as a useless reply. But, in context, a human would see it's a funny, sarcastic reply and it might get lots of likes or other engagement. So you need both AI and actual I working together to create useful data.

[Joel R]

1. For social discussion, I wonder if slowly catching up on all posts is part of the fun? Like if you miss a month of your favorite TV show, do you skip through the highlights reel of the missed episodes, or binge watch and catch up on every missed moment?  

I'm going to be curious how my users actually utilize this in the real world, especially on the social side.  

2. For topics that are solutions oriented, I think this is a very thoughtfully implemented and builds upon a clear trend by IPS of helping us build communities of success that are solution oriented.  

[Matt]

I think it depends on what you’re doing. For example, if you have chosen a book you’ll want to read it page by page but if you’re searching for the right book, you might read the blurb of use the chapter index to see if it’s right for you.

[Rikki]

On 11/29/2023 at 4:18 PM, Matt said:

We are happy with using signals and a mathematical model for this feature.

Just to be clear, I think this part is a good idea - my suggestion was to go a step further and summarize the contents of those posts pulled out by your own model using AI, rather than just outputting the highlighted posts as-is one after the other.

I agree that just feeding an entire topic consisting of 50% junk into an AI summary probably wouldn't give great results. But if you use your signals to pull out the noteworthy posts and then summarize those, I think the result would be likely be pretty good. AI is really good at summarizing text, after all.

[Fast Lane!]

On 11/29/2023 at 12:58 PM, Rikki said:

I clicked into this very excited, thinking you'd be showing off an AI-generated 'executive summary' of topics. Genuinely curious: why not feed those posts with high impact signals into an LLM to summarize? Not long ago your approach would have been great, but now I think will be in danger of feeling pretty clunky, manual and old fashioned. This seemed like it'd be the perfect opportunity to use a modern tool to solve the problem.

We do this at work, makes catching up on looooong Space threads easy when you pop in.  Ref: https://blog.google/products/workspace/hybrid-work-ai-tools-io-22/

[Fast Lane!]

Google Cloud can provide this as a service (amongst other services likely):

https://cloud.google.com/document-ai/docs/workbench/build-summarizer-processor

 

AWS perhaps as well:

https://aws.amazon.com/blogs/machine-learning/use-a-generative-ai-foundation-model-for-summarization-and-question-answering-using-your-own-data/

Edited December 1, 2023 by Fast Lane!

[Matt]

5 hours ago, Rikki said:

Just to be clear, I think this part is a good idea - my suggestion was to go a step further and summarize the contents of those posts pulled out by your own model using AI, rather than just outputting the highlighted posts as-is one after the other.

I agree that just feeding an entire topic consisting of 50% junk into an AI summary probably wouldn't give great results. But if you use your signals to pull out the noteworthy posts and then summarize those, I think the result would be likely be pretty good. AI is really good at summarizing text, after all.

I still think that a summary will rob the community of its character and voices. 

It’s still important to see the author next to their content IMO. 

[PoC2]

Auto-summaries. Excellent idea!

If this ended up being AI-based it would definitely be "Cloud-only" (and I would very much understand why).

Either way (I'm not on Cloud for technical reasons), it's great to see Invision not resting on their laurels and still innovating.

[PoC2]

I'm not a great fan of AI but it seems to be a hydra developing at a rate of knots. Some thought it would take 10 years to achieve what these malgorithms have done in 10 months. If Q* gets released it's yet another game changer, and not a great one for workers.

[Joel R]

18 hours ago, Matt said:

I think it depends on what you’re doing. For example, if you have chosen a book you’ll want to read it page by page but if you’re searching for the right book, you might read the blurb of use the chapter index to see if it’s right for you.

I suppose this is the difference between: 

- A returning member at large who wants to quickly scan for interesting and missed content 

- A returning member who posted in the topic and is therefore more invested 

In regards to the second point, Ive always believed that members like seeing their own content. Ideally, the topic summary would always include that specific members contributions (to a point). There's a psychology of reinforcement of seeing your own contributions or posts, which could be a powerful way of encouraging even more engagement.  

Edited December 1, 2023 by Joel R