[dzink]

Congratulations on launching and on the architecture choice and dev insights!

With regards to the product, there are several consequences you are about to see and need prepare to mitigate. I worked on something similar and need to rewrite it, thus the learnings.

1. The site looks like a dream come true for political recruiters/promoters. You have a place full of users self-identifying with political labels in great detail and you can reach out to anyone. You will see a ton of spam from companies, parties, surveyors, you name it. You may want to throttle it by limiting number of new people that can be reached by a user in 24 hrs.

2. Scammers or predators could bait people into in-person meetings or worse. All they have to do is mimic the typical profile of their targets. There is no verification of truth in others, but there is a lot of personal exposure of yourself.

3. It was strange to see people’s likes and dislikes as a profile. It gives you a lot to potentially dislike in a person, but nothing really to draw you to them as a person. In a community, the more preferences one lists, the more odds they have of alienating their neighbors, instead of bringing them together. Labels and stereotypes have one thing in common - they pidgeonhole people instead of giving them personality.

You have shown you can execute. The above is just feedback on the inherent intended or unintended consequences behind this new funnel for mixing people. Social products can be quite dangerous but also fun. It’s critical to reduce downside risk.

[pitherandd]

This is very good insight and I definitely agree. With regard to your first point, I'll look to add a limit to how many users a person can message per day.

I spent some time already and plan to spend a lot more on moderation tools to mitigate the effects of your second point. Ideally fake profiles can be both reported and detected, but as of yet I have not invested too much time into these efforts, other than adding basic reporting features and a moderation page for those with sufficient power.

I tried to mitigate your third point by only allowing users to see your total cluster vote as opposed to your individual card votes. As a result, there is a bit of plausible deniability as your vote for each card is somewhat masked by the average. This could be furthered by reducing the cluster count and thereby making each cluster larger and mask more individual card votes.

[smichel17]

This is a total thought experiment and I have not thought deeply about it.

I wonder if you could encourage healthy debates by grouping people based on cluster, such that:

- Eveybody agrees with each other person in the group on the majority of topics.

- Everybody disagrees with the majority opinion on some topic(s).

- Every topic has two people who disagree with the majority.

- No two people share the exact same views.

I'm imagining probably 6-7 people per group. Make sure they know the matching rules.

Basically, I wonder if you could cut through tribalism and encourage nuanced discussion by making it clear from the beginning that everyone mostly agrees on everything, but everyone also has some points of disagreement. Then it becomes more of a "figure out where we disagree and why" type discussion. Maybe seed it with one topic in the center of the cluster that everyone agrees on, to give a "safe" starting talking point.

I could also see it devolving to a conformist atmostphere where nobody wants to voice their minority view, but I'm hoping that would be mitigated by everyone knowing that they all have a minority view, and that one other person shares their particular minority views.

[dzink]

On second thought: 4. I would go for specific leading topics for a person and leave clustering for the rest. Clustering removes the fine granularity you spend so much time voting on in the beginning. The cards also rely on gut instinct plus fat fingers - i found myself mis-swiping on several and didn’t really care to go back and fix them. Instead, you should allow statements of explanation behind the most pertinent topics or highlighting the most relevant topics after one is done swiping and then prioritize matching on those, with the rest lumped into clusters.

5. The topics do conflate two different mental systems and they clash badly: Interests are things you want to pursue more of together. Politics are divisive topics. You and I share an interest in building products that help people with niche interests find each other, we should be able to find each other with your product. Instead, we found each other with the HN discussion. That’s because depth and insights surfaces with content generated by the person, while swiping is low threshold but also too low signal and the gut instinct of one strongly opposing swipe would be to “skip” that person. If you want connection, let users focus on a few key topics to them and let them pick a few they definitely want to avoid in another person, then let people reach out based on your statements of purpose on a topic. If you want to dig into these problems over chat, call, or email, feel free to reach out to me at hnusername at dreamlist dot com.

5 The down side with people matching is the reason why there are no people marching IPOs out there. Once you are successful, you have to leave. The dating sites have all aggregated into one big company which is a cash cow. Don’t converge towards dating unless you want acquired and users churning massively. The interest matching is slightly different, in that a good match doesn’t preclude you from staying on the platform. The politics ones always turn into a dumpster fire, as political preferences can be fleeting based on your in-group. I would stick to interests for things you want to do together.

[codetrotter]

> I tried to mitigate your third point by only allowing users to see your total cluster vote as opposed to your individual card votes

I find these clusters really strange. For example, this cluster:

> War on Drugs, War on Terror, Fossil Fuels, Drone Strikes, Dark Humor, Military, Police Officers

And

> Yoga, Lower Drinking Age, Gym, Alcohol, TikTok

What does Yoga and Gym have to do with lowering drinking age, or with TikTok? Very weird. So if I create a profile and I am positive towards Yoga and Gym then people are going to think that I want to lower the drinking age? And likewise, they are going to think that just because I support having a military force (which I do), that I also support the war on drugs (which I don’t).

Would be better to not do clustering at all than to group a whole lot of unrelated stuff together and give the wrong impression I think.

[pitherandd]

So, the clustering situation is a bit of a double-edged sword. Here's the pros and cons I've found:

Pros:

1) Allows for instant similarity-searching. I am using the Postgres CUBE data structure to index users' votes and it has a limit of 100 items. So, if I didn't use clusters then there could only be 100 cards maximum, but ideally you would have even less than that because the CUBE can start to slow down when you approach that limit.

2) It's also a bit of a privacy feature as people can only see how you voted along cluster lines, and not how you voted on individual cards. This provides the aforementioned plausible deniability.

Cons:

1) Not all clusters are ideal, as you've seen. I spent a lot of time exploring different clustering algorithms and none of them were perfect. Some cards were naturally a part of multiple clusters and others didn't align to any at all. I'm sure a lot of this comes down to card choice, which I definitely could improve.

2) Can be confusing to users as opposed to just listing theirs and others votes on cards.

---

If you'd like to help and create better clusters, I'd definitely be open to tweaking them. Most of the required data can be found by navigating to the Cards page. For example, if you go there and click "War on Drugs" and then click "View Correlations", you'll find that "War on Terror" correlates the most with "War on Drugs". This data can then be used to try and create your own clusters. I've found it to be a very tricky puzzle satisfying all the constraints.

In the end, for performance reasons, I felt like I had to choose between either having clusters or only 50-75 cards, and I chose clusters. There's probably a better way of doing it, but I was unable to find it at the time.

[noxToken]

Have you manually reviewed some of these clusters? Automatic clustering for subjective topics tend get gamed by users (if they can be influenced by users) or labeled incorrectly.

[pitherandd]

Yeah, the first pass for the clustering was SciKit's Agglomerative Hierarchical Clustering Function and then the clusters were manually tweaked in an attempt to increase accuracy. Clusters are also manually updated and curated, so hopefully they shouldn't be able to be gamed at the moment.

[matt_f]

Upvoted: People who take time from their day to give positive constructive feedback and encouragement to a total stranger about his/her project are what make this place a good community.