Three levels of comfortable communication with strangers online

 

Tablo reader up chevron

 Three levels of comfortable communication with strangers online

When you are already an adult, it can be quite difficult to meet new people, let alone find new friends at Xossipz. At the same time, the need to be heard, to know that you are not alone, to share your opinion and discuss it with someone who understands this topic and is close to you remains with many. And this does not depend on marital status, the presence of children, a successful career or not, and other factors.

Many services are trying to close this need. There are 3 main types:

  • classic dating (goal - romantic dating, sex, etc.)
  • services about "find new friends" (the goal is communication of interests, friendly meetings, etc.)
  • services about networking and expansion of professional connections

Dating, from the point of view of the product and the interaction of people, is very interesting, but, unfortunately, they are quite boring from the point of view of the application of machine learning and behavioral factors - they have a very strong and banal feature that drags on all the optimization - a photo. We will make a separate post about dating.

We undertook to solve the problem of "new friends". Here, the value of the photo, although it plays some role, is already influencing a large number of other behavioral factors. And in the course of many experiments over the past year, we have come to 3 levels of comfort for the interaction of strangers online.

Level 1. List of people. The most uncomfortable.

What most friends search services do - look, here are the people, for some reason they must be interesting to you, write to them.

And writing to a stranger is very difficult - what to write? What if he doesn't answer? And if he answers, and the conversation goes to a dead end? And so on.

When the Aura just started, we also unknowingly started with this - let's make embeddings based on the behavioral characteristics of each user and get an n-dimensional user space, where there will be people nearby who should definitely be interesting to each other? Easily! The only problem is that friends and sincere communication, it turned out, does not start.

Rather, it turned out to be a kind of social exploration - we made a separate People section and put people nearby by geolocation, people with similar interests based on ML, people who rated posts and comments in it. And these sections can even be filtered by gender, age, and community that the user is a member of. It's interesting to watch, but the very first writing to a stranger is still uncomfortable.

Level 2. Do everything for me.

Here we have already decided to take everything into our own hands and unite people in semi-automatic and automatic modes. SmartMatching, Automated Communities and the level of similarity with each user appeared.

SmartMatching is such a Tinder mechanic: you choose a target for what a person is for you (relationships, friendships or online communication), then the algorithm selects and shows you people one at a time, and you say OK or not OK. And if both said OK to each other, a notification is sent and the chat automatically opens.

Here, communication is already more comfortable - both know that the other also said OK, but the conversation still needs to be built and invented by yourself. In the beginning, we even wrote the first 2 messages for users, so that they had something to discuss, but then it had to be turned off due to the polar perception of this function.

We did not stop there and did Automatic Communities - you just need to activate this function and 2 times a week the algorithm selects people of interest to you and invites you all into 1 post for communication, which is seen only by members of this community. Moreover, he suggests what topic to discuss, and this topic is different each time. This method is also more comfortable, but the most frequent discussions under such posts still boil down to figuring out why the algorithm brought us together.

The level of similarity is an auxiliary indicator that can be found for any user - it shows how much your interests coincide with this user according to your behavior in the Aura. Also embeddings, n-dimensional spaces, etc.

Level 3. The most comfortable and effective. Posts and comments.

It turned out that if you let users write their message as a post and show it to strangers with the ability to comment, plus add ML for the correct selection of posts, real magic arises.

So we came to our massive scenario of communication and acquaintance of strangers. The result is such an open ecosystem where everyone can be heard regardless of whether they have subscribers or how many. And it is in this ecosystem that communication is very comfortable - both from the side of the author of the post and from the side of the commentators. And a big plus - you recognize a person not only from a photo or blindly trust the algorithm, you see what a person writes and how he communicates with others.

There are already many stories when users found real friends, met offline and even found relationships (it is much more reliable to pick a couple from photos, posts and comments than just from a photo).

But this method also poses a lot of challenges - spammers, insults, inappropriate content, etc. In order to combat this, we had to build a complex system, both algorithmic and manual moderation.

Also, to improve quality and safety, we ask everyone to confirm their phone number and set a real photo. It turned out that if you have your photo on your avatar, you are less likely to swear or insult someone in the comments.

People are unique in their behavior. But we sincerely want to find an inspiring social circle for everyone. Chat, meet new people - support and human relationships are valuable at all times.

Comment Log in or Join Tablo to comment on this chapter...
~

You might like Marcelo's other books...