domhel.bsky.social
Software Engineer, Student in M.Sc. Autonomous Systems, Building multiple Start-Ups
-> https://stallfreunde.app - Digital Horse Stable, 100+ active users
-> https://worldbummlr.com - AI travel assistant
44 posts
7 followers
2 following
Regular Contributor
Active Commenter
comment in response to
post
This is because I am currently in step 3. I have deployed it to my apps Worldbummlr and Stallfreunde and am collecting data.
Once I'm done, if you are interested, I can make a follow-up post with the results with analyzing the data.
worldbummlr.com
stallfreunde.app
comment in response to
post
Thanks for reading, happy to hear your thoughts on this.
You might be wondering why the steps are not detailed enough.
comment in response to
post
9. Deploy and redo, hopefully the numbers are better now. You can also include A/B testing to test different click paths.
comment in response to
post
6. Find the chains with the highest number of transitions until reaching the destination page.
7. Check the average time spent in the transition pages.
8. Optimize your app to reduce the number of transitions and time spent on transition pages.
comment in response to
post
so you can limit yourself to a Markov Chain of n-th order, where n is the number of previous states being used).
5. Find destination pages where diagonal elements have the maximum probability for the state transitions.
comment in response to
post
Row 1, column 2 is the transition probability from state 2 to state 1 (e.g page B to page A). Row 3, column 4 is the transition probability from state 4 to state 3, and so on.
* 4.3. Calculate all other tables in dependence of all previous states (this might be a lot,
comment in response to
post
* 4.2. Calculate the first state transition table from each state to a new state. It will be a table with {number of states} rows and {number of states} columns. The diagonal elements are the state transitions to the same page.
comment in response to
post
4. Calculate the state transition tables.
* 4.1. For each session uuid, save the initial state = the first page.
comment in response to
post
3. Capture enough data (I guess ~100 data points per page are sufficient).
comment in response to
post
2. On every page navigation, log the page name, timestamp and session uuid to your database. If it's a meaningful interaction with the page, log the same page again (to detect destination pages).
comment in response to
post
1. When the app starts and when the app is returned from inactive to foreground, generate a session uuid. This way we can map click paths anonymously and GDPR compliant.
comment in response to
post
or not visible enough.
And the last insight is that if we have multiple paths leading to the same page, we can find out which path is preferred by the user and we can focus on that one.
Okay, enough intro. How is it done?
It's more simple than it sounds. Let me lay out the steps:
comment in response to
post
However, this is not to be confused with "information pages" where the user simply looks at some info and navigates back, which would make the page still relevant.
Another insight is that if the probability to reach a certain page is really low, then this page might be irrelevant for our app,
comment in response to
post
but at as fast as possible.
A third useful insight we get from this is the number of false navigation. If the user navigates from A to B, and the next most likely action is to go back to A right after that, then we know that the user misclicked and our navigation button might be labeled poorly.
comment in response to
post
Therefore we minimize the number of state transitions.
Additionally, if we cannot get the number of state transitions to 1, we need to minimize the time the user stays on each state that is not the destination state. In practice, this means that the user has to navigate through the app,
comment in response to
post
If the probability to stay on page X is higher than to navigate to a different page, then X is considered as the destination page, the "useful page".
We want to optimize our app so that the user reaches the destination page with as little navigation as possible.
comment in response to
post
"destination page". This is a helpful metric if people open your app to perform a certain task. The click path could be A -> B -> C -> C. In our algorithm we will be able to detect destination pages by the transition probability of page X with itself, so X -> X.
comment in response to
post
50% of the users might go from A to B, 35% from A to C and 10% from A to D and 5% stay on A.
My idea is the following: we first collect lots of data on how users navigate through the app. If the user performs an action on a page that does not lead to a new page, it is considered the
comment in response to
post
your users through the app.
Basic example: user opens the app and is now in the initial state, page A. Let's say the user can navigate to three different pages B, C and D. The state transition probability describes how likely it is to transition from page A to page B, C, D, or A again. For example,
comment in response to
post
then close it again, but can also be used otherwise.
I recently learned about Markov Chains. Basically, they describe the probabilities of a state transition depending on a number of previous states. In the context of Flutter and multi page apps in general, they describe the click path of
comment in response to
post
Spirit of the Samurai (Dec 12) - Stop-motion hack-n-slash Metroidvania with brutal combat
Also check out Antonblast (Dec 3) - Wario-like platformer where you demolish cities to fight Satan
Both match your Ori/Doom/Metroid vibes perfectly! 🔥
comment in response to
post
this will become huge in the future, especially with VR. If you start a VR-gamify-learn company now and combine it with AI, the chances of you greatly benefiting humanity are high and you'll get paid for it
comment in response to
post
do I need to know the Bluetooth PHY if I build a mobile app with Bluetooth?
comment in response to
post
Does this count as a game? marketmelody.net
comment in response to
post
what exactly do you consider robotic foundation models?
comment in response to
post
I'm also looking around, but it doesn't seem to be the case. Feels like it's mostly politics and other boring stuff here
comment in response to
post
this is really cool
comment in response to
post
yes same just happened to me :D
comment in response to
post
I'm glad it's not, Flutter is a framework for developing apps ^^