Update to the Note Like Aggregation Tool
[Note] I have updated the Note Like aggregation tool.
I have added several features and improvements to the "Note Like aggregation tool" that I released previously.
Here is an introduction to the contents of this update.
■ New Feature: Find "Common Like" users
This is the newly added feature.
By entering two Note users, you can now display "common reader users who have liked both accounts."
This feature answers the question, "What kind of readers do I share with this creator?"
Based on how many times they have liked each of the 12 most recent articles of the two users being searched, the match rate is calculated with a maximum of 24 likes as 100%. They are listed in descending order of match rate, so you can see at a glance who the particularly enthusiastic common readers are.
■ Convenient Integration: One-click common like search from "Like Aggregation" results
A "Find Common Likes" button has been added to the existing "Like Aggregation" result list.
When you click this, the user you are currently searching for is automatically set as User A, and the user whose button you clicked is set as User B, immediately starting the common like search.
When you are curious about "how much do my reader demographics overlap with this person?", you can check it without manual input.
■ Performance Improvement: GAS caching feature
I have added a caching feature to the backend Google Apps Script.
The results of a search are held on the server side for a certain period, and if a request for the same user ID comes in, it returns the result from the cache without hitting the API.
As a result, for the second search onwards, the results return in about 1 to 2 seconds.
This only supports the search pattern for the 12 most recent articles. This is because the cache capacity on the GAS side is too small, so large data cannot be cached in the first place, and the benefits are not that significant. Well, please think of this as a minor feature addition.
■ AI Development
Well, I am currently studying AI for work, and as a practice exercise, I tried having an AI build "everything" from scratch. Since I went to the trouble, I'm releasing the feature lol
I had AI generate ideas from the initial concept. I asked it to come up with about five patterns of interesting feature ideas, and I adopted the plan I liked best from among them.
I then had it build that exactly as planned.
I didn't do any programming myself, leaving everything to AI as much as possible. I only provided instructions in Japanese. It took 5 minutes for the concept and maybe 15 minutes for the build? (This doesn't include the time I spent agonizing over it, lol—that actually took the longest.) From setting up the environment to pushing a pull request to GitHub, there was nothing for me to do.
The only work I did was probably deploying it on the GAS side. Aside from giving instructions, I just clicked buttons to approve or say OK.
Oh, I did do the testing myself, but it feels like it could even automate testing if I gave it browser permissions. (I'm too scared to grant browser control permissions, so I haven't tried that yet.)
I didn't like the design, so I spent the most time being particular about giving various instructions to improve the level of completion, and it ended up taking two or three hours to finish.
Can you really get this far without moving your hands?
You still need to understand some technical terms; you can't achieve full automation just by pressing buttons "without thinking." But it feels like we're almost there.
By the way, this article was also written in that flow; I asked it to turn the work I did this time into a note article, and it wrote almost all of it. (I added this chapter and the images later.)
■ Links
You can use the Like aggregation tool here.
https://licodeenar.github.io/note-likes/
The common Like search tool is here.
https://licodeenar.github.io/note-likes/compare.html
Please give it a try!