Here is an interesting project I worked on that used DNS lookup data to determine anomalous traffic, specifically bots that and other bot networks that may cause DDOS attacks.
It used a clustering technique that would build profiles of behavior that deviate from the normal. This also allowed for potentially checking for a safety score before interacting with a network IP.
This has to be one of the oddest applications of machine learning that I have used on a dataset.
Domain names are a huge industry and low character domain names, especially in the .COM space are expensive and most of the ones that are English words or mean anything in any language are long gone.
The idea here is that there are still small words that don’t mean anything but are pronounceable. In this particular implementation the solution was to use a phonetic model to analyze what a pronounceable name looks like. Once the phonetics of a word are understood it is all about scoring combinations of letter and determining which ones have the highest probability of being pronounceable.
One example is “smilyr” , which is pronounceable but doesn’t mean anything. It also got picked up on Domain Name Wire here
The web is full of summarization utilities, most free and some paid , but most of them either aren’t maintained after a few months or they require a backend API that is almost always very slow.
I skim a lot of articles on the web and after trying out a lot of summarizers on the web I decided to write my own dirty version. Here are the deets
- Pure JS only, does not use a backend api, so is pretty quick
- Does entity extraction ( kind of ) , so one can read summaries by topic
- Use an article extraction algorithm based on a modified version of TFIDF
If you run any kind of service on the web , one would have spent a lot of time tuning their landing page obessively. However one often neglected area is the support page or a FAQ section. This is where you address common concerns, questions or objections that customers have.
As with many other product experiments, this was a pain point for me, due the lack of a free form FAQ builder on the web. Some of the things I was trying to solve include
- Simple : super simple FAQ builder
- Easy : Maintain your FAQ with zero code / IT involvement.
- Solve issues : Build support workflows for common issues across your system.
- Insights driven : Learn and discover user funnels and pain points
- Personalization first : Surface your most requested answers and solutions automatically
Here is a quick example that shows a simple FAQ built using a WYSIWIG style editor . If you find this useful, drop me a note and I’d be happy to set you up with an account.