I knew that this was possible but I always thought it took much more... effort? How do we mitigate this, then? Run our posts through an LLM?
Reproducing Hacker News writing style fingerprinting
151–160 of 168 posts
Re: Reproducing Hacker News writing style fingerprinting
#152That makes me wonder two things.
Firstly, if your can use this to find LLM generated content, which I guess would need similar instructions. Imagine instructing it to talk like a pirate, it would be quite different from a generic response.
Secondly, if you want to make an alt account harder to cross-correlate with your main, would rewriting your comments with an LLM work against this method? And if so, how well?
Re: Reproducing Hacker News writing style fingerprinting
#153Earlier quoted context omitted.
Should is a commonly used word and a fine one. You should feel free to use it. If someone gets hot under the collar because you said he should do something then he is an idiot. "Ought to" is essentially a synonym. Anyone that gets upset when you said they should do something but is fine when you say that they ought to do something is truly a moron.
> If someone [test] then he is an idiot. > Anyone [test] when [test] is truly a moron. These structures are worse habits in communication than subtle, colloquially interchangeable word choices.
This isn't a habit of communication. I honestly mean it: if you get upset that someone said you "should" do something, but you are fine when they say you "ought to" do it, then you must be stupid. They mean the same thing in modern English.
Re: Reproducing Hacker News writing style fingerprinting
#154Earlier quoted context omitted.
Should is a commonly used word and a fine one. You should feel free to use it. If someone gets hot under the collar because you said he should do something then he is an idiot. "Ought to" is essentially a synonym. Anyone that gets upset when you said they should do something but is fine when you say that they ought to do something is truly a moron.
The only time to avoid command words like should is when the person could conceivably see them as a command. Because then you're being a dick. Otherwise, if someone wants to take the time to dissect meaning from add-on meaningless words like should in a sentence, they should find something better to do with their time. Or just ask instead of being a moron.
How are you being a dick?! There are loads of reasons why you may want or need to instruct someone to do something. I prefer the imperative mood. It is more direct. "Sudo make me a cup of tea".
Re: Reproducing Hacker News writing style fingerprinting
#155We can improve this. antirez has made a highly compelling poc but it could be refined for authorship attribution judging by the number of misses in the comments here, and how this compares to greater accuracy of the original post to which antirez refers. I’m no expert, but some ideas:
- remove super high frequency non specific words from the comparison bags, because they don’t distinguish much, have less semantic value and may skew the data
- remove stop words (NLP definition of stop words)
- perform stemming/tokenization/depluralization etc (again, NLP standard)
- implement commutativity and transitivity in the similarity function
- consider words as hyperlinks to the sets of people who use them often enough, and do something Pageranky to refine similarity
- consider word bigrams, etc
- weight variations and misspellings higher as distinguishing signals
What are your ideas ?
Re: Reproducing Hacker News writing style fingerprinting
#156Earlier quoted context omitted.
> Probably, we make the same type of mistakes while using English. That is most likely the case. Case in point: My native language doesn't have articles, so locally they're a common source of mistakes in English.
It would be fun to have a tool try guess your native language, based on your English writing.
I noticed that it also depends on the vendor of the autocorrect/dictionary you're using.
The project referenced in the post put me next to Brits on the similarity list and indeed I am using an English(UK) dictionary. Meanwhile this iteration aligns me with Americans despite the only change being the vendor (formerly Samsung, now Google).
I guess the Samsung keyboard corrects to proper Bri'ish.
I picked up the language as a child from a collection of people, half of whom weren't native speakers, so I don't speak any specific dialect.
Re: Reproducing Hacker News writing style fingerprinting
#157I wonder how much accuracy would be improved if expanding from single words to the most common pairs or n-tuples. You would need more computation to hash, but I bet adding frequency of the top 50 word-pairs and top 20 most common 3-tuples would be a strong signal. ( The nothing the accuracy is already good of course. I am indeed user eterm. I think I've said on this account or that one before that I don't sync passwo…
Maybe there isn't enough data for each user for pairs, but I thought about mixing the two approaches (but had no time to do it), that is, to have 350 components like now, for the single word frequency, plus other 350 for the most common pairs frequency. In this way part of the vector would remain a high enough signal even for users with comparable less data.
I've been thinking some more about this, and it occurred to me that you'd want to encode sentence boundaries as a pseudo-word in the n-tuples.
I then realised that "[period] " would likely dominate most common pairs, and that a lot of time could be saved by simply recording the first word of sentences as their own vector set, in addition but separate to the regular word vector.
Whether this would be a stronger or weaker signal per-vector-space than the tail of words in the regular common-words vector I don't know.
Re: Reproducing Hacker News writing style fingerprinting
#158> Well, the first problem I had, in order to do something like that, was to find an archive with Hacker News comments. Luckily there was one with apparently everything posted on HN from the start to 2023, for a huge 10GB of total data. This is actually super easy. The data is available in BigQuery.[0] It's up to date, too. I tried the following query, and the latest comment was from yesterday. SELECT id, text, `by` A…
…i can’t believe i’ve been running a script to ingest the data for the last six hours. thank you.
Re: Reproducing Hacker News writing style fingerprinting
#159Such a nice scientific way to detect and mute those who go against the agenda's grain, oh I mean don't contribute anything meaningful to the community
Re: Reproducing Hacker News writing style fingerprinting
#160The "analyze" feature works pretty well. My comments underindex on "this" - because I have drilled into my communication style never to use pronouns without clear one-word antecedents, meaning I use "this" less frequently that I would otherwise. They also underindex on "should" - a word I have drilled OUT of my communication style, since it is judgy and triggers a defensive reaction in others when used. (If required,…
Now if you only underindex on "underindex"... There's a good alternative that everyone understands, "use less"
If we weren't on HN, I might be inclined to agree