Has anybody considered adding keyword (or personal keyword) tagging based on ML? I mean a simple multiple Bayes filter would be really useful.
Has anybody considered adding keyword (or personal keyword) tagging based on ML? I mean a simple multiple Bayes filter would be really useful.
Do you mean a story code tag, labeling a story as focusing on "Machine Learning"? If so, I'm not sure how wide-spread a need it would address, as much of that is covered under "AI" (i.e. there aren't too many stories about the creation of better algorithms).
In either case, your best bet is to ask Lazaar himself, presenting your reasons for including it (aside from your one story). If he thinks there's a demand for that type of story, he'll make that decision. That's how many of the current story codes were added.
However, if you're talking about an SOL algorithm to filter story results for particular individuals, you may have a harder time convincing the Admins, as SOL users are notorious reluctant to have their reading 'monitored', packaged and sold to the highest bigger, as happens on many other sites.
That's another thing that Lazeez is best set to decide as to what best fits the site and its users.
If I read his post correctly, he is asking Lazeez to implement a different search technique using Bayes filters.
I had to look up what they are and it appears to be a way of analyzing a complex database. I didn't understand what I read, so I may have missed what he is asking Lazeez to do.
It sounds to me like he wants to be able to use a key word that he assigns to a story in order to search for a listing of the stories that have his key word.
If I read his post correctly, he is asking Lazeez to implement a different search technique using Bayes filters.
I had to look up what they are and it appears to be a way of analyzing a complex database. I didn't understand what I read, so I may have missed what he is asking Lazeez to do.
It sounds to me like he wants to be able to use a key word that he assigns to a story in order to search for a listing of the stories that have his key word.
That's largely what I assumed, without looking up the particular algorithm. The point, though, is he's asking for something much more complex than simply adding a tag, he's asking Lazeez to completely rewrite the tagging and sorting system so it operates like Google's or Amazons (with all of their abuses). There's also a fear, among many of us old-timers here, that that's the first step in treating us readers as commodities, and to sell our 'reading' data to whoever asks as a means of monetizing the site. As they say, once you build the infrastructure, hackers will figure out a way to take advantage of it.
That's largely why Lazeez hasn't done a lot of things here, because he's very conscious of his reader's distrust of other sites, especially since a database of readers' reading habits would reveal an entire string of potential sexual peccadillos.
Machine Learning typically refers to taking a corpus and throwing it at an algorithm to train it to make determinations about new things.
In other words, the proposal sounds like "looking at the already-existing corpus of stories and their tags, train an algorithm to evaluate incoming stories and suggest tags". Which, based on my experience with Machine Learning, feels like a mighty stretch for any sort of reasonable answer.
Go back and read what he said and parse his statement:
Has anybody considered adding keyword (or personal keyword) tagging
We have an existing search algorithm that allows searches on keywords or phrases that exist within stories. Personal keyword implies a reader could add a keyword and link it to a specific story. Personally, I see no gain to that new capability and it would require an immense amount of work on Lazeez's part.
Has anybody considered adding keyword (or personal keyword) tagging based on ML?
https://en.wikipedia.org/wiki/Recursive_Bayesian_estimation
I'm having a real problem with keywords and training a computer to do something on this website. The only rationale on the site for keywords is a search function. Is he asking for something that would learn to do better search routines?
I mean a simple multiple Bayes filter would be really useful.
I would have to ask, useful in what way?
Well Bison9, please explain how your idea would make things better on the site.