Hacker News new | past | comments | ask | show | jobs | submit login

Change of subject: I'm very interested in Bayesian modelling. Would you be as kind as to provide me with some resources to learn the ins and outs of this?

Also, I feel like akinator.com is using a Bayesian tree. Would you confirm this? What algorithm do you think they are using?




Would you be as kind as to provide me with some resources to learn the ins and outs of this?

I thought PG's explanation in "A Plan for Spam" was pretty good: http://www.paulgraham.com/naivebayes.html

Also, I feel like akinator.com is using a Bayesian tree. Would you confirm this?

It's possible, but I doubt it.

What algorithm do you think they are using?

A plain binary search tree would work perfectly. I guess it could be broken by people entering deliberately wrong data, but..

It's clearly some kind of search tree, because the questions are influenced by previous answers.

You could do a Bayesian version, but I'm not sure it would perform any better.


Thank you a lot :) I'm further bombarding you with questions:

I'm trying to list the usages of Bayesian inferences in industry. So, there is spam filtering, I bet there would be recommandation engines (ex: amazon's product recommandation, ad targeting). Would you know of other uses?

Do you know of some kind of tutorial of how one particular bayesian engine is applied to a particular dataset in a commercial venture?

Also, I have been looking at weka (http://www.cs.waikato.ac.nz/ml/weka/), and j48 in particular. j48 is an implementation of the C4.5 algorithm, (which is close to ID3 and C5). Here is how I learned about C4.5: http://www.cs.umd.edu/~samir/498/10Algorithms-08.pdf. Would you happen to know if one of these algorithms have commercial applications?

In a different subject, Judea Pearl has been awarded the Turing award for using Bayesian networks to calculate probability of causality. Which is different than traditional Bayesian analysis, where we calculate probabilities of a property being true. In his theory, he claims one can calculate the probability of a property being a necessary condition, the probability of a sufficient condition to another property (and the probability of both combined). I'm really having trouble understanding the ins and outs of this too. Have you heard about an usage of this theory? Do you think it has a viable shot at having commercial applications?


It appears you are looking for Deep Insight into field x where x = "applications of Bayesian filtering". This is not a reliable way to do this (random posts on non specialist forums). A more reliable path is to study the topic exhaustively over n years where n > 1. Take classes, or do a ph.d, or find a good academic library and read the related textbooks and journals. Thus developing expertise, apply imagination => profit!


We use it on our phishing filter for our form builder app. We have a tool which we use to review forms created in our site and teach it if a form is a phishing form or not. It works so well that we trust it to auto-suspend phishing accounts. Number of false positives is less than 1%.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: