A simple bayesian spam classifier.
Bayespam is inspired by Naive Bayes classifiers, a popular statistical technique of e-mail filtering.
Here, the message to be identified is cut into simple words, also called tokens. That are compared to all the corpus of messages (spam or not), to determine the frequency of different tokens in both categories.
A probabilistic formula is used to calculate the probability that the message is a spam. When the probability is high enough, the classifier categorizes the message as likely a spam, otherwise as likely a ham. The probability threshold is fixed at 0.8 by default.
Learn more about Bayespam here: https://docs.rs/bayespam.
Add to your Cargo.toml
manifest:
[dependencies]
bayespam = "1.1.0"
Add a model.json
file to your package root.
Then, you can use it to score and identify messages:
extern crate bayespam;
use bayespam::classifier;
fn main() -> Result<(), std::io::Error> {
// Identify a typical spam message
let spam = "Lose up to 19% weight. Special promotion on our new weightloss.";
let score = classifier::score(spam)?;
let is_spam = classifier::identify(spam)?;
println!("{:.4?}", score);
println!("{:?}", is_spam);
// Identify a typical ham message
let ham = "Hi Bob, can you send me your machine learning homework?";
let score = classifier::score(ham)?;
let is_spam = classifier::identify(ham)?;
println!("{:.4?}", score);
println!("{:?}", is_spam);
Ok(())
}
$> cargo run
0.9993
true
0.6311
false
You can train a new model from scratch, save it as JSON to reload it later:
extern crate bayespam;
use bayespam::classifier::Classifier;
use std::fs::File;
fn main() -> Result<(), std::io::Error> {
// Create a new classifier with an empty model
let mut classifier = Classifier::new();
// Train the classifier with a new spam example
let spam = "Don't forget our special promotion: -30% on men shoes, only today!";
classifier.train_spam(spam);
// Train the classifier with a new ham example
let ham = "Hi Bob, don't forget our meeting today at 4pm.";
classifier.train_ham(ham);
// Identify a typical spam message
let spam = "Lose up to 19% weight. Special promotion on our new weightloss.";
let score = classifier.score(spam);
let is_spam = classifier.identify(spam);
println!("{:.4}", score);
println!("{}", is_spam);
// Identify a typical ham message
let ham = "Hi Bob, can you send me your machine learning homework?";
let score = classifier.score(ham);
let is_spam = classifier.identify(ham);
println!("{:.4}", score);
println!("{}", is_spam);
// Serialize the model and save it as JSON into a file
let mut file = File::create("my_super_model.json")?;
classifier.save(&mut file, false)?;
Ok(())
}
$> cargo run
0.9999
true
0.0001
false
$> cat my_super_model.json
{"token_table":{"forget":{"ham":1,"spam":1},"only":{"ham":0,"spam":1},"meeting":{"ham":1,"spam":0},"our":{"ham":1,"spam":1},"dont":{"ham":1,"spam":1},"bob":{"ham":1,"spam":0},"men":{"ham":0,"spam":1},"today":{"ham":1,"spam":1},"shoes":{"ham":0,"spam":1},"special":{"ham":0,"spam":1},"promotion:":{"ham":0,"spam":1}}}
Contributions via issues or pull requests are appreciated.
Bayespam is distributed under the terms of the MIT License.