danielrapp / twss.js

A node.js "that's what she said" classifier

This is a node.js module that classifies if a sentence can be replied with "that's what she said".

Installation

npm install twss

Usage

var twss = require('twss');

twss.is("Nice weather we're having today"); // false
twss.is("Can you make it harder?");         // true
twss.is("You're not going fast enough!");   // true

Settings

twss.algo

You change algorithm from the default naive bayes classifier (nbc) to a k-nearest neighbor algorithm (knn).

twss.algo = 'nbc';
twss.algo = 'knn';

twss.threshold

If you want more obscure jokes to be accepted, you can set the "probability the sentence can be replied with twss" threshold. Be aware that a too low threshold may result in a lot of false-positives, and a too high threshold may result in a lot of false-negatives.

twss.threshold = 0.5;
twss.is("You're hardly my first."); // false

twss.threshold = 0.3;
twss.is("You're hardly my first."); // true

Additional functions

twss.probability

If you'd just like the probability that a sentence can be replied with "That's what she said" you can use the twss.probability function.

twss.probability("Behold, I come quickly."); // 0.956323045469951

Or just use the alias twss.prob

twss.prob("The juice keeps coming out of the wrong hole!") // 0.9961630818418142

twss.js's Issues

Short sentences biased

Short sentences trigger relentlessly. I.e. "oh wow lol", "oh god", "oh what", "oh why", "oh, right.", "always good", "no one does", "much better", "all my want", "god damn", "OH", "oh my", "I'm in awe", "i guess", "it isn't", "ok", "aint hard" and for some reason "I will never program in java"

CLI utility

adding the possibiliy to use twss in a terminal environment could be interesting and would be easier to integrate with other programs.
I'm willing to send a pull request if I can find some time to do it.

Normal sentence data is biased

The data for normal sentences (collected from fmylife.com) is biased towards the word "was" (and probably a lot of other things). A good resource for new normal sentences may be wikipedia.

Training data is too generic, lots of false positives

Training data (at least) needs work - even at 90% threshold, I get false positives on simple non-twss data:

twss.is("hello there, is this car still available?");
>> true

Add support for polling github issues

• poll github issues
• on new issue:
  • calculate twss-score
  • if score > limit, reply accordingly (cool message needed!)
• see how long it takes to be banned

k-NN doesn't work at all

It just returns this error:

ReferenceError: trainingPrompt is not defined
    at Object.isTwss (/Users/Daniel/twss.js/lib/classifier/knn.js:35:7)
    at Object.is (/Users/Daniel/twss.js/lib/twss.js:49:33)
    at [object Context]:1:3
    at Interface.<anonymous> (repl.js:171:22)
    at Interface.emit (events.js:64:17)
    at Interface._onLine (readline.js:153:10)
    at Interface._line (readline.js:408:8)
    at Interface._ttyWrite (readline.js:585:14)
    at ReadStream.<anonymous> (readline.js:73:12)
    at ReadStream.emit (events.js:81:20)

Missing OR operator in punctuation regex

https://github.com/DanielRapp/twss.js/blob/master/lib/document.js#L4

Missing the | Between the double quote and the single quote characters.

Crowdsourcing site

This might be useful:

• shows latest tweets with very high and very low score
• users can click +/-
• stuff with many votes into one direction and nearly none into the other comes into the rating DB

As it would also provide quite some entertainment, I guess you'd get enough users for this. :)