Famous sorting algorithms based on vote popularity and time


This library houses 3 popularity estimating algorithms employed by bigger news sites used to sort for best content:

  1. wilsonScore - Reddit's best comment scoring system
  2. redditHot - Reddit's hot post scoring system for news posts
  3. hackerHot - Hackernews' scoring system

Algorithms may cause scores to decay based on distance to post time.

Algorithms that are designed to decay based on time needs continual recomputation of scores. An example of doing so would be keeping track of, and periodically computing the score(s) required in a node process on a set of suitable candidates:

var decay = require('decay')
  , hotScore = decay.redditHot();
setInterval(function () {
  candidates = []; // perhaps get recent posts saved in db here 
  candidates.forEach(function (c) {
    c.score = hotScore(c.upVotes, c.dnVotes,;
    // save so that next GET /entry/ gets an updated ordering 
}, 1000 * 60 * 5); // run every 5 minutes, say 

Algorithms that produce a time agnostic popularity score is typically good for comments. For best results, simply recompute the score at every new vote:

var decay = require('decay')
  , wilsonScore = decay.wilsonScore();
// assume req.entry is the item being voted on'/entry/upvote', middleWare, function (reqres) {
  // call wilsonScore with ups, downs, post_date to recompute 
  req.entry.score = wilsonScore(req.entry.upVotes + 1, req.entry.dnVotes, req.entry.postDate);
  // save new score in database so that new pageviews sort  

Decay exports 3 scoring function factories.

Two of these algorithms decay with time, and the other is based purely on statistical popularity.

// 1. zero decay 
var wilsonScore = decay.wilsonScore(zScore);
var score = wilsonScore(upVotes, downVotes);
// 2. decays 
var redditHotScore = decay.redditHot(halflife);
var score = redditHotScore(upVotes, downVotes, date);
// 3. decays 
var hackerHotScore = decay.hackerHot(decay);
var score = hackerHotScore(upVotes, date);

AKA Reddit's Best comment sorting system.

Statistically, it is the lower bound of the Wilson Score interval at the alpha level based on supplied Z score.

The optional zScore parameter can be passed as to the exported wilsonScore factory. The Z score is a statistical value which roughly means how many standard deviations of safety you want, so it maps directly onto the confidence level of the Wilson Score interval.

It will default to z=1.96 if left out, representing a 95% confidence level in the lower bound. Otherwise, values through 1.0 (69%), to 3.3 (99.9%) good alternatives.

Based on the difference between ups/downs, and decays with time. Causes hive mind effects in large crowds.

An optional halflife parameter can be passed to the exported redditHot factory. The half-life defaults to 45000 [s]. For info on the effects on this parameter read the original blog post about it.

Based on simply the amount of upvotes, and decays with time. Prone to advertising abuse.

An optional gravity parameter (defaulting to 1.8) can be passed to the exported hackerHot factory. For info on the effects of this parameter read the original blog post about it.

$ npm install decay

MIT-Licensed. See LICENSE file for details.