Skip to content

Instantly share code, notes, and snippets.

@berstend
Created December 16, 2018 19:18
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save berstend/5864e430e9d29aa9ad9f46bcc27fa4e3 to your computer and use it in GitHub Desktop.
Save berstend/5864e430e9d29aa9ad9f46bcc27fa4e3 to your computer and use it in GitHub Desktop.
Given star ratings and votes, calculate bayesian estimate ratings to sort by popularity (taking number of votes into consideration)
/*
bayesian rating = (v ÷ (v+m)) × R + (m ÷ (v+m)) × C
where:
R = stars for the entry
v = number of votes for the entry
m = minimum votes required to be listed (eg. 10)
C = the mean stars across the whole list (eg. 4.1)
This rating only includes entries that have at least 10 votes.
The bayesian estimate is a statistical technique used to reduce the noise
due to low sample counts. In effect, the less an entry has votes, the more
it is pulled towards the mean (e.g. 4.1). In other words, these are the entry
that many people agree are great.
https://www.animenewsnetwork.com/encyclopedia/ratings-anime.php#best_bayesian
*/
const calculateBayesianEstimate = (R,v,m,C)=>{
return (v / (v + m)) * R + (m / (v + m)) * C
}
// calculate mean (= average)
const mean = arr=>arr.reduce((p,c)=>p + c, 0) / arr.length;
const entries = [{
id: 1,
stars: 4.5,
votes: 700
}, {
id: 2,
stars: 3.7,
votes: 1000
}, {
id: 3,
stars: 5,
votes: 10
}, {
id: 4,
stars: 4.8,
votes: 169
}, {
id: 5,
stars: 2.3,
votes: 8000
}, ]
const totalVotesMean = mean(entries.map(e=>e.votes))
const totalStarsMean = mean(entries.map(e=>e.stars))
const minVotes = totalVotesMean * 0.1 // 10 percent of mean votes is minimum
console.log({
totalVotesMean,
totalStarsMean,
minVotes
})
// => {totalVotesMean: 1975.8, totalStarsMean: 4.0600000000000005, minVotes: 197.58}
for (const entry of entries) {
const rank = calculateBayesianEstimate(entry.stars, entry.votes, minVotes, totalStarsMean)
entry.rank = rank
}
console.log(entries.sort((a,b)=>a.rank - b.rank).reverse())
/* =>
[
{
"id": 1,
"stars": 4.5,
"votes": 700,
"rank": 4.403144900733082
},
{
"id": 4,
"stars": 4.8,
"votes": 169,
"rank": 4.401153363522287
},
{
"id": 3,
"stars": 5,
"votes": 10,
"rank": 4.105283746025629
},
{
"id": 2,
"stars": 3.7,
"votes": 1000,
"rank": 3.759393777451194
},
{
"id": 5,
"stars": 2.3,
"votes": 8000,
"rank": 2.3424199336877467
}
]
*/
@berstend
Copy link
Author

This took way too long to figure out 😅

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment