/ Tag

A last.fm tag cloud generator build on Vue

A last.fm tag cloud generator build on Vue

lastfm-tag-cloud

A last.fm tag cloud generator build on Vue!

How are the tags chosen & scaled?

Initially, the sample of artists (up to the size and of the time period you specify) is iterated through. For each artist, their top tags are fetched, using artist.getTopTags.

Each tag in the response has a count for that artist.

Note: This count doesn't seem to be documented anywhere. They cap out at 100, so I am working under the assumption that they're a kind of confidence % as to how apropriate that tag is for that artist.

For each tag on the artist, if I have not seen it before, I initialise a library_total metric for that tag with an initial value of 0.

The product of the tag's score on that artist and the user's scrobbles of that artist is then added to that tag's library_total metric.

Once all of the artists are iterated through, the tags are pruned to the top 100 by this library_total metric. This is done to avoid hitting rate limits on the last.fm API in the next step, where I have to call tag.getInfo for every tag.

Each of these 100 tags is then scored as per the following code snippet [source]:

score_tags(result){
    for (var tag of result.tags) {
        result.scores[tag] = 0
        /**First, each tagging is weighted by the product of:
         *  - How many times the user has listened to the artist on which the tag was used,
         * and...
         *  - The "count" of that tag on the artist.
         *    I am assuming that this "count" is a confidence % given by last.fm as to the accuracy of the tag on that artist.
         *    I can't find any doccumentation, but this would make sense, as they cap out at 100.
         */
        for (var tagging of result.taggings[tag]) {
            result.scores[tag] += tagging.count/100 * result.listens[tagging.artist]
        }
        /**The sum of all these weighted taggings is then scaled by:
         * 1. How many of the uses of that tag overall fall within the user's library sample (its "uniqueness" to the sample).
         * 
         * 2. How many artists within the sample are tagged with that tag (its "spread" over the sample).
         * 
         * 3. The base 10 logarithm of how many people have used that tag overall (its "reach"; see last.fm API docs).
         *    Base 10 is used so 100 people using the tag makes it twice as significant as 10 people using the tag; a nice balance.
         *    It's also conveniently provided as a function by Math.
         */
        result.scores[tag] = result.scores[tag] 
                                    * (result.tag_meta[tag].library_total / result.tag_meta[tag].total) 
                                    * result.taggings[tag].length * result.taggings[tag].length
                                    * Math.log10(result.tag_meta[tag].reach)
    }
}

I've tried to make this take into account the "uniqueness" of the tag to a user's library, as if they were all just scored by frequency the biggest tag on everyone's clouds would probably just be "all".

GitHub

Comments