New Trending Algorithm - Attempt Number 4

Next attempt for the Tending algorithm:

Score calculation with linear gradation between 0 and 30 votes and scaling down below 10 votes:

def _score_with_linear_median_and_min(active_votes, created_timestamp, timescale=480000):
    """Calculate trending/hot score with linear median and min rshares."""
    if active_votes:
        rshares = [int(vote['rshares']) for vote in active_votes]
        length = len(rshares)
        median_rshares = sorted(rshares)[length // 2]
        median_rshares *= 1 if length > 10 else 0.00001 # under 10 votes scale down
        weighted_median_rshares = int(median_rshares * min(length / 30, 1))  # Linear weighted median
    else:
        weighted_median_rshares = 0
    return _score(weighted_median_rshares, created_timestamp, timescale)

This is the code extract for the calculation used now. As always, it will be a while before you see it on the test trending page.

Thoughts

The last attempt (blue line below) consisted of first calculating the median of all votes and then weakening the score of posts with fewer votes with log10. However, this resulted in posts with particularly few votes receiving a higher score. In other words, the exact opposite of what was intended.

Therefore, I have considered that posts with very few votes should not play a role in the trending score at all. For posts with fewer than 30 votes, the value of the median increases slowly until the median is no longer weakened from 30 votes. The number 30 is chosen arbitrarily here.

Here you can see a comparison of the three methods:

I'm curious to see how the trendings will develop now :-)

Steem Search on https://moecki.online/

My Full Node on https://api.moecki.online

My Python Pricefeed for Witnesses

Vote for me as Witness

27.09.2024