MediumNLPnlp-146

BM25F Field-Weighted Score

Problem

Return a field-weighted BM25F score. For each field i add weights[i] * idf * (tf_i * (k1 + 1)) / (tf_i + k1 * (1 - b + b * len_i / avg_len_i)).

idf = log(1 + (N - df + 0.5) / (df + 0.5)), and fields with term frequency 0 contribute nothing.

Hint

The idf factor is shared; each field only changes the saturation term and its weight.

case 1

in: [[2,0],[50,100],[50,100],[1,0.5],10,100]

expected: 3.233921799539688

case 2

in: [[1,1],[25,50],[50,50],[1,0.3],1,2]

expected: 1.1023276129550097

case 3

in: [[0,0],[50,50],[50,50],[1,1],5,10]

expected: 0

case 4

in: [[2,2],[50,50],[50,50],[0,1],10,100]

expected: 3.233921799539688

Starter codePython

import math
def bm25f_score(term_freqs, field_lens, avg_field_lens, weights, df, n_docs, k1=1.5, b=0.75):
    # Your code here
    pass

4 test cases · run your code against them in the editor