MediumNLPnlp-143

Reranker Feature Build

Problem

Build a four-element reranker feature vector: [retrieval_score, query-overlap count, query length, passage length].

The overlap count is the number of distinct query terms that occur anywhere in the passage.

Hint

Deduplicate the query tokens before counting overlap, but keep the raw lengths.

case 1

in: [0.8,["cat","sat"],["the","cat","sat","down"]]

expected: [0.8,2,2,4]

case 2

in: [0.5,["dog"],["cat","dog"]]

expected: [0.5,1,1,2]

case 3

in: [0,[],["a"]]

expected: [0,0,0,1]

case 4

in: [1,["a","a"],["a"]]

expected: [1,1,2,1]

Starter codePython

def reranker_feature_build(retrieval_score, query_tokens, passage_tokens):
    # Your code here
    pass

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