Increase filename boost (#785)

jtibshirani · web-flow · commit 640102a4a30e · 2024-05-28T14:51:34.000-07:00
When we introduced filename boosting in BM25, we set it to a very conservative
weight. This PR increases the weight from 2.0 -&gt; 5.0, which improves results on
relevant evals.

Relates to SPLF-88
diff --git a/build/scoring_test.go b/build/scoring_test.go
@@ -78,7 +78,7 @@ func TestBM25(t *testing.T) {
 			content:  exampleJava,
 			language: "Java",
 			// bm25-score:1.69 (sum-tf: 7.00, length-ratio: 2.00)
-			wantScore: 1.69,
+			wantScore: 1.82,
 		}, {
 			// Matches only on content
 			fileName: "example.java",
@@ -99,15 +99,15 @@ func TestBM25(t *testing.T) {
 			content:  exampleJava,
 			language: "Java",
 			// bm25-score:1.07 (sum-tf: 2.00, length-ratio: 2.00)
-			wantScore: 1.07,
+			wantScore: 1.55,
 		},
 		{
 			// Matches only on filename, and content is missing
 			fileName: "a/b/c/config.go",
 			query:    &query.Substring{Pattern: "config.go"},
 			language: "Go",
 			// bm25-score:1.91 (sum-tf: 2.00, length-ratio: 0.00)
-			wantScore: 1.91,
+			wantScore: 2.08,
 		},
 	}
 
diff --git a/score.go b/score.go
@@ -132,7 +132,7 @@ func (d *indexData) scoreFileUsingBM25(fileMatch *FileMatch, doc uint32, cands [
 		term := string(cand.substrLowered)
 
 		if cand.fileName {
-			termFreqs[term] += 2
+			termFreqs[term] += 5
 		} else {
 			termFreqs[term]++
 		}

Original file line number	Diff line number	Diff line change
`@@ -132,7 +132,7 @@ func (d indexData) scoreFileUsingBM25(fileMatch FileMatch, doc uint32, cands [`
`132`	`132`	`term := string(cand.substrLowered)`
`133`	`133`
`134`	`134`	`if cand.fileName {`
`135`		`- termFreqs[term] += 2`
	`135`	`+ termFreqs[term] += 5`
`136`	`136`	`} else {`
`137`	`137`	`termFreqs[term]++`
`138`	`138`	`}`