forked from
tangled.org/core
Monorepo for Tangled
appview/indexer/repos: boost popular repos in search
repos with more stars go higher up in the search results.
the final result order is a bit magic right now, generally speaking:
forks go lower, more stars go higher, for now.
some more areas of improvement:
- boosting for prefix matches
- boosting for exact matches
- boosting for matches in repo name over repo desc/topic/website etc.
Signed-off-by: oppiliappan <me@oppi.li>
+135
-2
+6
appview/indexer/bleve/query.go
+6
appview/indexer/bleve/query.go
+33
-2
appview/indexer/repos/indexer.go
+33
-2
appview/indexer/repos/indexer.go
···
414
414
indexerQuery.AddMustNot(mustNots...)
415
415
416
416
// use a disjunction where:
417
-
// - Non-forks get normal relevance score
418
-
// - Forks match but get penalized with lower boost
417
+
// - repos with more stars get higher boost
418
+
// - non-forks get a boost
419
+
// - boosts stack
419
420
finalQuery := bleve.NewDisjunctionQuery()
420
421
421
422
// add the main query
···
429
430
notForkQuery.AddMustNot(isForkQuery)
430
431
notForkQuery.SetBoost(2.0)
431
432
finalQuery.AddQuery(notForkQuery)
433
+
434
+
// add boosted queries for repos with more stars
435
+
// 10-99 stars
436
+
tier2Query := bleve.NewBooleanQuery()
437
+
tier2Query.AddMust(indexerQuery)
438
+
min10 := float64(10)
439
+
max99 := float64(99)
440
+
starRange2 := bleveutil.NumericRangeQuery("star_count", &min10, &max99)
441
+
tier2Query.AddMust(starRange2)
442
+
tier2Query.SetBoost(1.5)
443
+
finalQuery.AddQuery(tier2Query)
444
+
445
+
// 100-999 stars
446
+
tier3Query := bleve.NewBooleanQuery()
447
+
tier3Query.AddMust(indexerQuery)
448
+
min100 := float64(100)
449
+
max999 := float64(999)
450
+
starRange3 := bleveutil.NumericRangeQuery("star_count", &min100, &max999)
451
+
tier3Query.AddMust(starRange3)
452
+
tier3Query.SetBoost(2.5)
453
+
finalQuery.AddQuery(tier3Query)
454
+
455
+
// 1000+ stars
456
+
tier4Query := bleve.NewBooleanQuery()
457
+
tier4Query.AddMust(indexerQuery)
458
+
min1000 := float64(1000)
459
+
starRange4 := bleveutil.NumericRangeQuery("star_count", &min1000, nil)
460
+
tier4Query.AddMust(starRange4)
461
+
tier4Query.SetBoost(4.0)
462
+
finalQuery.AddQuery(tier4Query)
432
463
433
464
// use minimum of 1 to ensure all results match at least one clause
434
465
finalQuery.SetMin(1)
+96
appview/indexer/repos/indexer_test.go
+96
appview/indexer/repos/indexer_test.go
···
637
637
assert.Equal(t, uint64(1), result.Total)
638
638
assert.Contains(t, result.Hits, int64(2))
639
639
}
640
+
641
+
func TestStarCountBoosting(t *testing.T) {
642
+
ix, cleanup := setupTestIndexer(t)
643
+
defer cleanup()
644
+
645
+
ctx := context.Background()
646
+
647
+
err := ix.Index(ctx,
648
+
models.Repo{
649
+
Id: 1,
650
+
Did: "did:plc:alice",
651
+
Name: "repo",
652
+
Description: "testing",
653
+
RepoStats: &models.RepoStats{StarCount: 5000},
654
+
},
655
+
models.Repo{
656
+
Id: 2,
657
+
Did: "did:plc:bob",
658
+
Name: "repo",
659
+
Description: "testing",
660
+
RepoStats: &models.RepoStats{StarCount: 150},
661
+
},
662
+
models.Repo{
663
+
Id: 3,
664
+
Did: "did:plc:charlie",
665
+
Name: "repo",
666
+
Description: "testing",
667
+
RepoStats: &models.RepoStats{StarCount: 5},
668
+
},
669
+
models.Repo{
670
+
Id: 4,
671
+
Did: "did:plc:dana",
672
+
Name: "repo",
673
+
Description: "testing",
674
+
RepoStats: &models.RepoStats{StarCount: 25},
675
+
},
676
+
)
677
+
require.NoError(t, err)
678
+
679
+
// search for "testing" - should rank by star count when all else equal
680
+
result, err := ix.Search(ctx, models.RepoSearchOptions{
681
+
Keywords: []string{"testing"},
682
+
Page: pagination.Page{Limit: 10},
683
+
})
684
+
require.NoError(t, err)
685
+
assert.Equal(t, uint64(4), result.Total)
686
+
687
+
// verify that repos with more stars rank higher than those with fewer
688
+
popularIdx := -1
689
+
smallIdx := -1
690
+
for i, hit := range result.Hits {
691
+
if hit == 1 { // 5000 stars
692
+
popularIdx = i
693
+
}
694
+
if hit == 3 { // 5 stars
695
+
smallIdx = i
696
+
}
697
+
}
698
+
assert.True(t, popularIdx < smallIdx, "repo with 5000 stars should rank above repo with 5 stars")
699
+
}
700
+
701
+
func TestStarBoostingWithForkPenalty(t *testing.T) {
702
+
ix, cleanup := setupTestIndexer(t)
703
+
defer cleanup()
704
+
705
+
ctx := context.Background()
706
+
707
+
err := ix.Index(ctx,
708
+
models.Repo{
709
+
Id: 1,
710
+
Did: "did:plc:alice",
711
+
Name: "original-popular",
712
+
Description: "test project",
713
+
Source: "",
714
+
RepoStats: &models.RepoStats{StarCount: 100},
715
+
},
716
+
models.Repo{
717
+
Id: 2,
718
+
Did: "did:plc:bob",
719
+
Name: "fork-very-popular",
720
+
Description: "test project",
721
+
Source: "did:plc:someone/original",
722
+
RepoStats: &models.RepoStats{StarCount: 1000},
723
+
},
724
+
)
725
+
require.NoError(t, err)
726
+
727
+
result, err := ix.Search(ctx, models.RepoSearchOptions{
728
+
Keywords: []string{"project"},
729
+
Page: pagination.Page{Limit: 10},
730
+
})
731
+
require.NoError(t, err)
732
+
733
+
// fork with 1000 stars (4.0x) vs non-fork with 100 stars (2.0 * 2.5 = 5.0x)
734
+
assert.Equal(t, int64(1), result.Hits[0], "non-fork with fewer stars can still rank higher due to combined boost")
735
+
}