summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorBjørn Erik Pedersen <bjorn.erik.pedersen@gmail.com>2023-02-23 15:20:31 +0100
committerBjørn Erik Pedersen <bjorn.erik.pedersen@gmail.com>2023-02-23 17:06:22 +0100
commite442a63bb7659d95aec2d48bf954cd9d61163559 (patch)
treed8868929c74430e530f41ff8a359f451fd981729
parentd5601e8391470be952ba48948c5a976884fea871 (diff)
related: Add config option cardinalityThreshold
Fixes #10744
-rw-r--r--docs/content/en/content-management/related.md4
-rw-r--r--related/inverted_index.go65
-rw-r--r--related/inverted_index_test.go35
-rw-r--r--resources/page/pages_related.go4
4 files changed, 106 insertions, 2 deletions
diff --git a/docs/content/en/content-management/related.md b/docs/content/en/content-management/related.md
index 98af0a36f..c25b0ed61 100644
--- a/docs/content/en/content-management/related.md
+++ b/docs/content/en/content-management/related.md
@@ -160,6 +160,10 @@ applyFilter
weight
: An integer weight that indicates _how important_ this parameter is relative to the other parameters. It can be 0, which has the effect of turning this index off, or even negative. Test with different values to see what fits your content best.
+
+cardinalityThreshold (default 0)
+: {{< new-in "0.111.0" >}}. A percentage (0-100) used to remove common keywords from the index. As an example, setting this to 50 will remove all keywords that are used in more than 50% of the documents in the index.
+
pattern
: This is currently only relevant for dates. When listing related content, we may want to list content that is also close in time. Setting "2006" (default value for date indexes) as the pattern for a date index will add weight to pages published in the same year. For busier blogs, "200601" (year and month) may be a better default.
diff --git a/related/inverted_index.go b/related/inverted_index.go
index eab97098a..967855133 100644
--- a/related/inverted_index.go
+++ b/related/inverted_index.go
@@ -135,9 +135,21 @@ type IndexConfig struct {
// This field's weight when doing multi-index searches. Higher is "better".
Weight int
+ // A percentage (0-100) used to remove common keywords from the index.
+ // As an example, setting this to 50 will remove all keywords that are
+ // used in more than 50% of the documents in the index.
+ CardinalityThreshold int
+
// Will lower case all string values in and queries tothis index.
// May get better accurate results, but at a slight performance cost.
ToLower bool
+
+ // Counts the number of documents in the index.
+ numDocs int
+}
+
+func (cfg *IndexConfig) incrNumDocs() {
+ cfg.numDocs++
}
// Document is the interface an indexable document in Hugo must fulfill.
@@ -169,6 +181,9 @@ type InvertedIndex struct {
minWeight int
maxWeight int
+
+ // No modifications after this is set.
+ finalized bool
}
func (idx *InvertedIndex) getIndexCfg(name string) (IndexConfig, bool) {
@@ -202,8 +217,11 @@ func NewInvertedIndex(cfg Config) *InvertedIndex {
// Add documents to the inverted index.
// The value must support == and !=.
func (idx *InvertedIndex) Add(ctx context.Context, docs ...Document) error {
+ if idx.finalized {
+ panic("index is finalized")
+ }
var err error
- for _, config := range idx.cfg.Indices {
+ for i, config := range idx.cfg.Indices {
if config.Weight == 0 {
// Disabled
continue
@@ -211,6 +229,7 @@ func (idx *InvertedIndex) Add(ctx context.Context, docs ...Document) error {
setm := idx.index[config.Name]
for _, doc := range docs {
+ var added bool
var words []Keyword
words, err = doc.RelatedKeywords(config)
if err != nil {
@@ -218,22 +237,60 @@ func (idx *InvertedIndex) Add(ctx context.Context, docs ...Document) error {
}
for _, keyword := range words {
+ added = true
setm[keyword] = append(setm[keyword], doc)
}
if config.Type == TypeFragments {
if fp, ok := doc.(FragmentProvider); ok {
for _, fragment := range fp.Fragments(ctx).Identifiers {
+ added = true
setm[FragmentKeyword(fragment)] = append(setm[FragmentKeyword(fragment)], doc)
}
}
}
+
+ if added {
+ c := &idx.cfg.Indices[i]
+ (*c).incrNumDocs()
+ }
}
}
return err
}
+func (idx *InvertedIndex) Finalize(ctx context.Context) error {
+ if idx.finalized {
+ return nil
+ }
+
+ for _, config := range idx.cfg.Indices {
+ if config.CardinalityThreshold == 0 {
+ continue
+ }
+ setm := idx.index[config.Name]
+ numDocs := config.numDocs
+ if numDocs == 0 {
+ continue
+ }
+
+ // Remove high cardinality terms.
+ for k, v := range setm {
+ percentageWithKeyword := int(math.Ceil(float64(len(v)) / float64(numDocs) * 100))
+ if percentageWithKeyword > config.CardinalityThreshold {
+ delete(setm, k)
+ }
+ }
+
+ }
+
+ idx.finalized = true
+
+ return nil
+
+}
+
// queryElement holds the index name and keywords that can be used to compose a
// search for related content.
type queryElement struct {
@@ -548,12 +605,16 @@ func DecodeConfig(m maps.Params) (Config, error) {
}
}
for i := range c.Indices {
- if c.Indices[i].Type == "" {
+ icfg := c.Indices[i]
+ if icfg.Type == "" {
c.Indices[i].Type = TypeBasic
}
if !validTypes[c.Indices[i].Type] {
return c, fmt.Errorf("invalid index type %q. Must be one of %v", c.Indices[i].Type, xmaps.Keys(validTypes))
}
+ if icfg.CardinalityThreshold < 0 || icfg.CardinalityThreshold > 100 {
+ return Config{}, errors.New("cardinalityThreshold threshold must be between 0 and 100")
+ }
}
return c, nil
diff --git a/related/inverted_index_test.go b/related/inverted_index_test.go
index d38a7f6eb..c7348e088 100644
--- a/related/inverted_index_test.go
+++ b/related/inverted_index_test.go
@@ -86,6 +86,41 @@ func (d *testDoc) PublishDate() time.Time {
return d.date
}
+func TestCardinalityThreshold(t *testing.T) {
+ c := qt.New(t)
+ config := Config{
+ Threshold: 90,
+ IncludeNewer: false,
+ Indices: IndexConfigs{
+ IndexConfig{Name: "tags", Weight: 50, CardinalityThreshold: 79},
+ IndexConfig{Name: "keywords", Weight: 65, CardinalityThreshold: 90},
+ },
+ }
+
+ idx := NewInvertedIndex(config)
+ hasKeyword := func(index, keyword string) bool {
+ _, found := idx.index[index][StringKeyword(keyword)]
+ return found
+ }
+
+ docs := []Document{
+ newTestDoc("tags", "a", "b", "c", "d"),
+ newTestDoc("tags", "b", "d", "g"),
+ newTestDoc("tags", "b", "d", "g"),
+ newTestDoc("tags", "b", "h").addKeywords("keywords", "a"),
+ newTestDoc("tags", "g", "h").addKeywords("keywords", "a", "b", "z"),
+ }
+
+ idx.Add(context.Background(), docs...)
+ c.Assert(idx.Finalize(context.Background()), qt.IsNil)
+ // Only tags=b should be removed.
+ c.Assert(hasKeyword("tags", "a"), qt.Equals, true)
+ c.Assert(hasKeyword("tags", "b"), qt.Equals, false)
+ c.Assert(hasKeyword("tags", "d"), qt.Equals, true)
+ c.Assert(hasKeyword("keywords", "b"), qt.Equals, true)
+
+}
+
func TestSearch(t *testing.T) {
config := Config{
Threshold: 90,
diff --git a/resources/page/pages_related.go b/resources/page/pages_related.go
index 327369603..217aced47 100644
--- a/resources/page/pages_related.go
+++ b/resources/page/pages_related.go
@@ -236,5 +236,9 @@ func (s *RelatedDocsHandler) getOrCreateIndex(ctx context.Context, p Pages) (*re
s.postingLists = append(s.postingLists, &cachedPostingList{p: p, postingList: searchIndex})
+ if err := searchIndex.Finalize(ctx); err != nil {
+ return nil, err
+ }
+
return searchIndex, nil
}