Egothor/Radixor
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
9c5b9e331bd1cf2fe67b8062206a0685f868e2d9
Radixor/docs/benchmarks/languages/norwegian-bokmal.md
Leo Galambos 05f3855b99 feat(benchmarks): expand multilingual stemming quality evaluation 
* cover all Radixor dictionary languages
* add PRIMARY_OUTPUT, ANY_CANDIDATE, and ALL_CANDIDATES policies
* measure pairwise over-stemming and under-stemming
* add balanced accuracy and complementary quality metrics
* compare single-output and multi-output stemmers fairly
* improve result validation, reporting, and documentation
* move stemming quality tests into the standard test source set
* preserve the existing JMH benchmark structure and badge output
2026-07-20 23:20:17 +02:00

28 KiB
Raw Blame History

Norwegian Bokmal Stemmer Benchmarks

This page reports same-language stemming benchmarks for Norwegian Bokmal. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.

All speed values are environment-specific and were measured on the hardware and JVM listed in the benchmark overview. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.

Radixor must not be read as simply "slower" when a narrow competitor has a lower timing row. In these tables Radixor is the quality-oriented baseline: its exact-root accuracy is typically close to 100%, while many faster rule-based, light, minimal, or possessive filters reach that speed by doing much less linguistic work and often score far lower in All exact and Changed exact. The Radixor rows in this benchmark refresh use the contracted compiled patch trie: compilation collapses uniform patch-command subtrees into accepting leaves, reducing hot lookup depth while preserving the preferred stemming result measured by the accuracy pass. The EnglishRadixorDictionaryCoverageBenchmark table shows the resulting quality/speed envelope explicitly. The same interpretation applies to this language page: speed rows must be read together with the accuracy table above them.

Dictionary Corpus

Resource Dictionary rows Complete quality tokens Already-root tokens Changed speed tokens
NB_NO 17,929 90,757 33,376 57,381

Radixor Patch Command Distribution

Radixor stores the preferred transformation for each normalized dictionary word form as a compiled patch command. This distribution shows which runtime command class is selected by the trained trie for the complete language dictionary. The total number of preferred patch commands analyzed for this language is 90,757.

Command class Meaning Word forms Share
AppendCharacterCommand Appends one character to the end of the word form. 1,500 1.653%
BackwardCompoundCommand Applies a multi-step backward patch made from skip, delete, insert, and replace operations. 4,296 4.734%
DeleteSuffixCommand Deletes one or more trailing characters from the word form. 47,619 52.469%
PreserveCommand Returns the word form unchanged because it already matches the preferred root. 34,420 37.925%
ReplaceLastCharacterCommand Replaces the final character of the word form. 2,922 3.220%

Accuracy

Accuracy is computed from JMH auxiliary counters in the current report. The counters are deterministic for a fixed corpus and stemmer; percentages divide matching counters by evaluated counters from the same report and are not timing metrics.

Stemmer All exact Changed exact Root preserved Note
Radixor 96.852% 97.637% 95.503% Full Radixor dictionary patch-command stemmer.
Lucene NorwegianMinimalStemFilter 57.107% 53.913% 62.599% Minimal suffix reducer; narrow baseline, not a full stemmer.
Official Snowball direct 54.824% 51.791% 60.040% Official Snowball generated Java stemmer; rule-based suffix algorithm.
Lucene SnowballFilter 54.803% 51.780% 60.001% Lucene TokenFilter integration path around the Snowball algorithm.
Lucene NorwegianLightStemFilter 52.136% 50.616% 54.749% Light suffix stemmer; intentionally narrower than a dictionary-derived stemmer.

Speed

Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 1 fork, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.

Stemmer Benchmark method Score ms/op Error ms ns/token Relative vs Radixor Note
Radixor norwegianBokmalRadixor 3.631 1.377 63.3 1.000 Full Radixor dictionary patch-command stemmer.
Lucene NorwegianMinimalStemFilter norwegianBokmalLuceneNorwegianMinimalStemFilter 2.910 0.177 50.7 0.801 Minimal Norwegian suffix reducer.
Lucene NorwegianLightStemFilter norwegianBokmalLuceneNorwegianLightStemFilter 3.335 0.116 58.1 0.919 Light Norwegian suffix stemmer.
Official Snowball direct snowballDirect[NORWEGIAN_BOKMAL] 4.277 0.082 74.5 1.178 Official Snowball generated Java stemmer; direct API.
Lucene SnowballFilter luceneSnowballFilter[NORWEGIAN_BOKMAL] 6.077 0.208 105.9 1.674 Lucene TokenFilter path around Snowball; includes TokenStream overhead.

Interpretation Notes

  • Radixor is a dictionary-derived patch-command stemmer. Its quality depends on the language resource used to train the compiled trie.
  • Light, minimal, plural, and possessive filters are narrow baselines. They can be fast because they intentionally perform less linguistic work.
  • Lucene TokenFilter rows include TokenStream, attribute, and required normalization overhead. Direct rows measure exposed direct APIs.
  • Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
  • Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.

Stemming Quality

Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language NB_NO using the complete validated stemming-quality result matrix. Every usable dictionary row is one gold-standard group of forms expected to share a morphological family or lemma. Exact equality with a predetermined lemma is not required. Same-row pairs are positive pairs; pairs from different rows are negative pairs.

ALL_WORDS includes every valid group and its original forms. LOWERCASE_GROUPS_ONLY excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. Download the complete machine-readable result snapshot.

Evaluation Scope and Key Findings

The dictionary resource is src/main/resources/nb_no/stemmer.gz. The following findings compare only deterministic PRIMARY_OUTPUT rows over identical included groups; candidate policies are reported separately as capability analyses.

  • ALL_WORDS: Radixor ranks first by balanced accuracy at 0.974783 among 5 deterministic stemmers. The runner-up is SNOWBALL NORWEGIAN BOKMAL DIRECT at 0.874964, a difference of 0.099819. This rank does not imply leadership in throughput or every secondary metric.
  • LOWERCASE_GROUPS_ONLY: Radixor ranks first by balanced accuracy at 0.975000 among 5 deterministic stemmers. The runner-up is SNOWBALL NORWEGIAN BOKMAL DIRECT at 0.874991, a difference of 0.100009. This rank does not imply leadership in throughput or every secondary metric.

ALL_WORDS

This mode contains 7 result rows, 5 evaluated stemmers, and 3 output policies. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.

PRIMARY_OUTPUT ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor PRIMARY_OUTPUT 0.974783 11482 / 2835618215 (0.000405%) 7170 / 142180 (5.042903%) 0.927078 0.935387 0.935488
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.874964 23997 / 2835618215 (0.000846%) 35554 / 142180 (25.006330%) 0.802095 0.781707 0.782399
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.874834 24046 / 2835618215 (0.000848%) 35591 / 142180 (25.032353%) 0.801759 0.781401 0.782091
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.850006 25171 / 2835618215 (0.000888%) 42651 / 142180 (29.997890%) 0.776381 0.745871 0.747464
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.832414 14772 / 2835618215 (0.000521%) 47654 / 142180 (33.516669%) 0.815763 0.751764 0.758263
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor PRIMARY_OUTPUT 0.921620 0.949571 0.999996 0.974783 0.999993 0.000007
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.816288 0.749937 0.999992 0.874964 0.999979 0.000021
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.815930 0.749676 0.999992 0.874834 0.999979 0.000021
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.798148 0.700021 0.999991 0.850006 0.999976 0.000024
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.864847 0.664833 0.999995 0.832414 0.999978 0.000022
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor PRIMARY_OUTPUT 0.927078 0.935387 0.943846 0.878617 0.935491 0.935488
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.802095 0.781707 0.762330 0.641641 0.782409 0.782399
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.801759 0.781401 0.762052 0.641229 0.782102 0.782091
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.776381 0.745871 0.717668 0.594732 0.747476 0.747464
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.815763 0.751764 0.697076 0.602261 0.758274 0.758263
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor PRIMARY_OUTPUT 0.935384 0.993354 0.994615 0.993984 0.993984
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.781696 0.988328 0.971120 0.979648 0.979648
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.781391 0.988295 0.971086 0.979615 0.979615
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.745859 0.987774 0.965622 0.976573 0.976573
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.751753 0.992089 0.962516 0.977079 0.977079
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor PRIMARY_OUTPUT 135010 11482 7170 2835606733 11482 / 2835618215 7170 / 142180
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 106626 23997 35554 2835594218 23997 / 2835618215 35554 / 142180
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 106589 24046 35591 2835594169 24046 / 2835618215 35591 / 142180
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 99529 25171 42651 2835593044 25171 / 2835618215 42651 / 142180
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 94526 14772 47654 2835603443 14772 / 2835618215 47654 / 142180

ANY_CANDIDATE ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ANY_CANDIDATE 1.000000 0 / 2835618215 (0.000000%) 0 / 142180 (0.000000%) 1.000000 1.000000 1.000000
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor ANY_CANDIDATE 1.000000 1.000000 1.000000 1.000000 1.000000 0.000000
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor ANY_CANDIDATE 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor ANY_CANDIDATE n/a n/a n/a n/a n/a
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor ANY_CANDIDATE 142180 0 0 2835618215 0 / 2835618215 0 / 142180

ALL_CANDIDATES ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ALL_CANDIDATES 0.999996 20161 / 2835618215 (0.000711%) 0 / 142180 (0.000000%) 0.898118 0.933794 0.935844
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor ALL_CANDIDATES 0.875811 1.000000 0.999993 0.999996 0.999993 0.000007
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor ALL_CANDIDATES 0.898118 0.933794 0.972422 0.875811 0.935848 0.935844
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor ALL_CANDIDATES n/a n/a n/a n/a n/a
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor ALL_CANDIDATES 142180 20161 0 2835598054 20161 / 2835618215 0 / 142180

Multi-output analysis

Alternative candidates are capability analyses, not replacements for the deterministic comparison.

Stemmer Under pairs repaired Best-case over pairs avoided All-candidate collisions added Multi-candidate forms Multi-candidate share Maximum candidates Total candidate assignments
Radixor 7170 11482 8679 4237 5.626079% 9 79825

LOWERCASE_GROUPS_ONLY

This mode contains 7 result rows, 5 evaluated stemmers, and 3 output policies. Applied-row and form counts are shown per row because adapters share the language corpus but policy rows remain independently auditable. Rankings are separated by output policy and ordered by unrounded balanced accuracy, followed by MCC, F1, over-stemming rate, over-stemming count, under-stemming rate, and stemmer. Balanced accuracy is a navigation metric, not a universally authoritative quality score.

PRIMARY_OUTPUT ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor PRIMARY_OUTPUT 0.975000 11482 / 2831176784 (0.000406%) 7104 / 142091 (4.999613%) 0.927151 0.935591 0.935695
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.874991 23997 / 2831176784 (0.000848%) 35524 / 142091 (25.000880%) 0.802043 0.781698 0.782388
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.874798 23993 / 2831176784 (0.000847%) 35579 / 142091 (25.039587%) 0.801914 0.781464 0.782161
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.849947 25118 / 2831176784 (0.000887%) 42641 / 142091 (30.009642%) 0.776513 0.745896 0.747500
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.832344 14719 / 2831176784 (0.000520%) 47644 / 142091 (33.530625%) 0.815950 0.751796 0.758325
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor PRIMARY_OUTPUT 0.921608 0.950004 0.999996 0.975000 0.999993 0.000007
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.816205 0.749991 0.999992 0.874991 0.999979 0.000021
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.816153 0.749604 0.999992 0.874798 0.999979 0.000021
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.798359 0.699904 0.999991 0.849947 0.999976 0.000024
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.865169 0.664694 0.999995 0.832344 0.999978 0.000022
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor PRIMARY_OUTPUT 0.927151 0.935591 0.944186 0.878976 0.935698 0.935695
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.802043 0.781698 0.762360 0.641630 0.782398 0.782388
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.801914 0.781464 0.762031 0.641314 0.782171 0.782161
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.776513 0.745896 0.717603 0.594765 0.747512 0.747500
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.815950 0.751796 0.696995 0.602302 0.758335 0.758325
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor PRIMARY_OUTPUT 0.935587 0.993348 0.994694 0.994020 0.994020
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 0.781688 0.988318 0.971127 0.979647 0.979647
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 0.781454 0.988310 0.971074 0.979616 0.979616
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 0.745885 0.987789 0.965603 0.976570 0.976570
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 0.751785 0.992107 0.962494 0.977076 0.977076
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor PRIMARY_OUTPUT 134987 11482 7104 2831165302 11482 / 2831176784 7104 / 142091
2 SNOWBALL NORWEGIAN BOKMAL DIRECT PRIMARY_OUTPUT 106567 23997 35524 2831152787 23997 / 2831176784 35524 / 142091
3 SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER PRIMARY_OUTPUT 106512 23993 35579 2831152791 23993 / 2831176784 35579 / 142091
4 NORWEGIAN BOKMAL LUCENE NORWEGIAN LIGHT STEM FILTER PRIMARY_OUTPUT 99450 25118 42641 2831151666 25118 / 2831176784 42641 / 142091
5 NORWEGIAN BOKMAL LUCENE NORWEGIAN MINIMAL STEM FILTER PRIMARY_OUTPUT 94447 14719 47644 2831162065 14719 / 2831176784 47644 / 142091

ANY_CANDIDATE ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ANY_CANDIDATE 1.000000 0 / 2831176784 (0.000000%) 0 / 142091 (0.000000%) 1.000000 1.000000 1.000000
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor ANY_CANDIDATE 1.000000 1.000000 1.000000 1.000000 1.000000 0.000000
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor ANY_CANDIDATE 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor ANY_CANDIDATE n/a n/a n/a n/a n/a
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor ANY_CANDIDATE 142091 0 0 2831176784 0 / 2831176784 0 / 142091

ALL_CANDIDATES ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ALL_CANDIDATES 0.999996 20161 / 2831176784 (0.000712%) 0 / 142091 (0.000000%) 0.898061 0.933756 0.935808
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor ALL_CANDIDATES 0.875743 1.000000 0.999993 0.999996 0.999993 0.000007
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor ALL_CANDIDATES 0.898061 0.933756 0.972405 0.875743 0.935811 0.935808
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor ALL_CANDIDATES n/a n/a n/a n/a n/a
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor ALL_CANDIDATES 142091 20161 0 2831156623 20161 / 2831176784 0 / 142091

Multi-output analysis

Alternative candidates are capability analyses, not replacements for the deterministic comparison.

Stemmer Under pairs repaired Best-case over pairs avoided All-candidate collisions added Multi-candidate forms Multi-candidate share Maximum candidates Total candidate assignments
Radixor 7104 11482 8679 4204 5.586637% 9 79733

Output Policies and Metric Definitions

PRIMARY_OUTPUT uses one deterministic stem per form and therefore defines a strict partition. ANY_CANDIDATE is an optimistic oracle-assisted pairwise upper bound: a same-group pair succeeds when candidates intersect, while a different-group pair succeeds when a non-colliding selection exists. Candidate choices may differ between pairs, so this is not deterministic runtime behaviour and need not represent one globally consistent assignment. ALL_CANDIDATES activates every returned candidate; forms are related when candidate sets intersect. Alternatives can reduce under-stemming but can introduce cross-group collisions, and the resulting relation can overlap and need not be a partition.

For each row, TP = underPossiblePairs - underErrorPairs, FN = underErrorPairs, FP = overErrorPairs, and TN = overPossiblePairs - overErrorPairs. TP and FN concern same-group pairs; FP and TN concern different-group pairs. Consequently, under-stemming and over-stemming use different denominators. Undefined values are rendered as n/a.

  • Under-stemming rate: FN / (TP + FN), the false-negative rate over same-group pairs.
  • Over-stemming rate: FP / (TN + FP), the false-positive rate over different-group pairs.
  • Pairwise precision: TP / (TP + FP), the fraction of predicted conflations that are gold-standard positive pairs.
  • Pairwise recall: TP / (TP + FN), the fraction of gold-standard positive pairs successfully connected.
  • Pairwise specificity: TN / (TN + FP), the fraction of different-group pairs correctly separated.
  • Balanced accuracy: (recall + specificity) / 2. It gives equal weight to positive and negative pair classes and is less dominated by the large true-negative class than ordinary accuracy. It does not replace the raw errors or other metrics.
  • Pairwise F-beta: ((1 + betaSquared) * TP) / (((1 + betaSquared) * TP) + (betaSquared * FN) + FP). F0.5 emphasizes precision and penalizes over-stemming more; F1 weights precision and recall equally; F2 emphasizes recall and penalizes under-stemming more.
  • MCC: (TP * TN - FP * FN) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN)). It uses all confusion counts and remains useful under class imbalance, except when its denominator is degenerate.
  • Jaccard index: TP / (TP + FP + FN).
  • Fowlkes–Mallows index: sqrt(precision * recall).
  • Pairwise accuracy: (TP + TN) / (TP + TN + FP + FN). It can be dominated by true-negative cross-group pairs.
  • Pairwise error rate: (FP + FN) / (TP + TN + FP + FN).

Adjusted Rand Index uses the gold/predicted contingency table and chance correction. Homogeneity is 1 - H(gold | predicted) / H(gold); completeness is 1 - H(predicted | gold) / H(predicted); V-measure is their harmonic mean; normalized mutual information uses the arithmetic-mean entropy normalization MI / ((H(gold) + H(predicted)) / 2). These partition-only metrics apply to PRIMARY_OUTPUT; candidate-relation rows show n/a.

Provenance

  • Authoritative source: docs/benchmarks/data/stemming-quality.csv
  • Source SHA-256: 5a93a6ab60e46489737cd649eb1ac48182114b9038f7f20195ab9d1c1fc0dd28
  • Evaluation command: ./gradlew stemmingQuality
  • Dictionary language: NB_NO
  • Processing modes: ALL_WORDS, LOWERCASE_GROUPS_ONLY
  • Stemmer versions and transitive artifacts: resolved by the repository's JMH Gradle configuration and gradle.lockfile
  • Radixor version, Git revision, generation date, JDK version, operating system, and dictionary revision: not recorded in the authoritative CSV
Reference in New Issue View Git Blame Copy Permalink