Egothor/Radixor
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
9c5b9e331bd1cf2fe67b8062206a0685f868e2d9
Radixor/docs/benchmarks/languages/ukrainian.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

33 KiB
Raw Blame History

Ukrainian Stemmer Benchmarks

This page reports same-language stemming benchmarks for Ukrainian. 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
UK_UA 1,493 15,737 2,985 12,752

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 15,737.

Command class Meaning Word forms Share
AppendCharacterCommand Appends one character to the end of the word form. 249 1.582%
BackwardCompoundCommand Applies a multi-step backward patch made from skip, delete, insert, and replace operations. 4,160 26.435%
DeleteSuffixCommand Deletes one or more trailing characters from the word form. 5,859 37.231%
PreserveCommand Returns the word form unchanged because it already matches the preferred root. 3,004 19.089%
ReplaceLastCharacterCommand Replaces the final character of the word form. 2,465 15.664%

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 99.307% 99.365% 99.062% Full Radixor dictionary patch-command stemmer.
Lucene HunspellStemFilter 86.815% 83.759% 99.866% Benchmark-only Ukrainian Hunspell dictionary compared via Lucene HunspellStemFilter.
Lucene MorfologikFilter 92.362% 90.637% 99.732% Dictionary-based path; Morfologik can emit multiple terms.
Morfologik direct 92.362% 90.637% 99.732% Direct dictionary lookup; first returned stem is used for quality when no ranking weight is exposed.

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 ukrainianRadixor 0.682 0.057 53.5 1.000 Full Radixor dictionary patch-command stemmer.
Lucene HunspellStemFilter luceneHunspellStemFilter 43.527 1.207 3413.3 63.799 Benchmark-only Ukrainian Hunspell dictionary compared via Lucene HunspellStemFilter.
Morfologik direct ukrainianMorfologikDirect 8.680 0.073 680.7 12.723 Direct Morfologik dictionary lookup; first returned stem is used for quality.
Lucene MorfologikFilter ukrainianLuceneMorfologikFilter 14.575 0.248 1143.0 21.364 Dictionary-based Morfologik TokenFilter; may emit multiple terms.

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 UK_UA 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/uk_ua/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 0.995343 among 4 deterministic stemmers. The runner-up is UKRAINIAN LUCENE MORFOLOGIK FILTER at 0.928768, a difference of 0.066575. This rank does not imply leadership in throughput or every secondary metric.
  • LOWERCASE_GROUPS_ONLY: Radixor ranks first by balanced accuracy at 0.995342 among 4 deterministic stemmers. The runner-up is UKRAINIAN LUCENE MORFOLOGIK FILTER at 0.928751, a difference of 0.066591. This rank does not imply leadership in throughput or every secondary metric.

ALL_WORDS

This mode contains 12 result rows, 4 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 0.995343 880 / 101387550 (0.000868%) 608 / 65340 (0.930517%) 0.987406 0.988637 0.988632
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.928768 828 / 101387550 (0.000817%) 9308 / 65340 (14.245485%) 0.956896 0.917054 0.919223
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.928646 828 / 101387550 (0.000817%) 9324 / 65340 (14.269972%) 0.956832 0.916912 0.919090
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.885793 794 / 101387550 (0.000783%) 14924 / 65340 (22.840526%) 0.933008 0.865139 0.871499
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor PRIMARY_OUTPUT 0.986588 0.990695 0.999991 0.995343 0.999985 0.000015
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.985438 0.857545 0.999992 0.928768 0.999900 0.000100
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.985434 0.857300 0.999992 0.928646 0.999900 0.000100
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.984495 0.771595 0.999992 0.885793 0.999845 0.000155
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor PRIMARY_OUTPUT 0.987406 0.988637 0.989871 0.977529 0.988639 0.988632
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.956896 0.917054 0.880397 0.846814 0.919270 0.919223
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.956832 0.916912 0.880190 0.846572 0.919137 0.919090
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.933008 0.865139 0.806475 0.762331 0.871568 0.871499
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor PRIMARY_OUTPUT 0.988630 0.997994 0.998266 0.998130 0.998130
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.917004 0.997990 0.971000 0.984310 0.984310
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.916862 0.997990 0.970876 0.984246 0.984246
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.865063 0.998114 0.949804 0.973360 0.973360
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor PRIMARY_OUTPUT 64732 880 608 101386670 880 / 101387550 608 / 65340
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 56032 828 9308 101386722 828 / 101387550 9308 / 65340
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 56016 828 9324 101386722 828 / 101387550 9324 / 65340
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 50416 794 14924 101386756 794 / 101387550 14924 / 65340

ANY_CANDIDATE ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ANY_CANDIDATE 1.000000 0 / 101387550 (0.000000%) 0 / 65340 (0.000000%) 1.000000 1.000000 1.000000
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 0.962151 122 / 101387550 (0.000120%) 4946 / 65340 (7.569636%) 0.982323 0.959732 0.960413
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 0.962029 122 / 101387550 (0.000120%) 4962 / 65340 (7.594123%) 0.982267 0.959599 0.960286
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 0.927570 326 / 101387550 (0.000322%) 9465 / 65340 (14.485767%) 0.962884 0.919443 0.922008
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 0.997984 0.924304 0.999999 0.962151 0.999950 0.000050
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 0.997983 0.924059 0.999999 0.962029 0.999950 0.000050
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 0.994199 0.855142 0.999997 0.927570 0.999903 0.000097
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 0.982323 0.959732 0.938156 0.922581 0.960438 0.960413
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 0.982267 0.959599 0.937954 0.922337 0.960310 0.960286
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 0.962884 0.919443 0.879752 0.850897 0.922053 0.922008
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE n/a n/a n/a n/a n/a
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE n/a n/a n/a n/a n/a
4 HUNSPELL UKRAINIAN LUCENE FILTER 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 65340 0 0 101387550 0 / 101387550 0 / 65340
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 60394 122 4946 101387428 122 / 101387550 4946 / 65340
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 60378 122 4962 101387428 122 / 101387550 4962 / 65340
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 55875 326 9465 101387224 326 / 101387550 9465 / 65340

ALL_CANDIDATES ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ALL_CANDIDATES 0.999993 1490 / 101387550 (0.001470%) 0 / 65340 (0.000000%) 0.982084 0.988727 0.988782
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 0.962145 1368 / 101387550 (0.001349%) 4946 / 65340 (7.569636%) 0.966650 0.950323 0.950669
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 0.962023 1368 / 101387550 (0.001349%) 4962 / 65340 (7.594123%) 0.966592 0.950191 0.950541
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 0.927565 1271 / 101387550 (0.001254%) 9465 / 65340 (14.485767%) 0.950501 0.912349 0.914347
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor ALL_CANDIDATES 0.977705 1.000000 0.999985 0.999993 0.999985 0.000015
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 0.977850 0.924304 0.999987 0.962145 0.999938 0.000062
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 0.977845 0.924059 0.999987 0.962023 0.999938 0.000062
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 0.977759 0.855142 0.999987 0.927565 0.999894 0.000106
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor ALL_CANDIDATES 0.982084 0.988727 0.995460 0.977705 0.988789 0.988782
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 0.966650 0.950323 0.934539 0.905349 0.950700 0.950669
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 0.966592 0.950191 0.934337 0.905109 0.950571 0.950541
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 0.950501 0.912349 0.877142 0.838825 0.914398 0.914347
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES n/a n/a n/a n/a n/a
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES n/a n/a n/a n/a n/a
4 HUNSPELL UKRAINIAN LUCENE FILTER 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 65340 1490 0 101386060 1490 / 101387550 0 / 65340
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 60394 1368 4946 101386182 1368 / 101387550 4946 / 65340
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 60378 1368 4962 101386182 1368 / 101387550 4962 / 65340
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 55875 1271 9465 101386279 1271 / 101387550 9465 / 65340

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
HUNSPELL UKRAINIAN LUCENE FILTER 5459 468 477 1322 9.280449% 6 15740
UKRAINIAN LUCENE MORFOLOGIK FILTER 4362 706 540 2207 15.493155% 6 16937
UKRAINIAN MORFOLOGIK DIRECT 4362 706 540 2207 15.493155% 6 16937
Radixor 608 880 610 190 1.333801% 2 14435

LOWERCASE_GROUPS_ONLY

This mode contains 12 result rows, 4 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 0.995342 880 / 101259406 (0.000869%) 608 / 65324 (0.930745%) 0.987403 0.988634 0.988629
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.928751 828 / 101259406 (0.000818%) 9308 / 65324 (14.248974%) 0.956884 0.917032 0.919202
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.928751 828 / 101259406 (0.000818%) 9308 / 65324 (14.248974%) 0.956884 0.917032 0.919202
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.885796 794 / 101259406 (0.000784%) 14920 / 65324 (22.839998%) 0.933007 0.865141 0.871500
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor PRIMARY_OUTPUT 0.986585 0.990693 0.999991 0.995342 0.999985 0.000015
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.985434 0.857510 0.999992 0.928751 0.999900 0.000100
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.985434 0.857510 0.999992 0.928751 0.999900 0.000100
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.984492 0.771600 0.999992 0.885796 0.999845 0.000155
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor PRIMARY_OUTPUT 0.987403 0.988634 0.989868 0.977524 0.988636 0.988629
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.956884 0.917032 0.880367 0.846777 0.919249 0.919202
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.956884 0.917032 0.880367 0.846777 0.919249 0.919202
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.933007 0.865141 0.806479 0.762334 0.871570 0.871500
Partition metrics (PRIMARY_OUTPUT only)
Rank Stemmer Output policy Adjusted Rand Index Homogeneity Completeness V-measure Normalized mutual information
1 Radixor PRIMARY_OUTPUT 0.988627 0.997992 0.998264 0.998128 0.998128
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 0.916982 0.997988 0.970978 0.984298 0.984298
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 0.916982 0.997988 0.970978 0.984298 0.984298
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 0.865065 0.998113 0.949788 0.973351 0.973351
Raw pair counts
Rank Stemmer Output policy TP FP FN TN Over error / possible Under error / possible
1 Radixor PRIMARY_OUTPUT 64716 880 608 101258526 880 / 101259406 608 / 65324
2 UKRAINIAN LUCENE MORFOLOGIK FILTER PRIMARY_OUTPUT 56016 828 9308 101258578 828 / 101259406 9308 / 65324
3 UKRAINIAN MORFOLOGIK DIRECT PRIMARY_OUTPUT 56016 828 9308 101258578 828 / 101259406 9308 / 65324
4 HUNSPELL UKRAINIAN LUCENE FILTER PRIMARY_OUTPUT 50404 794 14920 101258612 794 / 101259406 14920 / 65324

ANY_CANDIDATE ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ANY_CANDIDATE 1.000000 0 / 101259406 (0.000000%) 0 / 65324 (0.000000%) 1.000000 1.000000 1.000000
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 0.962142 122 / 101259406 (0.000120%) 4946 / 65324 (7.571490%) 0.982318 0.959722 0.960404
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 0.962142 122 / 101259406 (0.000120%) 4946 / 65324 (7.571490%) 0.982318 0.959722 0.960404
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 0.927552 326 / 101259406 (0.000322%) 9465 / 65324 (14.489315%) 0.962874 0.919422 0.921988
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 0.997983 0.924285 0.999999 0.962142 0.999950 0.000050
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 0.997983 0.924285 0.999999 0.962142 0.999950 0.000050
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 0.994198 0.855107 0.999997 0.927552 0.999903 0.000097
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 0.982318 0.959722 0.938141 0.922562 0.960428 0.960404
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 0.982318 0.959722 0.938141 0.922562 0.960428 0.960404
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 0.962874 0.919422 0.879722 0.850861 0.922033 0.921988
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE n/a n/a n/a n/a n/a
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE n/a n/a n/a n/a n/a
4 HUNSPELL UKRAINIAN LUCENE FILTER 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 65324 0 0 101259406 0 / 101259406 0 / 65324
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ANY_CANDIDATE 60378 122 4946 101259284 122 / 101259406 4946 / 65324
3 UKRAINIAN MORFOLOGIK DIRECT ANY_CANDIDATE 60378 122 4946 101259284 122 / 101259406 4946 / 65324
4 HUNSPELL UKRAINIAN LUCENE FILTER ANY_CANDIDATE 55859 326 9465 101259080 326 / 101259406 9465 / 65324

ALL_CANDIDATES ranking

Rank Stemmer Output policy Balanced accuracy Over-stemming Under-stemming F0.5 F1 MCC
1 Radixor ALL_CANDIDATES 0.999993 1490 / 101259406 (0.001471%) 0 / 65324 (0.000000%) 0.982079 0.988724 0.988779
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 0.962136 1368 / 101259406 (0.001351%) 4946 / 65324 (7.571490%) 0.966642 0.950311 0.950657
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 0.962136 1368 / 101259406 (0.001351%) 4946 / 65324 (7.571490%) 0.966642 0.950311 0.950657
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 0.927547 1271 / 101259406 (0.001255%) 9465 / 65324 (14.489315%) 0.950487 0.912326 0.914325
Classification metrics
Rank Stemmer Output policy Precision Recall Specificity Balanced accuracy Pairwise accuracy Error rate
1 Radixor ALL_CANDIDATES 0.977699 1.000000 0.999985 0.999993 0.999985 0.000015
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 0.977845 0.924285 0.999986 0.962136 0.999938 0.000062
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 0.977845 0.924285 0.999986 0.962136 0.999938 0.000062
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 0.977752 0.855107 0.999987 0.927547 0.999894 0.000106
Pair-relation metrics
Rank Stemmer Output policy F0.5 F1 F2 Jaccard Fowlkes–Mallows MCC
1 Radixor ALL_CANDIDATES 0.982079 0.988724 0.995459 0.977699 0.988787 0.988779
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 0.966642 0.950311 0.934522 0.905326 0.950688 0.950657
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 0.966642 0.950311 0.934522 0.905326 0.950688 0.950657
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 0.950487 0.912326 0.877111 0.838787 0.914376 0.914325
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
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES n/a n/a n/a n/a n/a
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES n/a n/a n/a n/a n/a
4 HUNSPELL UKRAINIAN LUCENE FILTER 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 65324 1490 0 101257916 1490 / 101259406 0 / 65324
2 UKRAINIAN LUCENE MORFOLOGIK FILTER ALL_CANDIDATES 60378 1368 4946 101258038 1368 / 101259406 4946 / 65324
3 UKRAINIAN MORFOLOGIK DIRECT ALL_CANDIDATES 60378 1368 4946 101258038 1368 / 101259406 4946 / 65324
4 HUNSPELL UKRAINIAN LUCENE FILTER ALL_CANDIDATES 55859 1271 9465 101258135 1271 / 101259406 9465 / 65324

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
HUNSPELL UKRAINIAN LUCENE FILTER 5455 468 477 1321 9.279292% 6 15730
UKRAINIAN LUCENE MORFOLOGIK FILTER 4362 706 540 2207 15.502950% 6 16928
UKRAINIAN MORFOLOGIK DIRECT 4362 706 540 2207 15.502950% 6 16928
Radixor 608 880 610 190 1.334645% 2 14426

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: UK_UA
  • 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