Papers › UCPhrase: Unsupervised Context-aware Quality Phrase Tagging

UCPhrase: Unsupervised Context-aware Quality Phrase Tagging

28 May 2021arXiv:2105.14078archive 2025-07-28

Xiaotao Gu, Zihan Wang, Zhenyu Bi, Yu Meng, Liyuan Liu, Jiawei Han, Jingbo Shang

Identifying and understanding quality phrases from context is a fundamental task in text mining. The most challenging part of this task arguably lies in uncommon, emerging, and domain-specific phrases. The infrequent nature of these phrases significantly hurts the performance of phrase mining methods that rely on sufficient phrase occurrences in the input corpus. Context-aware tagging models, though not restricted by frequency, heavily rely on domain experts for either massive sentence-level gold labels or handcrafted gazetteers. In this work, we propose UCPhrase, a novel unsupervised context-aware quality phrase tagger. Specifically, we induce high-quality phrase spans as silver labels from consistently co-occurring word sequences within each document. Compared with typical context-agnostic distant supervision based on existing knowledge bases (KBs), our silver labels root deeply in the input domain and context, thus having unique advantages in preserving contextual completeness and capturing emerging, out-of-KB phrases. Training a conventional neural tagger based on silver labels usually faces the risk of overfitting phrase surface names. Alternatively, we observe that the contextualized attention maps generated from a transformer-based neural language model effectively reveal the connections between words in a surface-agnostic way. Therefore, we pair such attention maps with the silver labels to train a lightweight span prediction model, which can be applied to new input to recognize (unseen) quality phrases regardless of their surface names or frequency. Thorough experiments on various tasks and datasets, including corpus-level phrase ranking, document-level keyphrase extraction, and sentence-level phrase tagging, demonstrate the superiority of our design over state-of-the-art pre-trained, unsupervised, and distantly supervised methods.

PaperPDFCode

Code

xgeric/UCPhrase-exp officialmentioned in papermentioned on GitHubpytorch report
xgeric/UCPhrase-reproduce officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Keyphrase ExtractionLanguage ModellingPhrase RankingPhrase TaggingSentence

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Keyphrase Extraction KP20k Wiki+RoBERTa F1@10 19.2 #1 of 7 Archive leaderboard report
Keyphrase Extraction KP20k Wiki+RoBERTa Recall 73.0 #1 of 7 Archive leaderboard report
Keyphrase Extraction KP20k UCPhrase F1@10 19.7 #2 of 7 Archive leaderboard report
Keyphrase Extraction KP20k UCPhrase Recall 72.9 #2 of 7 Archive leaderboard report
Keyphrase Extraction KP20k AutoPhrase F1@10 18.2 #3 of 7 Archive leaderboard report
Keyphrase Extraction KP20k AutoPhrase Recall 62.9 #3 of 7 Archive leaderboard report
Keyphrase Extraction KP20k Spacy F1@10 15.3 #4 of 7 Archive leaderboard report
Keyphrase Extraction KP20k Spacy Recall 59.5 #4 of 7 Archive leaderboard report
Keyphrase Extraction KP20k PKE F1@10 12.6 #5 of 7 Archive leaderboard report
Keyphrase Extraction KP20k PKE Recall 57.1 #5 of 7 Archive leaderboard report
Keyphrase Extraction KP20k TopMine F1@10 15.0 #6 of 7 Archive leaderboard report
Keyphrase Extraction KP20k TopMine Recall 53.3 #6 of 7 Archive leaderboard report
Keyphrase Extraction KP20k StanfordNLP F1@10 13.9 #7 of 7 Archive leaderboard report
Keyphrase Extraction KP20k StanfordNLP Recall 51.7 #7 of 7 Archive leaderboard report
Keyphrase Extraction KPTimes UCPhrase F1@10 10.9 #1 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes UCPhrase Recall 83.4 #1 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes AutoPhrase F1@10 10.3 #2 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes AutoPhrase Recall 77.8 #2 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes Wiki+RoBERTa F1@10 9.4 #3 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes Wiki+RoBERTa Recall 64.5 #3 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes TopMine F1@10 8.5 #4 of 4 Archive leaderboard report
Keyphrase Extraction KPTimes TopMine Recall 63.4 #4 of 4 Archive leaderboard report
Phrase Ranking KP20k Wiki+RoBERTa P@50K 98.5 #1 of 3 Archive leaderboard report
Phrase Ranking KP20k Wiki+RoBERTa P@5K 100.0 #1 of 3 Archive leaderboard report
Phrase Ranking KP20k UCPhrase P@50K 96.5 #2 of 3 Archive leaderboard report
Phrase Ranking KP20k UCPhrase P@5K 96.5 #2 of 3 Archive leaderboard report
Phrase Ranking KP20k TopMine P@50K 78.0 #3 of 3 Archive leaderboard report
Phrase Ranking KP20k TopMine P@5K 81.5 #3 of 3 Archive leaderboard report
Phrase Ranking KPTimes Wiki+RoBERTa P@50K 96.5 #1 of 4 Archive leaderboard report
Phrase Ranking KPTimes Wiki+RoBERTa P@5K 99.0 #1 of 4 Archive leaderboard report
Phrase Ranking KPTimes UCPhrase P@50K 95.5 #2 of 4 Archive leaderboard report
Phrase Ranking KPTimes UCPhrase P@5K 96.5 #2 of 4 Archive leaderboard report
Phrase Ranking KPTimes AutoPhrase P@50K 95.5 #3 of 4 Archive leaderboard report
Phrase Ranking KPTimes AutoPhrase P@5K 96.5 #3 of 4 Archive leaderboard report
Phrase Ranking KPTimes TopMine P@50K 71.0 #4 of 4 Archive leaderboard report
Phrase Ranking KPTimes TopMine P@5K 85.5 #4 of 4 Archive leaderboard report
Phrase Tagging KP20k UCPhrase F1 73.9 #1 of 4 Archive leaderboard report
Phrase Tagging KP20k UCPhrase Precision 69.9 #1 of 4 Archive leaderboard report
Phrase Tagging KP20k UCPhrase Recall 78.3 #1 of 4 Archive leaderboard report
Phrase Tagging KP20k Wiki+RoBERTa F1 61.0 #2 of 4 Archive leaderboard report
Phrase Tagging KP20k Wiki+RoBERTa Precision 58.1 #2 of 4 Archive leaderboard report
Phrase Tagging KP20k Wiki+RoBERTa Recall 64.2 #2 of 4 Archive leaderboard report
Phrase Tagging KP20k AutoPhrase F1 49.7 #3 of 4 Archive leaderboard report
Phrase Tagging KP20k AutoPhrase Precision 55.2 #3 of 4 Archive leaderboard report
Phrase Tagging KP20k AutoPhrase Recall 45.2 #3 of 4 Archive leaderboard report
Phrase Tagging KP20k TopMine F1 40.6 #4 of 4 Archive leaderboard report
Phrase Tagging KP20k TopMine Precision 39.8 #4 of 4 Archive leaderboard report
Phrase Tagging KP20k TopMine Recall 41.4 #4 of 4 Archive leaderboard report
Phrase Tagging KPTimes UCPhrase F1 73.5 #1 of 4 Archive leaderboard report
Phrase Tagging KPTimes UCPhrase Precision 69.1 #1 of 4 Archive leaderboard report
Phrase Tagging KPTimes UCPhrase Recall 78.9 #1 of 4 Archive leaderboard report
Phrase Tagging KPTimes Wiki+RoBERTa F1 63.2 #2 of 4 Archive leaderboard report
Phrase Tagging KPTimes Wiki+RoBERTa Precision 60.9 #2 of 4 Archive leaderboard report
Phrase Tagging KPTimes Wiki+RoBERTa Recall 65.6 #2 of 4 Archive leaderboard report
Phrase Tagging KPTimes AutoPhrase F1 45.9 #3 of 4 Archive leaderboard report
Phrase Tagging KPTimes AutoPhrase Precision 44.2 #3 of 4 Archive leaderboard report
Phrase Tagging KPTimes AutoPhrase Recall 47.7 #3 of 4 Archive leaderboard report
Phrase Tagging KPTimes TopMine F1 34.0 #4 of 4 Archive leaderboard report
Phrase Tagging KPTimes TopMine Precision 32.0 #4 of 4 Archive leaderboard report
Phrase Tagging KPTimes TopMine Recall 36.3 #4 of 4 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections