Downloaded 36 times
Uploaded byMarieke van Erp
Evaluating Named Entity Recognition and Disambiguation in News and Tweets
The document evaluates named entity recognition (NER) and disambiguation (NED) tools using news and tweets, comparing the performance of ten out-of-the-box tools via the NERD API and NERD-ML. It discusses the challenges of achieving high precision and recall rates, particularly for the 'misc' class, and notes the necessity of larger datasets. The results indicate progress in NER performance, while highlighting ongoing challenges in NED tasks and data standardization.
More Related Content
Similar to Evaluating Named Entity Recognition and Disambiguation in News and Tweets
Evaluating Named Entity Recognition and Disambiguation in News and Tweets
- 1. EVALUATING NAMED ENTITY RECOGNITIONAND DISAMBIGUATION IN NEWS AND TWEETS Giuseppe Rizzo Università degli studi di Torino Marieke van Erp VU University Amsterdam Raphaël Troncy EURECOM
- 2. EVALUATING NER &NED • NER typically an NLP task (MUC, CoNLL, ACE) • NED took flight with availability of large structured resources (Wikipedia, DBpedia, Freebase) • Tools for NER and NED have started popping up outside regular research outlets (TextRazor, DBpedia Spotlight, AlchemyAPI) • Unclear how well these tools perform
- 3. THIS WORK • Evaluation &comparison of 10 out-of-the-box NER and NED tools through NERD API as well as a combination of the tools in NERD-ML • Two types of data: Newswire & Tweets
- 4. • http://nerd.eurecom.fr • Ontology, RESTAPI & UI • Uniform access to 12 different extractors/linkers: AlchemyAPI, DBpedia Spotlight, Extractiv, Lupedia, OpenCalais, Saplo, SemiTags, TextRazor, THD, Wikimeta, Yahoo! Content Analysis, Zemanta
- 5. NERD-ML • The aim ofNERD-ML is to combine the knowledge of the different extractors into a better named entity recogniser • Uses NERD predictions, Stanford NER & extra features • Naive Bayes, k-NN, SMO
- 6. DATA • CoNLL 2003 EnglishNER with AIDA CoNLL-YAGO links to Wikipedia (5,648 NEs/4,485 links in test set) • Making Sense of Microposts 2013 (MSM’13) for NER on Twitter domain + 62 randomly selected tweets from Ritter et al.’s corpus with links to DBpedia resources (MSM: 1,538 NEs/Ritter: 177 links in test set)
- 7. PER LOC ORG Precision Precision MISC OVERALL 100 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 RESULTS NER NEWSWIRE PER LOC ORG Recall Recall MISC OVERALL PER LOC ORG MISC OVERALL F1 F1 AlchemyAPI DBpediaSpotlight Extractiv Lupedia OpenCalais Saplo Textrazor Yahoo Wikimeta Zemanta Stanford NER NERD-ML Run01 NERD-ML Run02 NERD-ML Run03 Upper Limit
- 8. PER LOC ORG Precision Precision MISC OVERALL 100 0 10 20 30 40 50 60 70 80 90 100 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 RESULTS NER MSM PER LOC ORG Precision Recall MISC OVERALL PER LOC ORG MISC OVERALL Precision F1 AlchemyAPI DBpediaSpotlight Extractiv Lupedia OpenCalais Saplo Textrazor Wikimeta Zemanta Ritter et al. Stanford NER NERD-ML Run01 NERD-ML Run02 NERD-ML Run03 Upper Limit
- 9. RESULTS NED DBpedia AlchemyAPI Extractiv LupediaTextrazor Yahoo Zemanta Spotlight AIDAYAGO 70.63 26.93 51.31 57.98 49.21 0.0 35.58 TWEETS 53.85 25.13 74.07 65.38 58.14 76.00 48.57
- 10. DISCUSSION • Still a waysto go, but for certain classes NER is getting close to really good results • MISC class is (and probably always will be?) hard • Bigger datasets needed (for tweets and NED) • NED task can use standardisation
- 11. THANK YOU FORLISTENING • Try out our code at: https://github.com/giusepperizzo/nerdml
- 12. ACKNOWLEDGEMENTS This research isfunded through the LinkedTV and NewsReader projects, both funded by the European Union’s 7th Framework Programme grants GA 287911 and ICT-316404).