Table 5 Data and evaluation methods for Portuguese
From: A review on Relation Extraction with an eye on Portuguese
| References | Data/corpora | Data size | Method | Evaluation | Performance (%) |
|---|---|---|---|---|---|
| Brucksen et al. [11] | HAREM/ ReRelEM Golden Collection | 4,417 words | Set of heuristics based on morphosyntactic and semantic information | Golden Standard annotated manually | All relations F\(=\) 36 % |
| Cardoso [15] | HAREM/ ReRelEM Golden Collection | 4,417 words | Set of grammar rules | Golden Standard annotated manually | All relations F\(=\) 45 % |
| Chaves [19] | HAREM/ ReRelEM Golden Collection | 4,417 words | Set of grammar rules | Golden Standard annotated manually | All relations F\(=\) 27 % |
| Xavier and de Lima [101] | Tourism category from Wikipedia | – | Semi-automatic method based on structure from Wikipedia and syntactic heuristics | Golden Standard the domain of Tourism | F\(=\) 85 % |
| Santos et al. [85] | Biographies texts from Wikipedia, CETEMPblico corpus | CETEMPblico \(=\) 110 sentences | Rule-base approach | Manual evaluation of the family relations | Wikipedia F\(=\) 29 % CETEMPblico F\(=\) 36 % |
| Ferreira et al. [39] | MedAlert corpus | 2,724,860 tokens | REMMA system | MedAlert Golden Standard composed by 20 texts annotated manually | Inclusion F\(=\) 89 % |
| Tanev et al. [97] | News articles for Portuguese about security and disaster-related topics | News articles = 3.4 million titles, disaster-related articles \(=\) 100 (April 2009) | Ontopopulis system | Comparative evaluation between Baseline Portuguese and the results | Dead F\(=\) 69 %, Wounded F\(=\) 51 %, Kidnapeed F\(=\) 67 %, Arrested F\(=\) 47 % |
| Fernandes et al. [38] | GLOBO QUOTES from Globo.com | Around 13.5 million tokens | Entropy Guided Transformation Learning | Baseline system manually constructed | Quotation-Author F\(=\) 79.02 % |