Onto.PT: Towards the Automatic Construction of a Lexical Ontology ...
Onto.PT: Towards the Automatic Construction of a Lexical Ontology ...
Onto.PT: Towards the Automatic Construction of a Lexical Ontology ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
122 Chapter 7. Moving from term-based to synset-based relations<br />
tb-triple = (documento hypernym-<strong>of</strong> recibo)<br />
(document hypernym-<strong>of</strong> receipt)<br />
A1: documento, declaração B1: recibo, comprovante, nota, quitação,<br />
senha<br />
A2: escritura, documento<br />
plausible sb-triples = {A1, B1}<br />
tb-triple = (planta part-<strong>of</strong> floresta)<br />
(plant part-<strong>of</strong> forest)<br />
A1: relação, quadro, planta, mapa B1: bosque, floresta, mata, brenha, selva<br />
A2: vegetal, planta<br />
A3: traçado, desenho, projeto, planta, plano<br />
plausible sb-triples = {A2, B1}<br />
tb-triple = (passageiro purpose-<strong>of</strong> carruagem)<br />
(passenger purpose-<strong>of</strong> carriage)<br />
A1: passageiro, viajante B1: carriagem, carruagem, carraria<br />
A2: passageiro, viador B2: carruagem, carro, sege, coche<br />
A3: passageiro, transeunte B3: carruagem, caleça, caleche<br />
B4: actividade, carruagem, operosidade,<br />
diligência<br />
plausible sb-triples = {A1, B1}, {A1, B2}, {A1, B3}, {A2, B1}, {A2, B2}, {A2, B3}<br />
tb-triple = (máquina hypernym-<strong>of</strong> câmara)<br />
(machine hypernym-<strong>of</strong> camera)<br />
A1: motor, máquina B1: câmara, parlamento, assembleia, assembléia<br />
B2: quarto, repartimento, apartamento,<br />
câmara, compartimento, aposento, recâmara,<br />
alcova<br />
plausible sb-triples = {}<br />
Figure 7.5: Example <strong>of</strong> gold entries.<br />
<strong>of</strong> using only <strong>the</strong> 452 tb-triples as a lexical network, we used all <strong>the</strong> tb-triples<br />
in CARTÃO (see section 4). After comparing <strong>the</strong> automatic attachments with<br />
<strong>the</strong> attachments in <strong>the</strong> gold reference, we computed typical information retrieval<br />
measures, including precision, recall and three variations <strong>of</strong> <strong>the</strong> F -score: F1 is <strong>the</strong><br />
classic, F0.5 favors precision, and RF1 uses a relaxed recall (RelRecall), instead <strong>of</strong><br />
<strong>the</strong> classic recall – RelRecall is 1 if at least one correct attachment is selected. For<br />
a tb-triple in <strong>the</strong> set <strong>of</strong> tb-triples to ontologise, ti ∈ T , <strong>the</strong>se measures are computed<br />
as follows:<br />
P recisioni = |<strong>Automatic</strong>Attachmentsi ∩ GoldAttachmentsi|<br />
|<strong>Automatic</strong>Attachmentsi|<br />
Recalli = |<strong>Automatic</strong>Attachmentsi ∩ GoldAttachmentsi|<br />
|GoldAttachmentsi|<br />
P recision = 1<br />
|T |<br />
Recall = 1<br />
|T |<br />
<br />
1, if |<strong>Automatic</strong>Attachmentsi ∩ GoldAttachmentsi| > 0<br />
RelRecalli =<br />
0, o<strong>the</strong>rwise<br />
RelRecall = 1<br />
|T |<br />
|T |<br />
<br />
RelRecalli<br />
i=1<br />
|T |<br />
|T |<br />
<br />
P recisioni<br />
i=1<br />
<br />
Recalli<br />
i=1<br />
Fβ = (1 + β 2 <br />
P recision × Recall<br />
) ×<br />
(β2 <br />
× P recision) + Recall