Títol: | Semi-supervised vs. cross-domain graph-based learning for sentiment classification |
Importa'l al teu calendari: |
---|---|---|
Tipus: | Ponència | |
Per: | Dra. Natalia Ponomareva | |
Lloc: | Claude Shannon | |
Dia/hora: | 12.30 24/02/2016 | |
Duració aproximada: | 1:30 hores | |
Persona de contacte: | Gómez Soriano, José Manuel ( ) | |
Resum: | Hola a todos: Os mando esta noticia para recordaros el seminario que dará Natalia Ponomareva, de la Universidad de Wolverhampton este miércoles a las 12.30 pero, además, para corregir el lugar, que será en la Claude Shannon. La descripción de la charla es: Sentiment analysis has been intensively researched in the last ten years, but there are still many issues to be addressed. One of the problems is the lack of labelled data to carry out precise supervised sentiment classification. In response, research has moved towards developing semi-supervised and cross-domain techniques. In this talk I analyse and compare both techniques in the graph-based framework. First, I present one of the most popular and widely used graph-based algorithm called label propagation together with its modifications. Second, a sentiment similarity metric used for constructing sentiment graphs is introduced. Finally, I compare the performance of graph-based algorithms in cross-domain and semi-supervised settings and propose recommendations for selecting the most pertinent learning approach given the data available. The recommendations are based on two domain characteristics, domain similarity and domain complexity, which have a significant impact on semi-supervised and cross-domain performance. |
[ Tancar ]