Conference (or journal) review forms represent a particular kind of sentiment-bearing documents, which has been to date left largely unexplored. We set up a review corpus including a subset or a whole of anonymized conference reviews from four conferences. Furthermore, by a thorough manual analysis of the review forms and guidelines of eleven conferences we identified a set of generic review criteria, which we used as aspects in aspect-based sentiment analysis (SA) of the review form texts. A sentiment lexicon was created specifically for the domain of conference paper reviews. The first step of the lexicon construction method was the manual definition of a two-level taxonomy. In the second phase, noun phrases and adjectives frequently appearing in the given portions of the review were automatically collected. The results of the lexicon-based SA method were compared with the numerical scores from the reviews. The precision of criterion identification was evaluated at 57.38% and the recall at 53.44%; the sentiment polarity was correct in over 75% of cases. This is an improvement on the result of a similar sentiment analysis carried out in a comparable study. © 2021 Copyright for this paper by its author

Aspect-based sentiment analysis of conference review forms with LD-enabled review criteria

Chiara Ghidini
2021

Abstract

Conference (or journal) review forms represent a particular kind of sentiment-bearing documents, which has been to date left largely unexplored. We set up a review corpus including a subset or a whole of anonymized conference reviews from four conferences. Furthermore, by a thorough manual analysis of the review forms and guidelines of eleven conferences we identified a set of generic review criteria, which we used as aspects in aspect-based sentiment analysis (SA) of the review form texts. A sentiment lexicon was created specifically for the domain of conference paper reviews. The first step of the lexicon construction method was the manual definition of a two-level taxonomy. In the second phase, noun phrases and adjectives frequently appearing in the given portions of the review were automatically collected. The results of the lexicon-based SA method were compared with the numerical scores from the reviews. The precision of criterion identification was evaluated at 57.38% and the recall at 53.44%; the sentiment polarity was correct in over 75% of cases. This is an improvement on the result of a similar sentiment analysis carried out in a comparable study. © 2021 Copyright for this paper by its author
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11582/333608
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact