UPSI Digital Repository (UDRep)
|
UPSI Digital Repository (UDRep)
|
|
|
|
||||||||||||||||||||||||
| Abstract : Perpustakaan Tuanku Bainun |
| Sentiment Analysis (SA) field relies on lexicons to analyze people's perceptions. Therefore, evaluating and benchmarking lexicons is crucial for SA development, falling under Multi-Criteria Decision-Making (MCDM). The Fuzzy-Decision-by-Opinion-Score Method (FDOSM) and Fuzzy-Weighted-Zero-Inconsistency (FWZIC) techniques are novel MCDM extensions in a fuzzy-environment that address complexity-problems. Hence, this study aims to use these techniques to evaluate and benchmark dialectal lexicons. The methodology for this research is presented in four phases. The first phase includes developing four main criteria, along with its 28 sub-criteria for lexicon evaluation; the second phase proposes a multidimensional decision-matrix based on evaluation criteria and available lexicons; in the third phase, the multidimensional evaluation and benchmarking results of MCDM are obtained by FDOSM and are achieved by FWZIC; in the fourth phase, the validation was conducted based on the fusion of Systematic Ranking and Sensitivity Analysis to test the weight effect. The findings reveal that variations in weighting can significantly impact sensitivity analysis outcomes, i.e., SUAR (Lex18) ranked (4th) under the original scenario but fell to (5th) in scenario 6. Meanwhile, PADIC (Lex17) ranked (21st) under the original and scenarios 1, 2, 3, and 5, then rose to (20th) ranking position in scenarios 4 and 6. Despite these variations, some lexicons consistently maintained their ranking, i.e., SANAD (Lex21) the (24th), and TSAC (Lex12) the (28th) both maintained their ranking position across all sensitivity analysis interchanged scenarios, indicating their robustness. This emphasizes the reliability of the evaluation process from expert voting to sensitivity analysis of the four main criteria: Labelled-Data, Labelling-Type, Labelling-Techniques, Labelling-Targets, and its derived 28 sub-criteria. The study highlights the significance of addressing issues such as data variation and measuring importance when developing Arabic SA lexicons. The MCDM approach, particularly using FWZIC and FDOSM, proved to be more accurate in handling uncertainty and vagueness compared to other methods. |
| This material may be protected under Copyright Act which governs the making of photocopies or reproductions of copyrighted materials. You may use the digitized material for private study, scholarship, or research. |