핵심 개념
Advance natural language processing through comparative opinion mining in Vietnamese product reviews.
초록
Overview of the VLSP 2023 - ComOM Shared Task
Task Description
Importance of comparative opinions in product reviews for manufacturers and consumers.
Data Extraction
Constructed dataset with 120 documents, 7427 non-comparative sentences, and 2468 comparisons.
Data Creation
Systematic approach to collect and annotate data from Vietnamese review websites.
Data Statistics
VCOM corpus divided into training, development, and test sets.
Evaluation Metrics
Exact Match, Proportional Match, and Binary Match strategies for evaluation.
Baseline Model
Pipeline and Generative baseline models for comparative opinion extraction.
Results
20 teams participated, with top teams achieving high F1 scores.
Conclusions
Successful completion of the VLSP 2023 - ComOM Shared Task.
통계
참가 팀 수: 48팀
총 제출 횟수: 52회
최고 F1 점수: 23.73%
최고 Precision: 28.62%
최고 Recall: 22.16%
최고 Micro-F1: 29.52%
최고 T4-F1: 31.72%
최고 E-CEE-MACRO-F1: 66.17%
최고 E-CEE-MICRO-F1: 65.45%
인용구
"Product reviews contain valuable information, reflecting users’ perspectives on diverse aspects of products."
"The ComOM shared task is aimed at streamlining and enhancing the crucial analytical process of comparative opinion mining."