摘要
意见目标抽取是自然语言处理领域中意见挖掘研究的重要环节。该文提出了一种基于泛化、繁殖和自举的意见目标抽取方法,在泛化过程中提炼原子意见目标和意见目标模式,在繁殖过程中对复合意见目标进行扩展,并采取自举机制实现了意见目标的递增学习。实验结果显示,经过第一轮自举过程后,该方法的F-1 score指标超出基线方法0.078;自举过程完成后,F-1 score指标提高了0.112。这说明,泛化处理对意见目标充分繁殖意义重大,自举过程则有助于充分发挥泛化能力和繁殖能力。
Opinion target extraction is a key step in opinion mining.A method was developed for opinion target extraction based on generalization,propagation and bootstrapping.The generalization module extracts atom opinion targets and opinion target patterns from the compound opinion targets,the propagation module synthesizes compound opinion targets with a reasoning mechanism,and the bootstrapping module provides multi-cycle incremental learning.Tests show that the F-1 score for this method outperforms the baseline by 0.078 in the first cycle and by 0.112 in the last cycle.Thus,generalization improves the propagation and the bootstrapping helps to maximize the contributions of the generalization and propagation.
出处
《清华大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2009年第S1期1333-1338,共6页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金项目(60703051)
关键词
自然语言处理
意见挖掘
意见目标抽取
文本挖掘
natural language processing
opinion mining
opinion target extraction
text mining