基于条件随机域模型的比较要素抽取研究

随着主观性评价文本数量的不断增长, 文本情感分析已经成为众多研究者关注的对象. 比较要素抽取是比较句情感分析的重要研究任务之一, 比较句的情感分析结果与比较要素相结合才更有意义. 为了提高比较要素抽取的性能, 本文提出在构建系统模型的过程中引入浅层句法信息、比较词候选信息和启发式位置信息等多种语言学相关特征, 并且在不增加领域知识的情况下, 有效提高系统的准确率和F1值, 同时本文提出的方法可以有效处理含有多个比较关系的句子. 实验结果表明, 将本文提出的特征应用于条件随机域 (Conditional random fields, CRFs)模型可以有效提高比较要素抽取的各项性能指标, 同时, 将本文的实验结果与2012 年中文情感分析评测结果的最大值进行了比较, 各项指标均超过最大值, 进一步证明了本文方法的有效性.

Extraction of Comparative Elements Using Conditional Random Fields

With the rapid growth of the number of evaluative texts on the Web, sentiment analysis has attracted the attention of researchers all over the world. Extraction of comparative elements is one of the important tasks for sentiment analysis of comparative sentences. It is more meaningful that results of sentiment analysis combine with comparative elements. To improve the performance of comparative elements extraction, this paper proposes to introduce shallow parsing features, comparative word candidates and heuristic position information to conditional random fields (CRFs) for building a system model. The proposed method is not only free from introducing domain knowledge but also can effectively deal with sentences containing a few comparative relationships. Experiment results show that the performance of system is improved when introducing proposed features to the CRFs model. Meanwhile, compared with the best results of the 2012 Chinese opinion analysis evaluation, the F1-scores of the proposed method are higher than the maximum value.