融合遗传算法的特定领域情感词库构建

Scheme for constructing domain-specific sentiment lexicon using genetic algorithm

为提高情感词库在特定领域情感分析的性能,针对情感词的强度和极性随着领域不同而变化的问题,采用遗传算法构建特定领域专用的情感词库。提出了基于遗传算法的情感词库构建框架,将词库预测特定领域文本情感趋向的准确率作为优化目标,并不断对情感词分值进行调整。利用遗传算法强大的搜索能力,实现对情感词分值的调整,结合情感词对文本的影响,设计并改进了变异策略以提升情感分类的准确率。设计了精英策略以提升算法的收敛速度。通过在中文和英文评论数据集上的对比实验表明,相较于已有的情感词库,构建的词库在特定领域文本情感分类的准确率和F1值都在80%以上,具有明显优势,证明了方法的有效性。该方法构建的情感词库在特定领域具有良好的性能,有效提升了情感词的覆盖率,能很好地扩展到其他领域。

To improve the performance of sentiment vocabulary in sentiment analysis in a specific field,we propose a domain-specific sentiment lexicon construction method in this paper.The method,based on genetic algorithm,can address the problem that strength and polarity of sentiment words vary from domains.First,using the genetic algorithm,we propose a framework for constructing sentiment lexicons,which takes the accuracy of a lexicon in predicting the emotional trend of texts in a specific field as the optimization objective.And the genetic algorithm is used to continuously adjust the values of sentiment words according to the sentiment classification accuracy.Then,the powerful search ability of genetic algorithm is used to adjust the values of sentiment words.Furthermore,a mutation strategy is designed to take the influence of sentiment words to the polarity of texts into consideration,which improves the classification accuracy.Finally,an elite strategy is designed to improve the convergence speed of the algorithm.The compared experiments are conducted on the Chinese and English comment datasets to verify the effectiveness of the proposed method.Compared with the existing sentiment lexicons,the sentiment lexicon built by the proposed method has obvious advantage in the accuracy and F1-Measure of the text sentiment classification.They are all more than 80%.The sentiment lexicon constructed by the method of this paper has good sentiment classification performance in specific domains,which effectively improves the coverage of sentiment words and can be well extended to other domains.