基于结构对应学习的跨语言情感分类研究
发布时间：2018-11-03 08:22
【摘要】：情感分类的主要目的是预测用户在互联网中发布情绪数据的极性(积极的或者消极的),各种语言的情感分析已经成为诸多应用的研究热点,然而由于不同语言的情感资源在质量和数量上的不平衡,通常使用源语言来改善目标语言的跨语言情感分类方法,来提高目标语言情感分类的准确性.传统的跨语言情感分类主要是通过机器翻译将目标语言映射到源语言中,但是分类的准确性严重受到机器翻译质量的影响.通过对跨领域文本分类的结构学习算法(SCL)的讨论和拉普拉斯映射对两种语言之间词对的影响,对跨语言结构对应学习算法(CLSCL)的改进,进而提出M-CLSCL算法,借助选出来的轴心词对来进行目标语言的情感分类,通过M-CLSCL方法与前述相关方法的实验结果进行比较,可以发现M-CLSCL提高了情感分类的准确性.
[Abstract]:The main purpose of affective classification is to predict the polarity (positive or negative) of emotional data published by users on the Internet. Emotional analysis of various languages has become a hot research topic in many applications. However, due to the imbalance between the quality and quantity of emotional resources in different languages, the source language is usually used to improve the cross-language affective classification method of the target language to improve the accuracy of the target language affective classification. The traditional cross-language affective classification mainly uses machine translation to map the target language to the source language, but the accuracy of classification is seriously affected by the quality of machine translation. By discussing the structure learning algorithm (SCL) of cross-domain text classification and the effect of Laplace mapping on the word pairs between two languages, the paper improves the cross-language structure corresponding learning algorithm (CLSCL), and then proposes the M-CLSCL algorithm. With the help of the selected axis word pairs to carry on the emotion classification of the target language, we can find that M-CLSCL improves the accuracy of the emotion classification by comparing the experimental results of the M-CLSCL method with the previous related methods.
【作者单位】： 河北工业大学计算机科学与软件学院;河北省大数据计算重点实验室(河北工业大学);
【基金】：天津市自然科学基金(14JCYBJC18500) 天津市应用基础与前沿技术研究计划(13JCQNJC00200)
【分类号】：TP391.1
，
本文编号：2307240
[Abstract]:The main purpose of affective classification is to predict the polarity (positive or negative) of emotional data published by users on the Internet. Emotional analysis of various languages has become a hot research topic in many applications. However, due to the imbalance between the quality and quantity of emotional resources in different languages, the source language is usually used to improve the cross-language affective classification method of the target language to improve the accuracy of the target language affective classification. The traditional cross-language affective classification mainly uses machine translation to map the target language to the source language, but the accuracy of classification is seriously affected by the quality of machine translation. By discussing the structure learning algorithm (SCL) of cross-domain text classification and the effect of Laplace mapping on the word pairs between two languages, the paper improves the cross-language structure corresponding learning algorithm (CLSCL), and then proposes the M-CLSCL algorithm. With the help of the selected axis word pairs to carry on the emotion classification of the target language, we can find that M-CLSCL improves the accuracy of the emotion classification by comparing the experimental results of the M-CLSCL method with the previous related methods.
【作者单位】： 河北工业大学计算机科学与软件学院;河北省大数据计算重点实验室(河北工业大学);
【基金】：天津市自然科学基金(14JCYBJC18500) 天津市应用基础与前沿技术研究计划(13JCQNJC00200)
【分类号】：TP391.1
，
本文编号：2307240
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2307240.html
