中文反语识别特征分析

doi:10.6040/j.issn.1672-3961.0.2018.341

摘要/Abstract

摘要：

以微博数据为研究对象，对反语识别特征进行研究。针对微博特点和反语识别特性，构建包括情感短语、表情符号等在内的多种特征。试验表明，在不平衡数据集上反语特征的识别准确率、召回率和F值等评价标准分别比现有反语特征分别提高了0.34%、0.74%和0.18%，而在平衡数据集上反语特征的识别准确率、召回率和F值则分别提高了0.44%、2.54%和0.14%。

关键词: 反语识别, 情感分析, 特征构建, 不平衡数据集, 平衡数据集

Abstract:

The research object was data in microblog. The features of irony detection were studied. In view of the characteristics of microblog and irony detection, a variety of features were constructed, such as emotional phrases, emoticons and so on. The experiments showed that the proposed irony features improved 0.34% on recognition accuracy, 0.74% on recall and 0.18% on F-measure, compared with the existing ones for the imbalanced datasets. The proposed irony features also improved 0.44% on recognition accuracy, 2.54% on recall and 0.14% on F-measure, compared with the existing ones for the balanced datasets.

Key words: irony detection, sentiment analysis, features construction, imbalanced dataset, balanced dataset

中图分类号:

TP391

周荣翔,贾修一. 中文反语识别特征分析[J]. 山东大学学报 (工学版), 2019, 49(1): 41-46.

Rongxiang ZHOU,Xiuyi JIA. Features analysis for Chinese irony detection[J]. Journal of Shandong University(Engineering Science), 2019, 49(1): 41-46.

图/表 11

表1

表2

表3

表4

表5

表6

表7

表8

图1

图2

图3

参考文献 18

1	WANG M, CAO D, LI L, et al. Microblog sentiment analysis based on cross-media bag-of-words model[C]//International Conference on Internet Multimedia Computing and Service. Xiamen: ACM, 2014: 76.
2	JIANG F, LIU Y, LUAN H, et al. Microblog sentiment analysis with emoticon space model[C]//Chinese National Conference on Social Media Processing. Berlin, Germany: Springer, 2014: 76-87.
3	OUG, CHEN W, LI B, et al. Clusm: an unsupervised model for microblog sentiment analysis incorporating link information[C]// International Conference on Database Systems for Advanced Applications. Bali, Indonesia: Springer, 2014: 481-494.
4	KAROUI J, FARAH B, MORICEAU V, et al. Towards a contextual pragmatic model to detect irony in tweets[C]//Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing. Beijing: ACL, 2018: 644-650.
5	BRUNTSCH R , RUCH W . Studying irony detection beyond ironic criticism: let′s include ironic praise[J]. Frontiers in Psychology, 2017, 8, 606. doi: 10.3389/fpsyg.2017.00606
6	TASLIOGLU H, KARAGOZ P. Irony detection on microposts with limited set of features[C]//Proceedings of the Symposium on Applied Computing. Marrakech, Morocco: ACM, 2017: 1076-1081.
7	SAVOV P, NIELEK R. Ridiculously expensive watches and surprisingly many reviewers: a study of irony[C]//International Conference on Web Intelligence. Omaha, USA: IEEE, 2017: 725-729.
8	刘正光. 反语理论综述[J]. 解放军外国语学院学报, 2002, 22 (4): 16- 18. doi: 10.3969/j.issn.1002-722X.2002.04.004
	LIU Zhengguang . A critique of irony theories[J]. Journal of PLA University of Foreign Language, 2002, 22 (4): 16- 18. doi: 10.3969/j.issn.1002-722X.2002.04.004
9	TANG Y J, CHEN H. Chinese irony corpus construction and ironic structure analysis[C]//Proceedings of the 25th International Conference on Computational Lingustics. Dublin, Ireland: ACL, 2014: 1269-1278.
10	邓钊, 贾修一, 陈家骏. 面向微博的中文反语识别研究[J]. 计算机工程与科学, 2015, 37 (12): 2312- 2317. doi: 10.3969/j.issn.1007-130X.2015.12.018
	DENG Zhao , JIA Xiuyi , CHEN Jiajun . A survey on chinese ironic detection in microblog[J]. Computer Engineering and Science, 2015, 37 (12): 2312- 2317. doi: 10.3969/j.issn.1007-130X.2015.12.018
11	WU C H, WU F Z, WU S X, et al. THU NGN at semeval-2018 task 3: tweet irony detection with densely connected lstm and multi-task learning[C]//Proceedings of the 12th International Workshop on Semantic Evaluation. New Orleans, USA: ACL, 2018: 51-56.
12	邢竹天, 徐扬. 面向网络文本的汉语反讽修辞识别方法研究[J]. 山西大学学报(自然科学版), 2015, 38 (3): 385- 391.
	XIN Zhutian , XU Yang . A study of chinese sarcasm detection methods on internet texts[J]. Journal of Shanxi University(Natural Science Edition), 2015, 38 (3): 385- 391.
13	CHARALAMPAKIS B, SPATHIS D, KOUSLIS E, et al. Detecting irony ongreek political tweets: a text mining approach[C]// In Proceedings of the 16th International Conference on Engineering Applications of Neural Networks. New York, USA: ACM, 2015, 17: 1-5.
14	REYESA, ROSSO P. Mining subjective knowledge from customer reviews: a specific case of irony detection[C]// Proceedings of the 2nd Workshop on Computational Approaches to Subjectivity and Sentiment Analysis. Portland, USA: ACL, 2011: 118-124.
15	BOUAZIZIM, OHTSUKI T. Sarcasm detection in twitter[C]// Global Communications Conference. San Diego, USA: IEEE, 2016: 1-6.
16	RAVIK , RAVI V . A novel automatic satire and irony detection using ensembled feature selection and data mining[J]. Knowledge-Based Systems, 2016, 120, 15- 33.
17	CARVALHOP. Clues for detecting irony in user-generated contents: oh…!! it′s "so easy"; -[C]// International CIKM Workshop on Topic-Sentiment Analysis for MASS Opinion. New York, USA: ACM, 2009: 53-56.
18	吕叔湘. 中国文法要略[M]. 北京: 商务印书馆, 1982.

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed

类别	程度副词
极量	分外，十分，备加，万分，倍加，异常，尤，尤为，尤其，最，最为，无比，极，极为，极其，极度，极端，格外，殊，深为，特，特别，甚，甚为，绝，绝伦，绝对，绝顶，至，至为，透，透顶，顶，顶顶，非常
高量	大，大为，大大，太，好，好不，好生，很，忒，怪，挺，愈，愈为，愈加，愈发，愈益，更，更为，更其，更加，多，多么，比较，益，益发，相当，较，较为，较比，越，越加，越发，过，过于，过分，颇，颇为，蛮，老
较中量	何其，何等，够，尽，全然，满，还，真
中量	几，几乎，差不多，差点儿
较低量	不大，不太，不很，不甚，不胜
低量	多少，微微，丝毫，些微，有些，有点，有点儿，略，略为，略微，略略，毫，稍，稍为，稍微，稍稍，稍许

类型	例子
n+q	凯歌阵阵
q+n	阵阵凯歌
a+(uj)+n	美丽(的)花,最好(的)恭维
b+n	中等身材
n+u	暴风雨般，绅士似的
ad+(uj)+vn	迅速发展，认真(的)咨询
eng+n	AA制
d+n	真是棒极了，真是调皮
d+v	互相支援，没法接受，强烈反对
a+(ul)+n	活跃气氛，学好功课，热爱家乡，拉仇恨，支持魅族，抛弃(了)魅族，没牛人
v+a	洗刷干净，喜欢清净
a+(ud)+v	关心(得)不够，油价上升，经济复苏
v+(ud)+d	高兴(得)很
v+r	鼓励他
ad+v	正确领导
v+(ud)+l	吃(得)很饱
v+m	有一套
a+(ud)+d	痛快极了，漂亮(的)很，好(得)很，标致极了
d+a	非常漂亮，比较合适，非常谦虚，比较空，有点特殊，过分良好
a+v	富裕起来
n+(v)+a	态度和蔼，情节较重，外形好，胆子(可)真大
v+(ul)+b	看(了)高兴
m+a	十分壮丽
m+f	100以上
n+f	天亮以后，尖子生里面

分类器	朴素贝叶斯	逻辑斯蒂回归	支持向量机	决策树	随机森林
一元文法特征	0.445	0.404	0.690	0.642	0.610
文献[10]的反语特征	0.494	0.257	0.728	0.687	0.798
9个反语特征	0.505	0.244	0.729	0.680	0.823

分类器	朴素贝叶斯	逻辑斯蒂回归	支持向量机	决策树	随机森林
一元文法特征	0.509	0.467	0.393	0.233	0.300
文献[10]的反语特征	0.560	0.513	0.489	0.339	0.321
9个反语特征	0.586	0.521	0.502	0.339	0.311

分类器	朴素贝叶斯	逻辑斯蒂回归	支持向量机	决策树	随机森林
一元文法特征	0.472	0.434	0.513	0.348	0.431
文献[10]的反语特征	0.525	0.342	0.585	0.453	0.472
9个反语特征	0.543	0.332	0.595	0.452	0.464