基于双通道LSTM模型的用户性别分类方法研究
|
摘要
微博用户性别分类旨在根据用户信息进行用户性别的识别。目前性别分类的相关研究主要针对单一类型的特征(文本特征或者社交特征)进行性别分类。与以往研究不同,文中提出了一种双通道LSTM(Long-Short Term Memory)模型,以充分结合文本特征(用户发表的微博文本)和社交特征(用户关注者的信息)进行用户性别分类方法的研究。首先,利用单通道LSTM模型分别学习两组文本特征,得到两种特征表示;然后,在神经网络中加入Merge层,结合两种特征表示进行集成学习,以充分学习文本特征和社交特征之间的联系。实验结果表明,相对于传统的分类算法,双通道LSTM模型分类算法能够获得更好的用户性别分类效果。
User gender classification aims at classifying the users into male and female with the provided information.Previous studies on gender classification mainly focus on a single type of features(i.e.,textual features or social features).Different from previous research,this paper proposed a new approach named dual-channel LSTM by making full use of the relationship between textual features(the text which user publishes)and social features(the followers which user concerns).Specifically,this paper first got two kinds of features using single-channel LSTM respectively.Then,it proposed a joint learning method to integrate the features.Lastly,it got the final classification results by the dual-channel LSTM.Empirical studies show that the dual-channel LSTM model achieves effective results for gender classification compared with traditional classification algorithms.
User gender classification aims at classifying the users into male and female with the provided information.Previous studies on gender classification mainly focus on a single type of features(i.e.,textual features or social features).Different from previous research,this paper proposed a new approach named dual-channel LSTM by making full use of the relationship between textual features(the text which user publishes)and social features(the followers which user concerns).Specifically,this paper first got two kinds of features using single-channel LSTM respectively.Then,it proposed a joint learning method to integrate the features.Lastly,it got the final classification results by the dual-channel LSTM.Empirical studies show that the dual-channel LSTM model achieves effective results for gender classification compared with traditional classification algorithms.
引文
[1]WEN K M,XU S,LI R X,et al.Survey of Microblog and Chinese Microblog Information Processing[J].Journal of Chinese Information Processing,2012,26(6):28-36.(in Chinese)文坤梅,徐帅,李瑞轩,等.微博及中文微博信息处理研究综述[J].中文信息学报,2012,26(6):28-36.
[2]ZHANG J F,XIA Y Q,YAO J M.A Review towards Microtext Processing[J].Journal of Chinese Information Processing,2012,26(4):21-27.(in Chinese)张剑锋,夏云庆,姚建民.微博文本处理研究综述[J].中文信息学报,2012,26(4):21-27.
[3]WANG J J,LI S S,HUANG L.User Gender Classification in Chinese Microblog[J].Journal of Chinese Information Processing,2014,28(6):150-155.(in Chinese)王晶晶,李寿山,黄磊.中文微博用户性别分类方法研究[J].中文信息学报,2014,28(6):150-155.
[4]DICKINSON M B,HU W.Gender Prediction on Twitter Using Stream Algorithms with N-Gram Character Features[J].Proceedings of International Journal of Intelligences Science,2012,2(4):143-148.
[5]MORGAN M S,DEREK R.Gender Inference of Twitter Users in Non-English Contexts[C]∥Proceedings of EMNLP.2013:1136-1145.
[6]GONCALVES C B,RATIKIEWICZ J,FLAMMINI A,et al.Predicting the political alignment of Twitter user[C]∥Proceedings of the International Conference on Social Computing.2011.
[7]LIU,RUTHS D.What’s in a name?Using first names as features for gender inference in Twitter[C]∥Analyzing Microtext:2013AAAI Spring Symposium.2013.
[8]EICHSTAEDT M C,KERN L,et al.Developing Age and Gender Predictive Lexica over Social Media[C]∥Proceedings of EMNLP.2014:1146-1151.
[9]FARNADI M G,VASUDEVAN G,DAVALOS S,et al.Age and gender identification in social media[C]∥Proceedings of CLEF 2014Evaluation Labs pages.2014:1129-1136.
[10]HOCHREITER,JURGEN S.Long Short-Term Memory[J].Neural Computation,1997,9(8):1735-1780.
[11]GRAVES A.Generating Sequences With Recurrent Neural Networks[J].arXiv preprint arXiv:1308.0850,2013.
[12]ANTOINE X B,YOSHUA B.Deep Sparse Rectifier Neural Networks[C]∥Proceedings of AISTATS.2011:315-323.
[13]HINTON G E,SRIVASTAVA N,KRIZHEVSKY A,et al.Improving Neural Networks by Preventing Co-adaptation of Feature Detectors[J].Computer Science,2012,3(4):212-223.
1)http://mallet.cs.umass.edu/
2)https://code.google.com/p/fudannlp/
[2]ZHANG J F,XIA Y Q,YAO J M.A Review towards Microtext Processing[J].Journal of Chinese Information Processing,2012,26(4):21-27.(in Chinese)张剑锋,夏云庆,姚建民.微博文本处理研究综述[J].中文信息学报,2012,26(4):21-27.
[3]WANG J J,LI S S,HUANG L.User Gender Classification in Chinese Microblog[J].Journal of Chinese Information Processing,2014,28(6):150-155.(in Chinese)王晶晶,李寿山,黄磊.中文微博用户性别分类方法研究[J].中文信息学报,2014,28(6):150-155.
[4]DICKINSON M B,HU W.Gender Prediction on Twitter Using Stream Algorithms with N-Gram Character Features[J].Proceedings of International Journal of Intelligences Science,2012,2(4):143-148.
[5]MORGAN M S,DEREK R.Gender Inference of Twitter Users in Non-English Contexts[C]∥Proceedings of EMNLP.2013:1136-1145.
[6]GONCALVES C B,RATIKIEWICZ J,FLAMMINI A,et al.Predicting the political alignment of Twitter user[C]∥Proceedings of the International Conference on Social Computing.2011.
[7]LIU,RUTHS D.What’s in a name?Using first names as features for gender inference in Twitter[C]∥Analyzing Microtext:2013AAAI Spring Symposium.2013.
[8]EICHSTAEDT M C,KERN L,et al.Developing Age and Gender Predictive Lexica over Social Media[C]∥Proceedings of EMNLP.2014:1146-1151.
[9]FARNADI M G,VASUDEVAN G,DAVALOS S,et al.Age and gender identification in social media[C]∥Proceedings of CLEF 2014Evaluation Labs pages.2014:1129-1136.
[10]HOCHREITER,JURGEN S.Long Short-Term Memory[J].Neural Computation,1997,9(8):1735-1780.
[11]GRAVES A.Generating Sequences With Recurrent Neural Networks[J].arXiv preprint arXiv:1308.0850,2013.
[12]ANTOINE X B,YOSHUA B.Deep Sparse Rectifier Neural Networks[C]∥Proceedings of AISTATS.2011:315-323.
[13]HINTON G E,SRIVASTAVA N,KRIZHEVSKY A,et al.Improving Neural Networks by Preventing Co-adaptation of Feature Detectors[J].Computer Science,2012,3(4):212-223.
1)http://mallet.cs.umass.edu/
2)https://code.google.com/p/fudannlp/