Amrita_student at SemEval-2018 Task 1: Distributed Representation of Social Media Text for Affects in Tweets

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 319–323 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics Amrita student at SemEval-2018 Task 1: Distributed Representation of Social Media Text for Affects in Tweets Nidhin A Unnithan, Shalini K., Barathi Ganesh H. B., M. Anand Kumar, K. P. Soman Center for Computational Engineering and Networking (CEN) Amrita School of Engineering, Coimbatore Amrita Vishwa Vidyapeetham, India nidhinkittu5470@gmail.com, shalinikholla@gmail.com, bharathiganesh.hb@gmail.com, m anandkumar@cb.amrita.edu, kp soman@amrita.edu Abstract In this paper we did an analysis of ”Affects in Tweets” which was one of the task con- ducted by SemEval 2018. Task was to build a model which is able to do regression and clas- sification of different emotions from the given tweets data set. We developed a base model for all the subtasks using distributed represen- tation (Doc2Vec) and applied machine learn- ing techniques for classification and regres- sion. Distributed representation is an unsu- pervised algorithm which is capable of learn- ing fixed length feature representation from variable length texts. Machine learning tech- niques used for regression is ’Linear Regres- sion’ while ’Random Forest Tree’ is used for classification purpose. Empirical results ob- tained for all the subtasks by our model are shown in this paper. 1 Introduction Most basic form of communication between hu- mans is through language. Thus it can act as a medium of how we are feeling at any particular instance. For example, if we are angry at someone rather than just hitting him first we would express our feeling through our words. Thus from a con- version we can make out the different emotions a person is going through at that time. Apart from this social media texts can be used for determining the class of a person as described in (Ganesh H. B. et al., 2016b). In this work we are doing 2 ordi- nal classification, 1 classification and 2 regression of different emotions that people exhibits through tweets obtained from twitter (Mohammad and Kir- itchenko, 2018; Bravo-Marquez et al., 2014; Mo- hammad et al., 2013) for three different languages namely Arabic, English and Spanish. The data set given has tweets from all the three languages for each subtask (Mohammad et al., 2018; Moham- mad and Kiritchenko, 2018). There is a total of five subtask an emotion intensity regression task, an emotion intensity ordinal classification task, a sentiment intensity regression task, a sentiment analysis ordinal classification task and an emotion classification task. We used distributed representation (Le and Mikolov, 2014; Ganesh H. B. et al., 2016a) to create feature vector which can be feed as in- put to machine learning algorithms for classifica- tion and regression. Bag-of-words is one of the most common method used to create fixed length feature vectors but the ordering and semantics of the words are ignored in this method. By us- ing Doc2Vec, an unsupervised learning algorithm, we can create fixed length features from variable length data. Thus by using Doc2Vec we can pre- serve the ordering as well …

Content truncated to 500 words