INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based on Genetic Programming for Twitter Sentiment Analysis

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 771–776, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics INGEOTEC at SemEval 2017 Task 4: A B4MSA Ensemble based on Genetic Programming for Twitter Sentiment Analysis Sabino Miranda-Jim´enez and Mario Graff∗and Eric S. Tellez CONACyT - INFOTEC, Aguascalientes, M´exico {sabino.miranda,mario.graff,eric.tellez}@infotec.mx Daniela Moctezuma CONACyT - CentroGEO, Aguascalientes, M´exico dmoctezuma@centrogeo.edu.mx Abstract This paper describes the system used in SemEval-2017 Task 4 (Subtask A): Mes- sage Polarity Classification for both En- glish and Arabic languages. Our pro- posed system is an ensemble of two lay- ers, the first one uses our generic frame- work for multilingual polarity classifica- tion (B4MSA) and the second layer com- bines all the decision function values pre- dicted by B4MSA systems using a non- linear function evolved using a Genetic Programming system, EvoDAG. With this approach, the best performances reached by our system were macro-recall 0.68 (En- glish) and 0.477 (Arabic) which set us in sixth and fourth positions in the results ta- ble, respectively. 1 Introduction Sentiment Analysis is the computational analysis of people’s feelings or beliefs expressed in texts such as emotions, opinions, attitudes, appraisals, etc. (Liu and Zhang, 2012). At the same time, with the growth of social media (review websites, microblogging sites, etc.) on the Web, Twitter has received particular attention because it is a huge source of opinionated information (6, 000 tweets each second) 1, and has potential uses for decision- making tasks from business applications to politi- cal campaigns. In this context, SemEval is one of the fo- rums that conducts evaluations of Sentiment Analysis Systems on Twitter at different levels such as polarity classification at global or topic- based message, tweet quantifications, among other tasks (Nakov et al., 2016; SemEval, 2017). ∗corresponding author: mario.graff@infotec.mx 1https://www.brandwatch.com/blog/44-twitter-stats- 2016/ In this research, the sentiment analysis task is faced as a classification problem, thus supervised learning techniques are used to tackle this prob- lem. Particularly, we used Support Vector Ma- chines (SVM) and a Genetic Programming system called EvoDAG (Graff et al., 2016, 2017). In this context, one crucial step is the procedure used to transform the data (i.e., tweets) into the inputs (vectors) of the supervised learning tech- niques used. Typically, Natural Language Pro- cessing (NLP) approaches for data representation use n-grams of words, linguistic information such as dependency relations, syntactic information, lexical units (e.g. lemmas, stems), affective lex- icons, etc.; however, selecting the best configura- tion of those characteristics could be a huge prob- lem. In fact, this selection can be seen as a combi- natorial optimization problem where the objective is to improve the accuracy (or any performance measure) of the classifier being used. The pro- posed system uses our generic framework for mul- tilingual polarity classification (B4MSA) (Tellez et al., 2016) to transform the data into the inputs of an SVM. Furthermore, B4MSA uses random search and hill climbing to find a suitable text transformation pipeline among the possible ones. Looking at systems that obtained the …

Content truncated to 500 words