Cross-Lingual Sentiment Prediction with Various Classifiers for Structural Correspondence Learning and Synonym Augmentation

The Cross-Lingual Sentiment Analysis(CLSA) is a process of utilizing information from resource rich language to predict sentiment of resource poor languages. The monolingual approaches based on machine learning cannot be viably applied for resource-scarce domain as sufficient information is not available. This has picked up specific interest in the field on Natural Language Processing. Subsequently, this article proposes Structural Correspondence Learning (SCL) that utilizes basic correspondence for cross-lingual opinion investigation with synonyms. The key thought behind SCL, is to exploit the knowledge from the source language space to target language space by minimizing loss of data[1]. In contrast to previous SCL methods where it uses logistic classifier to predict, here different classifiers like Random Forest, Gaussian Naïve Bayes, Bernoulli NB, Multinomial NB, K- nearest Neighbors and Decision Tree are trained. This work comprises of two categories: one, training the model with synonyms of pivot for different classifiers and another one, training the model without synonyms on target language using different classifiers. The experimental results show that target language appended with synonyms performs better compared to the one without synonyms for Random Forest.