A Hybrid Dependency-Based Approach for Urdu Sentiment Analysis

Abstract

In this era of internet, it is evident that social media is one of the biggest platforms acting as a source of producing a huge amount of raw data on a daily basis that contains the opinion of people from different races, cultures, and age groups on a wide range of topics. Data that is produced on the social media platforms could be utilised by businesses to extract information for improving their services and to reach a wider set of audiences based on users' opinions being shared on these social media sites. In order to utilise this huge amount of unstructured data generated continuously on social media, a deep understanding of Natural Language Processing (NLP) is vital. Most of the approaches proposed so far for sentiment analysis (SA) are based on word co-occurrence frequencies that are inefficient in practical cases. Considering this research gap, this paper proposes a framework for concept-level SA for better SA. We developed our Urdu language dataset by collecting from YouTube, which comprises talks and reviews on different topics like movies, political, and commercial products. Further dataset was integrated with language rules and Deep Neural Networks (DNN) to optimise the polarity detection. For SA, when a predefined rule is triggered, the framework enables the sentiments to flow from words to concepts based on dependency relations among different words of a sentence on the basis of Urdu language grammatical rules. When any of the predefined patterns is not triggered, the framework switches to its sub-symbolic counterpart and passes data to DNN for performing the classification of input sentence. The proposed framework outperforms state-of-the-art approaches, including Long short-term memory (LSTM), Convolutional Neural Network (CNN), State Vector Machine (SVM), Linear Regression (LR), and Multilayer Perceptrons (MLP), with a margin of 6–7%, respectively, when employed on our Urdu dataset.

BERT-Based Sentiment Analysis for Low-Resourced Languages: A Case Study of Urdu Language

“…Lal Khan et al [15] fine-tuned Multilingual BERT and used pre-trained FastText, character N-grams, word embeddings from BERT, and word N-grams from words for SA. Sehar et al [45] proposed a hybrid system that combines deep neural network techniques with dependency-based Urdu grammatical rules to analyze sentiment in Urdu. Habiba et al [46] proposed a rule-based classifier for SA in complicated R-Urdu feelings.…”

Innovations in Urdu Sentiment Analysis Using Machine and Deep Learning Techniques for Two-Class Classification of Symmetric Datasets

“…Research was performed to apply sentiment analysis on an Urdu tweets dataset, and five algorithms were applied and the results were compared [3]. Another study was conducted to apply machine learning and deep learning approaches such as SVM, LR, LSTM, and CNN on Urdu sentences for the analysis of polarity, with the results outperforming previous ones [4]. Research was performed by extracting a small number of tweets, both in Urdu and Roman Urdu, and the WEKA tool was applied but was not able to give results on all applied algorithms in Urdu [5].…”