Browsing by Author "Dr. Muhammad Waqas Anwar"

Now showing 1 - 12 of 12

A Comparative Analysis of Offensive Language Detection in Roman Urdu
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2022) Muneeb Rehman Khan; FA19-RCS-019; LHR TP 8305; Dr. Muhammad Waqas Anwar
The rise of social media has encouraged experts to conduct several studies based on user generated data. The widespread use of online platforms like Facebook, Twitter and other web forums has created challenges to keep check for inappropriate language. Pakistan has about 44 million online social network users that speak Roman Urdu to interact with each other. The prevalence of user-generated information on social media caused a massive rise in the content of offensive texts. However, this growth also offers avenues for trolls who contaminate these social platforms with their nasty remarks. In South Asia, Roman Urdu is the most widely used communication language on social media. Compared to other resource-rich languages such as English, less work has been done on offensive language classification in Roman Urdu. In this research (1) a lexicon of offensive terms in Roman Urdu is presented, (2) developed a dataset called RU-OLD “Roman Urdu Offensive Language Detection” with 15000 tweets, and (3) conducted a comparative analysis of machine learning and several deep learning architectures. Comparative analysis was performed by well-known machine learning classifiers such as Naive Bayes, RF, SVM, and Logistic Regression as well as cutting-edge deep neural networks such as Long Short-Term Memory, Convolutional Neural Networks, XLNET and BERT. XLNET trained with developed corpus outperformed the rest of the models and resulted in the highest accuracy of 93%. The results are suggestive of the scalability of the proposed approach on large scale to counter offensive language in Roman Urdu
An Empirical Study of Urdu Noun and Verb Phrase Chunking
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2020) Maryam Khurshid; FA17-RCS-025; LHR TP 5979; Dr. Muhammad Waqas Anwar
Urdu is a language which is a morphologically rich and weak resourced language. The distinguishing features such as free word order, context-sensitive orthography, flexible grammar rules and complex morphology makes the representation of the Urdu language a difficult problem area. In Urdu's hand-written text the words are written without any space among them. A computer needs a text file that needs a separator when a word ends with a non-joiner character. Without these separators, the words will join with one another that will not be understandable for language native speakers. Chunking is a basic technique used for entity detection that labels and segments the sequence of Multi tokens. Chunking technique helps in the progress of many Natural Processing Applications. Chunking is a mature field while dealing with other languages like Hindi, English, Chinese and Turkish but it still requires the attention of researchers in the Urdu language. The Native speakers of Urdu language are more than 70 Million. The study is about the noun and verb phrase chunking in the Urdu language. The intention of this work is to explore the corpus accuracy based on the Noun and verb phrase chunking of the Urdu language. Chunking is an NLP (natural language processing) function that focuses on splitting a text into syntactically linked non-overlapping and non-exhaustive word-groups i.e. a word could only be a part of one chunk but not all words are in chunks. Different experiments are conducted on this work by using a tag set of different input and output schemes with the same Methodology. Firstly, the corpus is selected then preprocessing is performed on that corpus. After that part of speech tagging and IOB tags are assigned to that corpus then Noun and verb phrases are detected by neural networks and machine learning techniques. After this Noun and Verb phrases are detected from a corpus. At last, evaluation will be done by using different Parameters like F-call, recall and Precision.
Aspect Based Sentiment Analysis for Urdu Language Using Rule-Based Approac
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2022) EHTESHAM HASHMI; FA20-RCS-009; Dr. Muhammad Waqas Anwar; LHR TP 8241
Aspect Based Sentiment Analysis (ABSA) also known as entity level analysis has gained the vital importance to extract the people’s sentiment, emotions or opinions towards some entity. This entity can be any organization, social platforms etc. In spite of the fact that Sentiment Analysis (SA) has been examined broadly within the English language domain with some eminent work in other dialects including Chinese, Arabic etc., many other resource poor languages including Urdu did not get much attention of researchers due to the lack of resources. Urdu is widely spoken language in all over the world and various social platforms are full of Urdu reviews containing people’s sentiments. In this study we have established an aspect level rule-based approach for the Urdu language. To perform our work, we have used the Urdu dialect dataset containing the “COVID-19” tweets. These tweets contain all the information related to Coronavirus and people’s views towards this disease. Sentiment lexicon has been used to extract for the opinion term present in a tweet and after getting the opinion term various aspect have been extracted associated with the opinion. Aspect extraction has been performed using various rule, these rules have been created using linguistic and syntactic context of the phrases present in the tweet and polarities are assigned accordingly. The proposed study focused to achieve four crucial modules: aspect term, aspect term polarity, aspect term category and aspect term category polarity. With the help of evaluation measures including F1-Score, Accuracy, Precision and Recall our work has achieved the promising results.
Cross Domain Sentiment Classification for Urdu Language
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2020) Sana Hassan; FA18-RCS-010; LHR TP 6410; Dr. Muhammad Waqas Anwar
An enormous amount of information is produced daily on the internet about different prod ucts and objects. People like to express their feelings, thoughts in their native language on different social sites. This bulk data needs to be interpreted. So, Sentiment Analysis (SA) is required which extracts people’s opinions, feelings, and thoughts. However, it is a highly domain-dependent task. Due to this, the issue of domain-transfer arises. If a classifier is tested with any different domain, other than on which it is trained, its performance is affected. Many tasks and frameworks are created in mostly English and western languages. Tasks intended for the English language cannot be applied for other languages, hence there is a need to work on different dialects. In this research, we performed a cross-domain sentiment Analysis on data set of Urdu language comprising of 9000 sentences from sports tweets in which there are two domains (Hockey and Cricket). Furthermore, preparing corpus for specific domains in the Urdu language we applied ma chine learning and deep learning approach. After this, we evaluated results using standard evaluation measures and a confusion matrix, Gated Recurrent Units (GRU) gives the highest accuracy of (77%)
Fake News Detection and Corpus Development for Urdu News
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2021) Muhammad Ismail Suri; SP17-RCS-009; LHR TP 7287; Dr. Muhammad Waqas Anwar
With the growing technological advancements in the present era, the curiosity to remain updated about the daily happenings in the world has opened up various opportunities for news dissemination. Access to information with ease has made the propagation of news much efficient and cost effective. However, online news is a double-edged sword that creates a room for spreading fake news which intends to mislead and exaggerate the reality. This issue is a rising problem leading towards several real world sociopolitical and cultural impacts. To curb the menace of fake news different organizations are working to handle it, but majority of the work is being done in English, unfortunately the language of Urdu in this dimension remains untouched. This research is focused on detecting fake news in Urdu language. The demand is to have a detecting mechanism that can predict between fake and real news. In South Asian languages, unavailability of datasets is the major hindrance of evaluating fake news. Therefore, my contribution is the induction of a dataset influenced by English fake news dataset for Urdu news; Corpus of Urdu News for Fake News Detection (COUNFND) which has been derived from digital journalism. It contains 480 documents with 120 claims processed by means of conducting traditional machine learning algorithms and text preprocessing techniques. The corpus holds great importance for the evaluation of fake news detection system specifically for Urdu Language. This thesis discusses different techniques that have been applied on the Urdu language datasets and furthermore achieving classification accuracy of about 73 percent effectively. Later on, a new dataset of fake news in Urdu Language has been introduced by Maaz Amjad (Bend the Truth), which has articles divided into fake and real classification, I have applied different supervised learning models and have achieved an accuracy of 0.89 on random forest using four grams techniques which is greater than the baseline accuracy provided with the dataset.
Neural Abstractive Text Summarization for Urdu Language
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2022) Adnan Sanaullah; FA18-RCS-012; LHR TP 8343; Dr. Muhammad Waqas Anwar
Text summarization achieved a lot of popularity in natural language processing because of the large amount of literature available on internet, especially for English language. Nowadays, most used technique is abstractive text summarization in which generated summaries are quite related to the human-written summaries. In this research, to create the summaries for the Urdu language the abstractive text summarization technique is used. In this technique, the Attention based sequence to sequence encoder decoder model are used to create the summaries. For the training of model for Urdu Language, two dataset which are BBC Urdu Dataset and Urdu News 1M are used. In order to evaluate the model, ROUGE metrics are used in which the model generated summary and human-written summary are compared and then performance of the model is measured. After training the model on both datasets, there is the quit the difference between the results of both the datasets which is due to the size of dataset. Although the model got 42.85 rouge-1 score on BBC Urdu Dataset and 66.67 on Urdu News 1M Dataset. Our model shows promising results on both the datasets but if the size of dataset increases the model performs better. We also discussed the problem faced during the completion of research and results of the model in this research work
Neural Architecture Based Urdu Named Entity Recognition
(Library Information Services COMSATS University Islamaabad Lahore Campus, 2020) Muhammad Ehsan; FA18-RCS-018; LHR TP 6413; Dr. Muhammad Waqas Anwar
In natural language processing, named entity recognition (NER) is the issue of figuring out and arrange named elements referenced in unstructured content into pre-characterized classifications. To compare, evaluate and develop a NER system, we need standard and high-quality benchmark resources. But unfortunately, corpora in a huge amount are available for many other languages but not for the Urdu language. Even though Urdu is a rich language in all aspects and has more than 300 million speakers all over the world. As state-of-the-art sequence labeling systems are heavily based on large amounts of domain specific knowledge and hand-crafted feature for efficient learning from small data. When the dataset is in a large amount and high quality then the Named Entity Recognition (NER) task can be solved in a better way and provides better results on a large amount of labeled data, but the availability of labeled data in most of the languages is limited for training purpose. One way is there are scarcely any limitations on identifiable words. Hence, concluded the little amount of exampled data is troublesome. Subsequently, deliberately built orthographic features and language-explicit information assets (e.g., gazetteers) are broadly utilized for understanding this undertaking. Tragically, for new languages and new domains, the task of identifying explicit language sources and feature extraction is mainly an expensive process. As a result, it makes efficient NER system development adaptation a challenge. Another way to makes an efficient NER system is the use of deep neural network models that have achieved highest performance in the domain of natural language processing (NLP). However, this performance highly depends on the availability of huge amount of annotated data. The other way to get a better generalization from a small amount of data is to deploy semi-supervised models that performs on small set of annotated examples and a large set of unannotated data. Due to limited annotated data for many Natural Language Processing (NLP) problems like NER for the Urdu language, we will solve the NER problem using deep neural algorithms. In deep learning methods, we perform 44 experiments by using recurrent neural network (RNN), long short-term memory (LSTM), Bidirectional LSTM, gated recurrent unit (GRU), Bidirectional GRU and GRUCell in a supervised learning setting to compare with ix the previous models. Our model outperformed the previous results with F1-score of 91.43%. We also use multi-layer perceptron in a semi-supervised setting. This model gives the results with F1-score of 51.00%. We use standard evaluation measures to check the performance of the NER system. However, deep learning results with semi-supervised settings unable to beat the results of deep learning results with supervised settings.
Resource Creation and Evaluation of Aspect Based Sentiment Analysis in Urdu
(Library Information Services COMSATS University Islamaabad Lahore Campus, 2020) LHR TP 6412; Dr. Muhammad Waqas Anwar; FA18-RCS-017; Sadaf Rani
Sentiment analysis is an emerging research area of Natural Language Processing (NLP). Along with the rise of people generated content on social sites, sentiment analysis has gained more importance. One of the challenging tasks of sentiment analysis is Aspect Based Sentiment Analysis (ABSA). ABSA is a task of identifying the sentiment at aspect level. Right now, mostly researchers have focused on the English language but very less attention on resource poor languages like Urdu, Arabic etc. To the best of our knowledge, there is no publicly available dataset on ABSA in Urdu language. In this study, we focused on the resource creation for the task of ABSA in Urdu language. This task is further divided into four subtasks i.e., Aspect Term, Aspect Term Polarity, Aspect Category and Aspect Category Polarity. Our dataset is comprised of 6672 sentences which are collected form twitter using Twitter API’s. For annotation, we prepared standard guidelines which are according to the SemEval. Evaluation is also performed by extracting n gram features at word level and char level. TF-IDF vectorizer is used to vectorize the data into machine readable form. Then we applied different machine learning algorithms. Experiments are performed by Naïve Bayes (NB), support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF) and performance is measured by four parameters i.e., accuracy, precision, recall, f1 measure. We achieved 71% accuracy in Aspect Term task, 61% in Aspect Category, 74% in Aspect Term Polarity and 80% in Aspect Category Polarity The evaluation results depict that there is need to do more work in this task. We provided baseline evaluation for researcher community to further compare their results of ABSA systems.
Structural based Sentiment Mining for Roman Urdu
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2020) Mubashir Ubaid Ullah; SP18-RCS-013; LHR TP 5986; Dr. Muhammad Waqas Anwar
Web-based data is increasing day by day and plays a vital role in developing people’s opinions. Sentiment mining/analysis is the natural language processing task that helps to identify, classify these opinions. Usually research focus is on resource-rich language for sentiment mining. In this thesis, we performed classification of various sentiments using feature selection techniques for a resource-poor language i.e. Roman Urdu. These classification techniques include chi-square, mutual information and select from model which are implemented on the Roman Urdu Dataset of 11k reviews. Well-known machine learning algorithms are applied for experimental analysis that includes Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Multinomial Naïve Bayes (MNB) and Multi-Layer Perceptron (MLP). These are applied for character-level & word-level features for n-gram variations that are bigram to 7-gram for character-level classification and Uni, Bi, Uni Bi gram, Uni-Bi-Tri gram & Uni-Bi-Tri-Four gram in terms of word-level classification. Results are being evaluated using accuracy, precision, recall & f1-score. The Highest accuracies for both word-level and character-level achieved are 83.93% and 83.72% which improves the baseline score that was 82.46% on feature union whereas F1-score is 90.51% & 90.42% respectively. Some renowned Neural Network techniques are also applied in this thesis which include CNN, LSTM, & Bi-LSTTM. We achieved maximum results by Bi-LSTM which gives 91.8% accuracy and 91.7% F1-score
Textual Cyberbullying Detection on Social Networks using Machine Learning and Deep Learning Models
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2022) Gull Bano Anwar; FA20-RCS-011; LHR TP 8359; Dr. Muhammad Waqas Anwar
Nowadays Cyberbullying on social media has become a major problem. Cyberbullying may cause many serious and negative mental, emotional and physical impacts on a person's life. However, Cyberbullying leaves a record that can demonstrate value and give proof to help stop digital abuse. The early detection of Cyberbullying on social media becomes crucial to moving the effect on the social media user. In this direction, many studies are conducted to detect Cyberbullying content automatically. The major concern and gap in Cyberbullying detection strategies is the lack of linguistic resources, especially for newly evolved languages. Roman Urdu is a newly emerged and widely used language on social network sites in Asian countries. The greatest strategy to prevent Cyberbullying is to use Machine Learning or Deep Learning with Natural Language Processing (NLP) tools to detect it automatically. The current research proposed an efficient framework to detect Cyberbullying, using NLP tools with Machine Learning and Deep Learning models. Using different preprocessing techniques, the proposed study is validated on a roman-Urdu-abusive-comment detector (RUACD) dataset. Data Preprocessing steps were followed that included text cleaning, tokenization, lemmatization, and removal of stop words. For experimental purposes, five machine learning models Support Vector Machine (SVM), Naïve Bayes (NB), Logistic Regression (LR), Random Forest (RF), and Decision Tree (DT) and 4 deep learning models are evaluated on the RUACD dataset. From experiments of machine learning models, current study finds that the SVM, LR, and DT outperformed and achieved promising results as SVM, LR and DT achieve 96.2%, 94.91, and 94.01 of test accuracy and from experiments of deep learning models, current study find that the DNN, LSTM, and RNN outperformed and achieved promising results as DNN, LSTM, and RNN achieves 90.4%, 86.5, and 85.4 of test accuracy. Ensemble of these outperformed models is formed separately and achieved 95.92% of test accuracy with machine learning ensemble name EN-SLD, and achieved 90.92% of test accuracy with deep learning ensemble name EN-DLR. At last the ensemble of both EN-SLD and EN DLR is formed and achieved 93.92% of test accuracy
Urdu to English Based Unsupervised Machine Translation
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2020) Ahmed Raza; SP18-RCS-005; LHR TP 5983; Dr. Muhammad Waqas Anwar
The process of automatically converting the text from one language to another natural language is Machine Translation. Machine Translation is a subfield of computational linguistic. There are two state-of-the-art machine translation techniques i,e Neural Machine Translation (NMT), and Statistical Machine Translation (SMT). In both techniques, a large corpus is required for the training of the translation model. Urdu counts in low resource languages due to the fewer resources available for computational work. To build a good translation system available resources are not enough. Many languages present in the world have a different structure. Like in Urdu and English, Urdu structure is based on Subject Object-Verb (SOV) and the English structure is based Subject-Verb-Object (SVO). In this study, we presented Urdu to English unsupervised translation model and the practical challenges faced during the work. We try to partially remove the need for parallel corpora and proposed a method to train a Machine Translation System in an unsupervised manner. The proposed system is aimed to provide Urdu to English translation through an unsupervised manner. For this propose, we use Artetxe Author developed a toolkit that is based on Unsupervised Neural Machine Translation (UNMT). This approach tested the models of UNMT which include denoising and on-the-fly back-translation. From denoising model obtain the BLEU score 4.14 and 5.11 for two language pairs UR-EN and EN-UR. From back translation obtain the BLEU score of 5.21 and 6.28 which are better than from the previous score. Back-translation results difference from denoising technique gains +1.07 and +1.17 for two language pairs Urdu to English and English to Urdu. We also faced many challenges during work and effects on pre-processing techniques. Our approach shows promising results in translation of Urdu text into English which is mostly neglected due to its complexities
Urdu-English Machine Transliteration using Neural Networks
(Library Information Services, COMSATS University Islamabad, Lahore Campus, 2019) Usman Mohy ud Din; FA16-RCS-007; LHR TP 5776; Dr. Muhammad Waqas Anwar
Machine translation has gained much attention in recent years. It is a sub-field of computational linguistic which focus on translating text from one language to other language. Among different translation techniques, neural network currently leading the domain with its capabilities of providing a single large neural network with attention mechanism, sequence-to-sequence and long-short term modelling. Despite significant progress in domain of machine translation, translation of out-of-vocabulary words(OOV) which include technical terms, named-entities, foreign words are still a challenge for current state-of-art translation systems, and this situation becomes even worse while translating between low resource languages or languages having different structures. Due to morphological richness of a language, a word may have different meninges in different context. In such scenarios, translation of word is not only enough in order provide the correct/quality translation. Transliteration is a way to consider the context of word/sentence during translation. For low resource language like Urdu, it is very difficult to have/find parallel corpus for transliteration which is large enough to train the system. In this work, we presented transliteration technique based on Expectation Maximization (EM) which is un-supervised and language independent. Systems learns the pattern and out-of-vocabulary (OOV) words from parallel corpus and there is no need to train it on transliteration corpus explicitly. This approach is tested on three models of statistical machine translation (SMT) which include phrase based, hierarchical phrase-based and factor based models and two models of neural machine translation which include LSTM and transformer model. On SMT models, there is gain of 0.63 to 0.91 in BLEU score while on NMT models, there is gain of 1.28 to 2.05 in BLEU which are better than previous baseline scores. Our approach shows promising results in translation of Urdu text into English which is mostly neglected due to its complexities. We also discussed the results, different challenges faced during this work and effect of right pre-processing techniques