Automatic summarization condenses content while retaining key ideas and details. Urdu, with over 230 million speakers globally, is one of the most widely spoken languages. The rise of Urdu content on social media platforms has driven the need for tools that enhance accessibility and engagement. The growing popularity of social media has increased the number of Urdu instructional videos. Well-written video descriptions can boost viewer engagement and improve search engine optimization; however, many lack these. Therefore, an automatic description generation system for Urdu videos is needed, which can be achieved by abstractive summarization of video transcripts. However, such public datasets are not available in Urdu. To address this problem, we investigate the usability of high-resource language datasets for Urdu abstractive text summarization. We created the first Urdu video transcription dataset Urdu How2 and evaluated its quality using intrinsic evaluation. We leverage transfer learning, a technique where knowledge from pretrained models (like mT5) is adapted to new tasks, to develop the uT5 model for generating Urdu text summaries. We further trained the model to improve its Urdu text generation capability. The machine-generated summaries are evaluated using ROUGE scores, human evaluation scores, and adversarial evaluation, providing a reliable assessment of the quality of generated descriptions and the robustness of the model against noisy text data. The human evaluation shows the proposed method generates accurate and coherent summaries compared to the translated ground truth. To the best of our knowledge, this is the first attempt to utilize a cross-lingual dataset for Urdu abstractive text summarization and video description generation. This research enhances Urdu content accessibility and lays the groundwork for advancing multilingual content generation and multimodal analysis in other low-resource languages.

In this era of multimedia dominance, the surge of multimodal content on social media has transformed our methods of communication and information exchange. With the widespread use of multimedia content, the ability to effectively summarize this multimodal content is crucial for enhancing consumption, searchability, and retrieval. The scarcity of such training datasets has been a barrier to research in this area, especially for low-resource languages like Urdu. To address this gap, this paper introduces “UrduMASD”, a video-based Urdu multimodal abstractive text summarization dataset. The dataset contains 15,374 collections of videos, audio, titles, transcripts, and corresponding text summaries. To ensure the quality of the dataset, intrinsic evaluation metrics such as Abstractivity, Compression, Redundancy, and Semantic coherence have been employed. It was observed that our dataset surpasses existing datasets on numerous key quality metrics. Additionally, we present baseline results achieved using both text-based and state-of-the-art multimodal summarization models. On adding visual information, an improvement of 2.6% was observed in the ROUGE scores, highlighting the efficacy of utilizing multimodal inputs for summarization. To the best of our knowledge, this is the first dataset in Urdu that provides video-based multimodal data for abstractive text summarization, making it a valuable resource for advancing research in this field.

Cybercrime is a serious and growing threat affecting millions of people worldwide. Detecting cybercrimes from text messages is challenging, as it requires understanding the linguistic and cultural nuances of different languages and regions. Roman Urdu is a widely used language in Pakistan and other South Asian countries, however, it lacks sufficient resources and tools for natural language processing and cybercrime detection. To address this problem, we make three main contributions in this paper. (1) We create and release CRU, a benchmark dataset for text-based cybercrime detection in Roman Urdu, which covers a number of cybercrimes as defined by the Prevention of Electronic Crimes Act (PECA) of Pakistan. This dataset is annotated by experts following a standardized procedure based on Pakistan’s legal framework. (2) We perform experiments on four pre-trained language models (PLMs) for cybercrime text classification in Roman Urdu. Our results show that xlm-roberta-base is the best model for this task, achieving the highest performance on all metrics. (3) We explore the utility of prompt engineering techniques, namely prefix and cloze prompts, for enhancing the performance of PLMs for low-resource languages such as Roman Urdu. We analyze the impact of different prompt shapes and k-shot settings on the performance of xlm-roberta-base and bert-base-multilingual-cased. We find that prefix prompts are more effective than cloze prompts for Roman Urdu classification tasks, as they provide more contextually relevant completions for the models. Our work provides useful insights and resources for future research on cybercrime detection and text classification in low-resource languages.

In an era where social media platform users are growing rapidly, there has been a marked increase in hateful content being generated; to combat this, automatic hate speech detection systems are a necessity. For this purpose, researchers have recently focused their efforts on developing datasets, however, the vast majority of them have been generated for the English language, with only a few available for low-resource languages such as Roman Urdu. Furthermore, what few are available have small number of samples that pertain to hateful classes and these lack variations in topics and content. Thus, deep learning models trained on such datasets perform poorly when deployed in the real world. To improve performance the option of collecting and annotating more data can be very costly and time consuming. Thus, data augmentation techniques need to be explored to exploit already available datasets to improve model generalizability. In this paper, we explore different data augmentation techniques for the improvement of hate speech detection in Roman Urdu. We evaluate these augmentation techniques on two datasets. We are able to improve performance in the primary metric of comparison (F1 and Macro F1) as well as in recall, which is impertinent for human-in-the-loop AI systems.

Abstract Roman Urdu is an informal form of the Urdu language written in Roman script, which is widely used in South Asia for online textual content. It lacks standard spelling and hence poses several normalization challenges during automatic language processing. In this article, we present a feature-based clustering framework for the lexical normalization of Roman Urdu corpora, which includes a phonetic algorithm UrduPhone, a string matching component, a feature-based similarity function, and a clustering algorithm Lex-Var. UrduPhone encodes Roman Urdu strings to their pronunciation-based representations. The string matching component handles character-level variations that occur when writing Urdu using Roman script. The similarity function incorporates various phonetic-based, string-based, and contextual features of words. The Lex-Var algorithm is a variant of the k-medoids clustering algorithm that groups lexical variations of words. It contains a similarity threshold to balance the number of clusters and their maximum similarity. The framework allows feature learning and optimization in addition to the use of predefined features and weights. We evaluate our framework extensively on four real-world datasets and show an F-measure gain of up to 15% from baseline methods. We also demonstrate the superiority of UrduPhone and Lex-Var in comparison to respective alternate algorithms in our clustering framework for the lexical normalization of Roman Urdu.

In recent years, web-based platforms have become popular for LaTeX document authoring and presentation due to their hassle-free environments and collaboration features. Unfortunately, currently available platforms are not accessible to persons with vision impairments (PVIs) due to poor screen reader and keyboard support. In this paper, we present Web-Accessible LaTeX-based Authoring and Presentation (Web-ALAP), a web version of ALAP that makes creating mathematical documents accessible to PVIs. Web-ALAP provides assistive features like speech-based prompts, automatic narration of the error message along with the line number for debugging, and essential keyboard shortcuts. More importantly, the integrated “Math Mode” library offers a natural language description of the mathematical content within a document, which is not done by other screen readers. This feature enables PVIs to find and fix semantic errors in equations as well. We discuss the design choices in Web-ALAP and evaluate its features with 10 visually impaired participants. The results show that users perform better while working with Web-ALAP as opposed to another web-based LaTeX editor (Overleaf).

LaTeX is widely used in STEM fields for creating high-quality documents that are converted to the Portable Document Format (PDF) for dissemination. Currently, available LaTeX systems do not guarantee that the generated PDFs are compliant with international accessibility standards. In this work, we present AGAP (Automated Generation of Accessible PDF) that automates and makes accessible the process of generating accessible PDFs from LaTeX. AGAP flags accessibility violations and provides guidance on how to fix them at compile time. AGAP allows interaction through speech synthesis and keyboard shortcuts thus making it fully accessible to persons with vision impairments (PVIs). Evaluating the accessible PDF generated using AGAP with a standard accessibility checker resulted in a much smaller number of violations as opposed to the PDF generated using another desktop LaTeX editor.

The task of automatic hate-speech and offensive language detection in social media content is of utmost importance due to its implications in unprejudiced society concerning race, gender, or religion.Existing research in this area, however, is mainly focused on the English language, limiting the applicability to particular demographics.Despite its prevalence, Roman Urdu (RU) lacks language resources, annotated datasets, and language models for this task.In this study, we: (1) Present a lexicon of hateful words in RU, (2) Develop an annotated dataset called RUHSOLD consisting of 10, 012 tweets in RU with both coarse-grained and fine-grained labels of hate-speech and offensive language, (3) Explore the feasibility of transfer learning of five existing embedding models to RU, (4) Propose a novel deep learning architecture called CNN-gram for hatespeech and offensive language detection and compare its performance with seven current baseline approaches on RUHSOLD dataset, and (5) Train domain-specific embeddings on more than 4.7 million tweets and make them publicly available.We conclude that transfer learning is more beneficial as compared to training embedding from scratch and that the proposed model exhibits greater robustness as compared to the baselines.

Assistive technologies such as screen readers and text editors have been used in past to improve the accessibility and authoring of scientific and mathematical documents. However, most screens readers fail to narrate complex mathematical notations and expressions as they skip symbols and necessary information required for the accurate narration of mathematical content. This study aims at evaluating a new Accessible LaTeX Based Mathematical Document Authoring and Presentation (ALAP) tool, which assist people with visual impairments in reading and writing mathematical documents. ALAP includes features like, assistive debugging, Math Mode for reading and writing mathematical notations, and automatic generation of an accessible PDF document. These features aim to improve the LaTeX debugging experience and make it simple for blind users to author mathematical content by narrating it in natural language through the use of integrated text to speech (TTS) engine. We evaluated ALAP by conducting a study with 18 visually impaired LaTeX users. The results showed that users preferred ALAP over another comparable LaTeX based authoring tool and were relatively more comfortable in completing the tasks while using ALAP.

Sentiment classification is a popular text mining task in which textual content (e.g., a message) is assigned a polarity label (typically positive or negative) reflecting the sentiment expressed in it. Sentiment classification is used widely in applications like customer feedback analysis where robustness and correctness of results are critical. In this article, we highlight that prediction accuracy alone is not sufficient for assessing the performance of a sentiment classifier; it is also important that the classifier is not biased toward positive or negative polarity, thus distorting the distribution of positive and negative messages in the predictions. We propose a measure, called Polarity Bias Rate, for quantifying this bias in a sentiment classifier. Second, we present two methods for removing this bias in the predictions of unsupervised and supervised sentiment classifiers. Our first method, called Bias-Aware Thresholding (BAT), shifts the decision boundary to control the bias in the predictions. Motivated from cost-sensitive learning, BAT is easily applicable to both lexicon-based unsupervised and supervised classifiers. Our second method, called Balanced Logistic Regression (BLR) introduces a bias-remover constraint into the standard logistic regression model. BLR is an automatic bias-free supervised sentiment classifier. We evaluate our methods extensively on seven real-world datasets. The experiments involve two lexicon-based and two supervised sentiment classifiers and include evaluation on multiple train-test data sizes. The results show that bias is controlled effectively in predictions. Furthermore, prediction accuracy is also increased in many cases, thus enhancing the robustness of sentiment classification.

This software usability study aims towards evaluating our LaTeX based extension, created to assist blind researchers and writers, in terms of authoring both continuous and non-continuous text [2]. Our extension includes features like speech based error prompts and navigation to the error location which are expected to improve the LaTeX code debugging experience and increase writing productivity. Upon testing our extension, it is observed that a majority of both LaTeX novice and expert users preferred using MS Word for writing continuous text, while the LaTeX experts preferred our extension for writing mathematical content.

Named entities such as people, locations, and organizations play a vital role in characterizing online content. They often reflect information of interest and are frequently used in search queries. Although named entities can be detected reliably from textual content, extracting relations among them is more challenging, yet useful in various applications (e.g., news recommending systems). In this paper, we present a novel model and system for learning semantic relations among named entities from collections of news articles. We model each named entity occurrence with sparse structured logistic regression, and consider the words (predictors) to be grouped based on background semantics. This sparse group LASSO approach forces the weights of word groups that do not influence the prediction towards zero. The resulting sparse structure is utilized for defining the type and strength of relations. Our unsupervised system yields a named entities' network where each relation is typed, quantified, and characterized in context. These relations are the key to understanding news material over time and customizing newsfeeds for readers. Extensive evaluation of our system on articles from TIME magazine and BBC News shows that the learned relations correlate with static semantic relatedness measures like WLM, and capture the evolving relationships among named entities over time.

Low-literate people are unable to use many mainstream social networks due to their text-based interfaces even though they constitute a major portion of the world population. Specialized speech-based networks (SBNs) are more accessible to low-literate users through their simple speech-based interfaces. While SBNs have the potential for providing value-adding services to a large segment of society they have been hampered by the need to operate in low-income segments on low budgets. The knowledge of influential users and communities in such networks can help in optimizing their operations. In this paper, we present a self-organizing map (SOM) for discovering and visualizing influential communities of users in SBNs. We demonstrate how a friendship graph is formed from call data records and present a method for estimating influences between users. Subsequently, we develop a SOM to cluster users based on their influence, thus identifying community-level influences and their roles in information propagation. We test our approach on Polly, a SBN developed for job ads dissemination among low-literate users. For comparison, we identify influential users with the benchmark greedy algorithm and relate them to the discovered communities. The results show that influential users are concentrated in influential communities and community-level information propagation provides a ready summary of influential users.

We present an unsupervised method to find lexical variations in Roman Urdu informal text.Our method includes a phonetic algorithm UrduPhone, a featurebased similarity function, and a clustering algorithm Lex-C.UrduPhone encodes roman Urdu strings to their phonetic equivalent representations.This produces an initial grouping of different spelling variations of a word.The similarity function incorporates word features and their context.Lex-C is a variant of k-medoids clustering algorithm that group lexical variations.It incorporates a similarity threshold to balance the number of clusters and their maximum similarity.We test our system on two datasets of SMS and blogs and show an f-measure gain of up to 12% from baseline systems.

In data mining we often have to learn from biased data, because, for instance, data comes from different batches or there was a gender or racial bias in the collection of social data. In some applications it may be necessary to explicitly control this bias in the models we learn from the data. This paper is the first to study learning linear regression models under constraints that control the biasing effect of a given attribute such as gender or batch number. We show how propensity modeling can be used for factoring out the part of the bias that can be justified by externally provided explanatory attributes. Then we analytically derive linear models that minimize squared error while controlling the bias by imposing constraints on the mean outcome or residuals of the models. Experiments with discrimination-aware crime prediction and batch effect normalization tasks show that the proposed techniques are successful in controlling attribute effects in linear regression models.

Social discrimination (e.g., against females) arising from data mining techniques is a growing concern worldwide. In recent years, several methods have been proposed for making classifiers learned over discriminatory data discrimination-aware. However, these methods suffer from two major shortcomings: (1) They require either modifying the discriminatory data or tweaking a specific classification algorithm and (2) They are not flexible w.r.t. discrimination control and multiple sensitive attribute handling. In this paper, we present two solutions for discrimination-aware classification that neither require data modification nor classifier tweaking. Our first and second solutions exploit, respectively, the reject option of probabilistic classifier(s) and the disagreement region of general classifier ensembles to reduce discrimination. We relate both solutions with decision theory for better understanding of the process. Our experiments using real-world datasets demonstrate that our solutions outperform existing state-of-the-art methods, especially at low discrimination which is a significant advantage. The superior performance coupled with flexible control over discrimination and easy applicability to multiple sensitive attributes makes our solutions an important step forward in practical discrimination-aware classification.

SMS are short-length text documents written in a colloquial style. SMS text processing is challenging because of low signal-to-noise ratio and multi-varied text composition in terms of language, vocabulary, style and quality. These challenges can be overcome by robust text normalization, which is a necessary step before any technique can be applied and evaluated on such data. In this paper, we present a rule-based model for multi-lingual SMS text normalization focusing on messages written in Romanized Urdu and English. Urdu in contrast to English is a morphologically rich language (MRL), i.e. it produces a very large number of word forms for a given root form, while Romanized Urdu is a way of writing Urdu in Latin script which does not follow standard rules for systematic communication. Hence, normalization or standardization of multi-lingual SMS text offers challenges associated with SMS text, multi-lingualism, MRLs and Latin script. Our SMS standardizer is based upon a tuned set of rules that range over various domains of natural language processing, and which tackle the challenges mentioned above effectively. We then implement the standardizer in the application of Keyword Extraction from SMS messages, where it produces significant improvement in performance by upto 23% in F-measure.

Dimensionality reduction (DR) through feature extraction (FE) is desirable for efficient and effective processing of text documents. Many of the techniques for text FE produce features that are not readily interpretable and require super-linear computation time. In this paper, we present a fast supervised DR/FE technique, named FEDIP, that is motivated by the notion of relatedness of terms to topics or contexts. This relatedness is quantified by using the discrimination information provided by a term for a topic in a labeled document collection. Features are constructed by pooling the discrimination information of highly related terms for each topic. FEDIP's time complexity is linear in the size of the vocabulary and document collection. FEDIP is evaluated for document classification with SVM and naive Bayes classifiers on six text data sets. The results show that FEDIP produces low-dimension feature spaces that yield higher classification accuracy when compared with LDA and LSI. FEDIP is also found to be significantly faster than the other techniques on our evaluation data sets.

In this paper, we propose and evaluate a self-optimization strategy for a clustering-based tag recommendation system. For tag recommendation, we use an efficient discriminative clustering approach. To develop our self-optimization strategy for this tag recommendation approach, we empirically investigate when and how to update the tag recommender with minimum human intervention. We present a nonlinear optimization model whose solution yields the clustering parameters that maximize the recommendation accuracy within an administrator specified time window. Evaluation on "BibSonomy'' data produces promising results. For example, by using our self-optimization strategy a 6% increase in average F1 score is achieved when the administrator allows up to 2% drop in average F1 score in the last one thousand recommendations.

Text classification is widely used in applications ranging from e-mail filtering to review classification. Many of these applications demand that the classification method be efficient and robust, yet produce accurate categorizations by using the terms in the documents only. We present a supervised text classification method based on discriminative term weighting, discrimination information pooling, and linear discrimination. Terms in the documents are assigned weights according to the discrimination information they provide for one category over the others. These weights also serve to partition the terms into two sets. A linear opinion pool is adopted for combining the discrimination information provided by each set of terms yielding a two-dimensional feature space. Subsequently, a linear discriminant function is learned to categorize the documents in the feature space. We provide intuitive and empirical evidence of the robustness of our method with three term weighting strategies. Experimental results are presented for data sets from three different application areas. The results show that our method's accuracy is higher than other popular methods, especially when there is a distribution shift from training to testing sets. Moreover, our method is simple yet robust to different application domains and small training set sizes.

Typically, spam filters are built on the assumption that the characteristics of e-mails in the training set is identical to those in individual users' inboxes on which it will be applied. This assumption is oftentimes incorrect leading to poor performance of the filter. A personalized spam filter is built by taking into account the characteristics of e-mails in individual users' inboxes. We present an automatic approach for personalized spam filtering that does not require users' feedback. The proposed algorithm builds a statistical model of significant spam and non-spam words from the labeled training set and then updates it in multiple passes over the unlabeled individual user's inbox. The personalization of the model leads to improved filtering performance. We evaluate our algorithm on two publicly available datasets. The results show that our algorithm is robust and scalable, and a viable solution to the server-side personalized spam filtering problem. Moreover, it outperforms published results on one dataset and its performance is equivalent to the others on the second dataset.

The volume of spam e-mails has grown rapidly in the last two years resulting in increasing costs to users, network operators, and e-mail service providers (ESPs). E-mail users demand accurate spam filtering with minimum effort from their side. Since the distribution of spam and non-spam e-mails is often different for different users a single filter trained on a general corpus is not optimal for all users. The question asked by ESPs is: How do you build robust and scalable automatic personalized spam filters? We address this question by presenting PSSF, a novel statistical approach for personalized service-side spam filtering. PSSF builds a discriminative classifier from a statistical model of spam and non-spam e-mails. A classifier is first built on a general training corpus that is then adapted in one or more passes of soft labeling and classifier rebuilding over each user's unlabeled e-mails. The statistical model captures the distribution of tokens in spam and non-spam e-mails. This model is robust in the sense that its size can be reduced significantly without degrading filtering performance. We evaluate PSSF on two datasets. The results demonstrate the superior performance and scalability of PSSF in comparison with other published results on the same datasets.