Abstract

Socio-political protests often lead to grave consequences when they occur. The early detection of such protests is very important for taking early precautionary measures. However, the main shortcoming of protest event detection is the scarcity of sufficient training data for specific language categories, which makes it difficult to train data-hungry deep learning models effectively. Therefore, cross-lingual and zero-shot learning models are needed to detect events in various low-resource languages. This paper proposes a multi-lingual crossdocument level event detection approach using pre-trained transformer models developed for Shared Task 1 at CASE 2022. The shared task constituted four subtasks for event detection at different granularity levels, i.e., document level to token level, spread over multiple languages (English, Spanish, Portuguese, Turkish, Urdu, and Mandarin). Our system achieves an average F1 score of 0.73 for document-level event detection tasks. Our approach secured 2 nd position for the Hindi language in subtask 1 with an F1 score of 0.80. While for Spanish, we secure 4 th position with an F1 score of 0.69. Our code is available at https://github. com/nehapspathak/campros/.

Abstract

Online social media platforms have evolved into a significant place for debate around socio-political phenomena such as government policies and bills. Studying online debates on such topics can help infer people’s perception and acceptance of the happenings. At the same time, various inauthentic users that often pollute the democratic discussion of the subject need to be weeded out from the debate. The characterization of a campaign keeping in mind various forms of involved actors thus becomes very important. On December 12, 2019, Citizenship Amendment Act (CAA) was enacted by the Indian Government, triggering a debate on whether the act was unfair. In this work, we investigate the user’s perception of the #CitizenshipAmendmentAct on Twitter, as the campaign unrolled with divergent discourse in the country. Keeping the campaign participants as the prime focus, we study 9,947,814 tweets produced by 275,111 users during the starting 3 months of protest. Our study includes the analysis of user engagement, content, and network properties with online accounts divided into authentic (genuine users) and inauthentic (bots, suspended, and deleted) users. Our findings show different themes in shared tweets among protesters and counter-protesters. We find presence of inauthentic users on both side of discourse, with counter-protesters having more inauthentic users than protesters. The follow network of the users suggests homophily among users on the same side of discourse and connection between various inauthentic and authentic users. This work contributes to filling the gap of understanding the role of users (from both sides) in a less studied geo-location, India.

Abstract

Polysemy is the phenomenon where a single word form possesses two or more related senses. It is an extremely ubiquitous part of natural language and analyzing it has sparked rich discussions in the linguistics, psychology and philosophy communities alike. With scarce attention paid to polysemy in computational linguistics, and even scarcer attention toward quantifying polysemy, in this paper, we propose a novel, unsupervised framework to compute and estimate polysemy scores for words in multiple languages. We infuse our proposed quantification with syntactic knowledge in the form of dependency structures. This informs the final polysemy scores of the lexicon motivated by recent linguistic findings that suggest there is an implicit relation between syntax and ambiguity/polysemy. We adopt a graph based approach by computing the discrete Ollivier Ricci curvature on a graph of the contextual nearest neighbors. We test our framework on curated datasets controlling for different sense distributions of words in 3 typologically diverse languages - English, French and Spanish. The effectiveness of our framework is demonstrated by significant correlations of our quantification with expert human annotated language resources like WordNet. We observe a 0.3 point increase in the correlation coefficient as compared to previous quantification studies in English. Our research leverages contextual language models and syntactic structures to empirically support the widely held theoretical linguistic notion that syntax is intricately linked to ambiguity/polysemy

Abstract

Music sharing trends have been shown to change during times of socio-economic crises. Studies have also shown that music can act as a social surrogate, helping to significantly reduce loneliness by acting as an empathetic friend. We explored these phenomena through a novel study of online music sharing during the Covid-19 pandemic in India. We collected tweets from the popular social media platform Twitter during India’s first and second wave of the pandemic (n = 1,364). We examined the different ways in which music was able to accomplish the role of a social surrogate via analyzing tweet text using Natural Language Processing techniques. Additionally, we analyzed the emotional connotations of the music shared through the acoustic features and lyrical content and compared the results between pandemic and pre-pandemic times. It was observed that the role of music shifted to a more community focused function rather than tending to a more self-serving utility. Results demonstrated that people shared music during the Covid-19 pandemic which had lower valence and shared songs with topics that reflected turbulent times such as Hardship and Exclusion when compared to songs shared during pre-Covid times. The results are further discussed in the context of individualistic versus collectivistic cultures.

Abstract

Code mixing is the linguistic phenomenon where bilingual speakers tend to switch between two or more languages in conversations. Recent work on code-mixing in computational settings has leveraged social media code mixed texts to train NLP models. For capturing the variety of code mixing in, and across corpus, Language ID (LID) tags based measures (CMI) have been proposed. Syntactical variety/patterns of code-mixing and their relationship vis-a-vis computational model’s performance is under explored. In this work, we investigate a collection of English(en)-Hindi(hi) code-mixed datasets from a syntactic lens to propose, SyMCoM, an indicator of syntactic variety in code-mixed text, with intuitive theoretical bounds. We train SoTA en-hi PoS tagger, accuracy of 93.4%, to reliably compute PoS tags on a corpus, and demonstrate the utility of SyMCoM by applying it on various syntactical categories on a collection of datas

Abstract

Code-Mixing is a phenomenon of mixing two or more languages in a speech event and is prevalent in multilingual societies. Given the low-resource nature of Code-Mixing, machine generation of code-mixed text is a prevalent approach for data augmentation. However, evaluating the quality of such machine generated code-mixed text is an open problem. In our submission to HinglishEval, a sharedtask collocated with INLG2022, we attempt to build models factors that impact the quality of synthetically generated code-mix text by predicting ratings for code-mix quality. HinglishEval Shared Task consists of two sub-tasks - a) Quality rating prediction); b) Disagreement prediction. We leverage popular codemixed metrics and embeddings of multilingual large language models (MLLMs) as features, and train task specific MLP regression models. Our approach could not beat the baseline results. However, for Subtask-A our team ranked a close second on F-1 and Cohen’s Kappa Score measures and first for Mean Squared Error measure. For Subtask-B our approach ranked third for F1 score, and first for Mean Squared Error measure. Code of our submission can be accessed here.

Abstract

Many populous countries including India areburdened with a considerable backlog of legalcases. Development of automated systems thatcould process legal documents and augmentlegal practitioners can mitigate this. However,there is a dearth of high-quality corpora that isneeded to develop such data-driven systems.The problem gets even more pronounced inthe case of low resource languages such asHindi. In this resource paper, we introducetheHindi Legal Documents Corpus (HLDC), acorpus of more than 900K legal documents inHindi. Documents are cleaned and structuredto enable the development of downstream ap-plications. Further, as a use-case for the cor-pus, we introduce the task of bail prediction.We experiment with a battery of models andpropose a Multi-Task Learning (MTL) basedmodel for the same. MTL models use sum-marization as an auxiliary task along with bailprediction as the main task. Experiments withdifferent models are indicative of the needfor further research in this area. We releasethe corpus and model implementation codewith this paper:https://github.com/Exploration-Lab/HLDC

Abstract

In this paper, we study the role of nodes and edges in a complex network in dictating the robustness of a community structure towards structural perturbations. Specifically, we attempt to identify all vital nodes, which, when removed, would lead to a large change in the underlying community structure of the network. This problem is critical because the community structure of a network allows us to explore deep underlying insights into how the function and topology of the network affect each other. Moreover, it even provides a way to condense large net- works into smaller modules where each community acts as a meta node and aids in more straightforward network analysis. If the community structure were to be compromised by either accidental or intentional per- turbations to the network, that would make such analysis difficult. Since identifying such vital nodes is computationally intractable, we propose a suite of heuristics that allow to find solutions close to the optimality. To show the effectiveness of our approach, we first test these heuristics on small networks and then move to more extensive networks to show that we achieve similar results. Further analysis reveals that the pro- posed approaches are useful to analyze the vulnerability of communities in networks irrespective of their size and scale. Additionally, we show the performance through an extrinsic evaluation framework – we employ two tasks, i.e., link prediction and information diffusion, and show that the effect of our algorithms on these tasks is higher than the other baselines.

Abstract

ackground: The adoption of nonpharmaceutical interventions and their surveillance are critical for detecting and stopping possible transmission routes of COVID-19. A study of the effects of these interventions can help shape public health decisions. The efficacy of nonpharmaceutical interventions can be affected by public behaviors in events, such as protests. We examined mask use and mask fit in the United States, from social media images, especially during the Black Lives Matter (BLM) protests, representing the first large-scale public gatherings in the pandemic. Objective: This study assessed the use and fit of face masks and social distancing in the United States and events of large physical gatherings through public social media images from 6 cities and BLM protests. Methods: We collected and analyzed 2.04 million public social media images from New York City, Dallas, Seattle, New Orleans, Boston, and Minneapolis between February 1, 2020, and May 31, 2020. We evaluated correlations between online mask usage trends and COVID-19 cases. We looked for significant changes in mask use patterns and group posting around important policy decisions. For BLM protests, we analyzed 195,452 posts from New York and Minneapolis from May 25, 2020, to July 15, 2020. We looked at differences in adopting the preventive measures in the BLM protests through the mask fit score. Results: The average percentage of group pictures dropped from 8.05% to 4.65% after the lockdown week. New York City, Dallas, Seattle, New Orleans, Boston, and Minneapolis observed increases of 5.0%, 7.4%, 7.4%, 6.5%, 5.6%, and 7.1%, respectively, in mask use between February 2020 and May 2020. Boston and Minneapolis observed significant increases of 3.0% and 7.4%, respectively, in mask use after the mask mandates. Differences of 6.2% and 8.3% were found in group pictures between BLM posts and non-BLM posts for New York City and Minneapolis, respectively. In contrast, the differences in the percentage of masked faces in group pictures between BLM and non-BLM posts were 29.0% and 20.1% for New York City and Minneapolis, respectively. Across protests, 35% of individuals wore a mask with a fit score greater than 80%. Conclusions: The study found a significant drop in group posting when the stay-at-home laws were applied and a significant increase in mask use for 2 of 3 cities where masks were mandated. Although a positive trend toward mask use and social distancing was observed, a high percentage of posts showed disregard for the guidelines. BLM-related posts captured the lack of seriousness to safety measures, with a high percentage of group pictures and low mask fit scores. Thus, the methodology provides a directional indication of how government policies can be indirectly monitored through social media

Abstract

Twitter is an extensively used micro-blogging site for publishing user’s views on recent happenings. This wide reachability of messages over large audience poses a threat, as the degree of personally identifiable information disclosed might lead to user regrets. The Tweet-Scan-Post system scans the tweets contextually for sensitive messages. The tweet repository was generated using cyber- keywords for personal, professional and health tweets. The Rules of Sensitivity and Contextuality was defined based on standards established by various national regulatory bodies. The naive sensitivity regression function uses the Bag-of-Words model built from short text messages. The imbalanced classes in dataset result in misclassification with 25% of sensitive and 75% of insensitive tweets. The system opted stacked classification to combat the problem of imbalanced classes. The system initially applied various state-of-art algorithms and predicted 26% of the tweets to be sensitive. The proposed stacked classification approach increased the overall proportion of sensitive tweets to 35%. The system contributes a vocabulary set of 201 Sensitive Privacy Keyword using the boosting approach for three tweet categories. Finally, the system formulates a sensitivity scaling called TSP’s Tweet Sensitivity Scale based on Senti-Cyber features composed of Sensitive Privacy Keywords, Cyber-keywords with Non-Sensitive Privacy Keywords and Non-Cyber-keywords to detect the degree of disclosed sensitive information.