Search

The ever-evolving social media discourse has witnessed an overwhelming use of memes to express opinions or dissent. Besides being misused for spreading malcontent, they are mined by corporations and political parties to glean the public's opinion. Therefore, memes predominantly offer affect-enriched insights towards ascertaining the societal psyche. However, the current approaches are yet to model the affective dimensions expressed in memes effectively. They rely extensively on large multimodal datasets for pre-training and do not generalize well due to constrained visual-linguistic grounding. In this paper, we introduce MOOD (Meme emOtiOns Dataset), which embodies six basic emotions. We then present ALFRED (emotion-Aware muLtimodal Fusion foR Emotion Detection), a novel multimodal neural framework that (i) explicitly models emotion-enriched visual cues, and (ii) employs an efficient cross-modal fusion via a gating mechanism. Our investigation establishes ALFRED's superiority over existing baselines by 4.94% F1. Additionally, ALFRED competes strongly with previous best approaches on the challenging Memotion task. We then discuss ALFRED's domain-agnostic generalizability by demonstrating its dominance on two recently-released datasets – HarMeme and Dank Memes, over other baselines. Further, we analyze ALFRED's interpretability using attention maps. Finally, we highlight the inherent challenges posed by the complex interplay of disparate modality-specific cues toward meme analysis. [ABSTRACT FROM AUTHOR]

Sarcasm detection and humor classification are inherently subtle problems, primarily due to their dependence on the contextual and non-verbal information. Furthermore, existing studies in these two topics are usually constrained in non-English languages such as Hindi, due to the unavailability of qualitative annotated datasets. In this work, we make two major contributions considering the above limitations: (1) we develop a Hindi-English code-mixed dataset, MaSaC,1 for the multi-modal sarcasm detection and humor classification in conversational dialog, which to our knowledge is the first dataset of its kind; (2) we propose MSH-COMICS,2 a novel attention-rich neural architecture for the utterance classification. We learn efficient utterance representation utilizing a hierarchical attention mechanism that attends to a small portion of the input sentence at a time. Further, we incorporate dialog-level contextual attention mechanism to leverage the dialog history for the multi-modal classification. We perform extensive experiments for both the tasks by varying multi-modal inputs and various submodules of MSH-COMICS. We also conduct comparative analysis against existing approaches. We observe that MSH-COMICS attains superior performance over the existing models by $>$ > 1 F1-score point for the sarcasm detection and 10 F1-score points in humor classification. We diagnose our model and perform thorough analysis of the results to understand the superiority and pitfalls. [ABSTRACT FROM AUTHOR]

With the COVID-19 pandemic, online university education assumed greater importance. We propose a practical e-learning model with suggestions to augment online education of all forms to support effective technical education. We also share our experiences of implementations of some of these suggestions in online teaching. [ABSTRACT FROM AUTHOR]

Retweets are essential to boost the popularity of a tweet, and a large number of fake retweeters can contribute heavily to this aspect. We define a fake retweeter as a Twitter account that retweets spammy tweets, retweets an abnormally large amount of tweets in a short period, or misuses a trending hashtag to promote events irrelevant to the topic of discussion. We introduce an up-to-date, temporally diverse, trend-oriented labeled dataset to address the problem of fake retweeter detection. We develop a novel classifier, called HawkesEye which makes predictions based on a temporal window, in contrast to existing approaches which require a graph-like relationship between tweet entities, or the presence of the entire retweeting timeline of a retweeter. HawkesEye utilizes both temporal and textual information using a class-specific topic model and Hawkes processes. Experiments on our curated dataset show significant improvement over four state-of-the-art methods, with precision and recall scores of 0.964 and 0.960 on a balanced dataset, respectively – HawkesEye beats the best baseline by 6.16% and 25.98% relative improvement in terms of precision and recall, respectively. We also diagnose our model to understand the advantages and pitfalls of the underlying mechanism. We believe that the extent of this study is not restricted to Twitter, but generalizable to other social media systems such as Facebook and Instagram with similar reposting capabilities. [ABSTRACT FROM AUTHOR]

The growth of online social media has led to a huge increase in the number of users who want to share and publicize various kinds of information. Twitter, the most popular micro-blogging platform, has become a hotbed for users who are involved in different activities such as news publishing, job hunting, recruiting, advertising and publicity. Retweeting a tweet is a major action to broadcast a user’s message out to millions of users. Retweet action has two major advantages: (i) gaining quick exposure to the content, and (ii) increasing likelihood of gaining new Twitter followers in return. The organic way of gaining a larger number of retweets is a time consuming process, which leads to the creation of unfair methods to gain retweets. Thus, Twitter users often approach various blackmarket services to gain retweets inorganically in a short duration. Blackmarkets spread their collusive ecosystem in such a way that Twitter is unable to detect them even after devoting significant effort to purge the platform off bots, trolls, and fake accounts. One major reason behind the evasion is that the collusive users involved in blackmarket services exhibit a mix of organic and inorganic behavior – they organically reweet some genuine tweets; at the same time, they inorganically retweet tweets submitted to blackmarket services. This paper is the first attempt to provide a thorough study of the collusive users involved in two types of blackmarket services – Premium and Freemium. We collect a novel dataset of collusive users comprising of users from both types of blackmarket services. We provide network-centric, profile-centric, timeline-centric and retweet-centric characteristics of these users and show how users involved in premium blackmarket services exhibit diverse behavior as compared to those involved in freemium services. We further employ human annotators to label collusive users into three types: bots, promotional customers, and normal customers. We then curate 63 novel features to run state-of-the-art classifiers in two settings – binary classification (collusive vs. genuine) and multi-class classification (bot, promotional customers, normal customers, and genuine users). Bagging achieves the best accuracy (macro F1-score of 0.892) in the former setting, whereas Random Forest outperforms others (macro F1-score of 0.791) in the latter setting. We also develop a chrome extension, SCoRe++ which can detect collusive retweeters in real time. [ABSTRACT FROM AUTHOR]

"Community" in social networks is a nebulous concept. A community is generally assumed to be formed by people who possess similar attributes or characteristics, also known as "homophily". Although there has been a lot of research on community detection based on network topology, the semantic interpretation of communities is rarely studied. The present work aims to understand the behavioral similarity of users present in their personal neighborhood communities formed by friends, relatives, or colleagues, and addresses two fundamental questions: (i) Are communities formed by users who possess similar behavioral traits? If so, does this apply to all those sub-networks, i.e., friends, relatives, and colleagues? (ii) Does adding behavioral node-specific attributes/features to the nodes in a network lead to better community detection? To better understand the psycho-sociological homophilic nature of personal networks, the personalities and values of Twitter users were analyzed using the well-established "Big-5 personality model" and "Schwartz sociological behavior model". Empirical results based on the psychosociological behavior show that friends networks exhibit homophily, whereas relatives and colleagues networks do not exhibit such homophilic behavior. It can also be observed that neurotic people tend to behave heterogeneously with people of various personality traits. In addition, it is shown that such empirical evidence can be used as features for the tasks of community detection and link prediction. [ABSTRACT FROM AUTHOR]

In this paper, the authors seek to answer one fundamental question - what brings people together to form a community? In this article, they explore the personalities (psychological) and values (sociological) of individuals in social network communities in order to understand such natural selection. [ABSTRACT FROM AUTHOR]

We investigate whether off-the-shelf summarization algorithms can be combined to produce better quality summaries. To this end, we propose ensemble schemes that can combine the outputs of multiple base summarization algorithms, to produce summaries better than what is generated by any of the base algorithms. [ABSTRACT FROM AUTHOR]

Detection of non-overlapping and overlapping communities are essentially the same problem. However, current algorithms focus either on finding overlapping or non-overlapping communities. We present a generalized framework that can identify both non-overlapping and overlapping communities, without any prior input about the network or its community distribution. To do so, we introduce a vertex-based metric, GenPerm, that quantifies by how much a vertex belongs to each of its constituent communities. Our community detection algorithm is based on maximizing the GenPerm over all the vertices in the network. We demonstrate, through experiments over synthetic and real-world networks, that GenPerm is more effective than other metrics in evaluating community structure. Further, we show that due to its vertex-centric property, GenPerm can be used to unfold several inferences beyond community detection, such as core-periphery analysis and message spreading. Our algorithm for maximizing GenPerm outperforms six state-of-the-art algorithms in accurately predicting the ground-truth labels. Finally, we discuss the problem of resolution limit in overlapping communities and demonstrate that maximizing GenPerm can mitigate this problem. [ABSTRACT FROM AUTHOR]

We report the unusual and exotic phenomenon of positive exchange bias observed in all ferromagnetic layered soft/hard Fe/Co spring magnet systems. Interestingly, the results are well reproducible both at low (5 K) and room temperatures (300 K). The effect of applied magnetic pinning field of 7 and −5 T reveals a noticeable shift of magnetic hysteresis loops along the field direction. We argue that the reason behind observed positive exchange bias is the establishment of antiferromagnetic (AFM) exchange coupling at the interface as a result of intermixing of Fe into the Co layer. The collective coupling of this intermix layers is AFM which facilitates positive shift of magnetic hysteresis loop. [ABSTRACT FROM AUTHOR]