Phishing scams on the Ethereum network have become a serious threat, especially with the influx of new users into the cryptocurrency market. Current detection methods are mainly focused on long-term consistent transaction patterns with smooth temporal dynamics. However, these methods often struggle to differentiate between phishing and non-phishing users, whose behaviours may appear deceptively similar. Additionally, they face challenges such as network sparsity and data leakage, leading to significant performance limitations. To address these issues, we introduce TEMPER, a novel sequential learning framework designed to jointly capture the subtle distinctions between long- and short-term user behaviours and their correlations to provide more comprehensive insights. TEMPER effectively generates distinguishable user embeddings, enabling the accurate identification of phishing users. Unlike previous approaches, TEMPER mitigates data leakage through a novel sequential transaction sampling algorithm and addresses network sparsity with short-term temporal learning. Through extensive experimentation on three real-world Ethereum datasets, TEMPER demonstrates its efficacy by achieving a 3-4% improvement in the F1-Score compared to existing baseline models, representing a significant advancement in Ethereum phishing user detection.

Phishing scams on Ethereum have expanded with the surge of the platform, posing substantial challenges due to the sheer similarity in user behaviours and sparse temporal instances. Current methods often fail to tackle these concerns and overlook the temporal sequence of transactions, resulting in suboptimal performance. In this paper, we aim to address these gaps by focusing on the alignment of two aspects: (1) User-specific local temporal behavior, and (2) Divergences from global activity patterns of the network. Hence, we introduce CATALOG (CApturing joint TemporAl dependencies from LOcal and Global user behaviour), a novel representation learning model that jointly captures the local and global user behviours and their correlations by leveraging a dual cross-attention mechanism paired with a bi-directional Masked Language Modelling (MLM) transformer. Our proposed model simultaneously learns from local behavioral shifts, global market trends, and contextually enriched embeddings, effectively distinguishing phishing from non-phishing users while addressing existing research gaps. Extensive experiments on real-world Ethereum transaction data show that our framework improves phishing detection by 7-8% in the F1-Score along with demonstrating the generalization to Ethereum versions 1.0 and 2.0.

Traffic forecasting problems involve jointly modeling the non-linear spatio-temporal dependencies at different scales. While graph neural network models have been effectively used to capture the non-linear spatial dependencies, capturing the dynamic spatial dependencies between the locations remains a major challenge. The errors in capturing such dependencies propagate in modeling the temporal dependencies between the locations, thereby severely affecting the performance of long-term predictions. While transformer-based mechanisms have been recently proposed for capturing the dynamic spatial dependencies, these methods are susceptible to fluctuations in data brought on by unforeseen events like traffic congestion and accidents. To mitigate these issues we propose an improvised spatio-temporal parallel transformer (STPT) based model for traffic prediction that uses multiple adjacency graphs passed through a pair of coupled graph transformer-convolution network units, operating in parallel, to generate more noise-resilient embeddings. We conduct extensive experiments on 4 real-world traffic datasets and compare the performance of STPT with several state-of-the-art baselines, in terms of measures like RMSE, MAE, and MAPE. We find that using STPT improves the performance by around \(10-34\%\) as compared to the baselines. We also investigate the applicability of the model on other spatio-temporal data in other domains. We use a Covid-19 dataset to predict the number of future occurrences in different regions from a given set of historical occurrences. The results demonstrate the superiority of our model for such datasets.

Network alignment, the task of identifying the same entities across multiple networks, has recently gained immense popularity in industry and academia. However, most existing alignment methods consider networks static, merely a snapshot of the time-evolving real networks where the nodes, links, and attributes are bound to change over time. The existing methods cannot capture these intrinsic dynamic network changes and cannot be applied directly to the evolving real-network scenario. Even extending these static methods to update their model parameters by retraining dynamically suffers from high computational complexity or simple sequential training suffers from compromised predictions due to distribution drifts in the networks. Moreover, dealing with a pair of evolving networks poses extra challenges of their domain differences and different evolution rates and patterns. Hence to overcome these challenges, we propose EvoAlign, an end-to-end information replay-based continual learning framework built upon Graph Neural Networks (GNNs) for the alignment of evolving networks. EvoAlign employs the concept of uncertainty in model predictions to identify and address two key aspects: detecting new patterns and preserving historical patterns. It also uses a novel shift-induced regularizer to handle distribution drift and domain differences in evolving networks. We empirically show that our method outperforms the existing state-of-the-art alignment methods on three real datasets.

Signed link prediction in graphs is an important problem that has applications in diverse domains. It is a binary classification problem that predicts whether an edge between a pair of nodes is positive or negative. Existing approaches for link prediction in unsigned networks cannot be directly applied for signed link prediction due to their inherent differences. Furthermore, signed link prediction must consider the inherent characteristics of signed networks, such as structural balance theory. Recent signed link prediction approaches generate node representations using either generative models or discriminative models. Inspired by the recent success of Generative Adversarial Network (GAN) based models in several applications, we propose a GAN based model for signed networks, SigGAN. It considers the inherent characteristics of signed networks, such as integration of information from negative edges, high imbalance in number of positive and negative edges, and structural balance theory. Comparing the performance with state-of-the-art techniques on five real-world datasets validates the effectiveness of SigGAN.

Name ambiguity is a prevalent problem in scholarly publications due to the unprecedented growth of digital libraries and number of researchers. An author is identified by their name in the absence of a unique identifier. The documents of an author are mistakenly assigned due to underlying ambiguity, which may lead to an improper assessment of the author. Various efforts have been made in the literature to solve the name disambiguation problem with supervised and unsupervised approaches. The unsupervised approaches for author name disambiguation are preferred due to the availability of a large amount of unlabeled data. Bibliographic data contain heterogeneous features, thus recently, representation learning-based techniques have been used in literature to embed heterogeneous features in common space. Documents of a scholar are connected by multiple relations. Recently, research has shifted from a single homogeneous relation to multi-dimensional (heterogeneous) relations for the latent representation of document. Connections in graphs are sparse, and higher order links between documents give an additional clue. Therefore, we have used multiple neighborhoods in different relation types in heterogeneous graph for representation of documents. However, different order neighborhood in each relation type has different importance which we have empirically validated also. Therefore, to properly utilize the different neighborhoods in relation type and importance of each relation type in the heterogeneous graph, we propose attention-based multi-dimensional multi-hop neighborhood-based graph convolution network for embedding that uses the two levels of an attention, namely, (i) relation level and (ii) neighborhood level, in each relation. A significant improvement over existing state-of-the-art methods in terms of various evaluation matrices has been obtained by the proposed approach.

With the increasing use of Twitter for encouraging users to instigate violent behavior with hate and racial content, it becomes necessary to investigate the uniqueness in the dynamics of the spread of tweets made during violent communal incidents and the challenges they pose in early identification of potential viral content. In this article, we study the spread of the tweets made during several violent communal incidents along four major dimensions — the underlying follower network of the users, their structural and engagement characteristics, the cascades, and the cognitive aspects of the content, each of which plays a vital role in the spread of content. Using large public and collected data, we compare these features with tweets related to other subjects from several major domains, such as non-violent political events, celebrities, and technology, that contribute to a large fraction of the viral content over Twitter. We discover that while the spread of cascades and the users involved may provide strong early evidence of the viral content for several domains, the early phases of the spread of viral tweets related to violent communal incidents are characterized by cascades with protracted growth involving fringe or low-importance users, which would possibly make early prediction difficult. Our findings indicate that an interplay of certain network and cascade properties, together with the cognitive characteristics of tweets and the behavioral patterns of the engaging users, may provide stronger early indicators of the virality of this content.

Identifying tweets related to infrastructure damage during a crisis event is an important problem. However, the unavailability of labeled data during the early stages of a crisis event poses major challenge in training suitable models. Several domain adaptation strategies have been proposed for text classification that can be used to train models using available source data of previous crisis events and apply on a target data related to a current event. However, these approaches are insufficient to handle the distribution drift in the source and target data along with the class imbalance in the target data. In this paper we introduce an Ensemble learning approach with a Decoupled Adversarial (EnDeA) model to classify infrastructure damage tweets in a target tweet dataset. EnDeA is an ensemble of three different models two of which separately learn the event invariant and specific features of a target data from a set of source and target data. The third model which is an adversarial model helps to improve the prediction accuracy of both models. Unlike the existing approaches that also identify the domain invariant and specific properties of target data for sentiment classification, our method works for short texts and can better handle the distribution drift and class imbalance problem. We rigorously investigate the performance of the proposed approach using multiple public datasets and compare it with several state-of-the-art baselines. We discover that EnDeA outperforms these baselines with around 20% improvement in the 1 scores.

Author name disambiguation (AND) is a challenging problem due to several issues such as missing key identifiers, same name corresponding to multiple authors, along with inconsistent representation. Several techniques have been proposed but maintaining consistent accuracy levels over all data sets is still a major challenge. We identify two major issues associated with the AND problem. First, the namesake problem in which two or more authors with the same name publishes in a similar domain. Second, the diverse topic problem in which one author publishes in diverse topical domains with a different set of coauthors. In this work, we initially propose a method named ATGEP for AND that addresses the namesake issue. We evaluate the performance of ATGEP using various ambiguous name references collected from the Arnetminer Citation (AC) and Web of Science (WoS) data set. We empirically show that the two aforementioned problems are crucial to address the AND problem that are difficult to handle using state‐of‐the‐art techniques. To handle the diverse topic issue, we extend ATGEP to a new variant named ATGEP‐web that considers external web information of the authors. Experiments show that with enough information available from external web sources ATGEP‐web can significantly improve the results further compared with ATGEP.

Twitter provides important information for emergency responders in the rescue process during disasters. However, tweets containing relevant information are sparse and are usually hidden in a vast set of noisy contents. This leads to inherent challenges in generating suitable training data that are required for neural network models. In this paper, we study the problem of retrieving the infrastructure damage information from tweets generated from different location during crisis using the model actively trained on past but similar events. We combine RNN and GRU based model coupled with active learning that gets trained on most uncertain samples and captures the latent features of different data distribution. It reduces the uses of around 90% less training data, thereby significantly reducing the manual annotation efforts. We use the model pre-trained using active learning based approach to retrieve the infrastructure damage tweets originated from different regions. We obtain a minimum of 18% gain on F1-measure and considerably on other metrics over recent state-of-the-art IR techniques.

Retrieving relevant information from social media based on specific requirements has become a focus area for researchers. In this paper, we propose a framework for online retrieval of tweets providing information about possible infrastructure damages, caused due to earthquakes and use the same to determine a damage score for the possibly affected locations. Identifying such tweets would not only provide a holistic view of the affected areas but would also help in taking necessary relief actions. Existing works on this topic fail to effectively capture the semantic variation in the tweets, possibly due to poor content quality, thereby providing scopes for further improvement in the mechanisms involved. Our proposed technique relies on a novel split-query based mechanism along with a pseudo-relevance feedback approach to identify the relevant tweets. The pseudo-relevance feedback approach expands on an initial set of seed tweets obtained using a semi-automatic query generation mechanism that couples topic based clustering with human annotation. Empirical validation of our proposed method on a manually annotated ground truth data reveals a considerable improvement in precision, recall and mean average precision over several baseline methods.

Quantifying scientific productivity has traditionally been one of the major areas of research. An important component in measuring the scientific productivity of a researcher has been the number and citation count of his publications. However, that citation based measures of scientific productivity are influenced by domain specific factors like the popularity of the research domain, the key topics within the domain, as well as temporal factors like aging, these measures may not suitably reflect the contribution of a researcher uniformly across all domains. In this paper, we introduce social trust on a researcher in a given domain as a measure of his scientific productivity and success. We argue that trust in scientific domain is a social component that indicates the productivity and can influence several parameters like the collaborations of the researchers as well as the citations received by their publications. Unlike citation count of publications, trust is not domain specific and hence can be used uniformly across all domains to measure the scientific productivity. Our proposed measure of trust relies on a trust-based network of authors (nodes), where a link between two nodes is based on social indicators like co-authorship and citation counts. We validate the correctness as well as effectiveness of our proposed approach empirically using the ArnetMiner dataset. Observations indicate that the proposed measure not only mitigates the aging issues prevalent in citation based measures but can also predict the possibility of future success of the researchers in terms of citation count.

In this paper, we introduce social yield, a measure of collaboration success of the collaborating authors in a coauthorship network. We then attempt to empirically observe the link dynamics in collaboration networks induced by the social yield of the collaborations. Observation indicate that certain observed behavior like presence of large number of small sized communities and highly dynamic behavior of the links in collaboration networks can be explained based on the distribution of social yield of these collaborations. It is also observed that the distribution of social yield among the collaborations also affects the resilience of the collaboration networks to targeted link removal.