arXiv subjects

Marya Bazzi

Publications and source records attributed to Marya Bazzi.

12 recordsLinked to original sources

arXiv2025

Categorising SME Bank Transactions with Machine Learning and Synthetic Data Generation

Despite their significant economic contributions, Small and Medium Enterprises (SMEs) face persistent barriers to securing traditional financing due to information asymmetries. Cash flow lending has emerged as a promising alternative, but its effectiveness depends on accurate modelling of transaction-level data. The main challenge in SME transaction analysis lies in the unstructured nature of textual descriptions, characterised by extreme abbreviations, limited context, and imbalanced label distributions. While consumer transaction descriptions often show significant commonalities across individuals, SME transaction descriptions are typically nonstandard and inconsistent across businesses and industries. To address some of these challenges, we propose a bank categorisation pipeline that leverages synthetic data generation to augment existing transaction data sets. Our approach comprises three core components: (1) a synthetic data generation module that replicates transaction properties while preserving context and semantic meaning; (2) a fine-tuned classification model trained on this enriched dataset; and (3) a calibration methodology that aligns model outputs with real-world label distributions. Experimental results demonstrate that our approach achieves 73.49% (+-5.09) standard accuracy on held-out data, with high-confidence predictions reaching 90.36% (+-6.52) accuracy. The model exhibits robust generalisation across different types of SMEs and transactions, which makes it suitable for practical deployment in cash-flow lending applications. By addressing core data challenges, namely, scarcity, noise, and imbalance, our framework provides a practical solution to build robust classification systems in data-sparse SME lending contexts.

cs.CE↗

arXiv2025

Measuring the co-evolution of online engagement with (mis)information and its visibility at scale

Online attention is an increasingly valuable resource in the digital age, with extraordinary events such as the COVID-19 pandemic fuelling fierce competition around it. As misinformation pervades online platforms, users seek credible sources, while news outlets compete to attract and retain their attention. Here we measure the co-evolution of online ``engagement'' with (mis)information and its ``visibility'', where engagement corresponds to user interactions on social media, and visibility to fluctuations in user follower counts. Using over 100 million COVID-related retweets across 3 years, we analyse how user interactions and follower dynamics differ for factual, misleading and uncertain content. We observe that during major events (e.g., vaccine rollouts), users spreading factual content see rapid follower gain spikes, whereas those sharing misleading content tend to sustain faster growth outside of these high-attention periods. We introduce two scalable modelling frameworks (simple contagion and biased convergence) that reproduce many observed differing follower growth rates using temporal retweet network dynamics, providing evidence that content visibility co-evolves with user engagement. Our modelling lends itself to studying other large-scale events where online attention is at stake, such as climate and political debates.

cs.SI↗

arXiv2025

Dynamic benchmarking framework for LLM-based conversational data capture

The rapid evolution of large language models (LLMs) has transformed conversational agents, enabling complex human-machine interactions. However, evaluation frameworks often focus on single tasks, failing to capture the dynamic nature of multi-turn dialogues. This paper introduces a dynamic benchmarking framework to assess LLM-based conversational agents through interactions with synthetic users. The framework integrates generative agent simulation to evaluate performance on key dimensions: information extraction, context awareness, and adaptive engagement. By simulating various aspects of user behavior, our work provides a scalable, automated, and flexible benchmarking approach. Experimental evaluation - within a loan application use case - demonstrates the framework's effectiveness under one-shot and few-shot extraction conditions. Results show that adaptive strategies improve data extraction accuracy, especially when handling ambiguous responses. Future work will extend its applicability to broader domains and incorporate additional metrics (e.g., conversational coherence, user engagement). This study contributes a structured, scalable approach to evaluating LLM-based conversational agents, facilitating real-world deployment.

cs.CL↗

arXiv2024

Modelling and Predicting Online Vaccination Views using Bow-tie Decomposition

Social media has become increasingly important in shaping public vaccination views, especially since the COVID-19 outbreak. This paper uses bow-tie structure to analyse a temporal dataset of directed online social networks that represent the information exchange among anti-vaccination, pro-vaccination, and neutral Facebook pages. Bow-tie structure decomposes a network into seven components, with two components "SCC" and "OUT" emphasised in this paper: SCC is the largest strongly connected component, acting as an "information magnifier", and OUT contains all nodes with a directed path from a node in SCC, acting as an "information creator". We consistently observe statistically significant bow-tie structures with different dominant components for each vaccination group over time. In particular, the anti-vaccination group has a large OUT, and the pro-vaccination group has a large SCC. We further investigate changes in opinions over time, as measured by fan count variations, using agent-based simulations and machine learning models. Across both methods, accounting for bow-tie decomposition better reflects information flow differences among vaccination groups and improves our opinion dynamics prediction results. The modelling frameworks we consider can be applied to any multi-stance temporal network and could form a basis for exploring opinion dynamics using bow-tie structure in a wide range of applications.

cs.SI↗

arXiv2023

How to Data in Datathons

The rise of datathons, also known as data or data science hackathons, has provided a platform to collaborate, learn, and innovate in a short timeframe. Despite their significant potential benefits, organizations often struggle to effectively work with data due to a lack of clear guidelines and best practices for potential issues that might arise. Drawing on our own experiences and insights from organizing >80 datathon challenges with >60 partnership organizations since 2016, we provide guidelines and recommendations that serve as a resource for organizers to navigate the data-related complexities of datathons. We apply our proposed framework to 10 case studies.

cs.AI↗

arXiv2022

Local2Global: A distributed approach for scaling representation learning on graphs

We propose a decentralised "local2global"' approach to graph representation learning, that one can a-priori use to scale any embedding technique. Our local2global approach proceeds by first dividing the input graph into overlapping subgraphs (or "patches") and training local representations for each patch independently. In a second step, we combine the local representations into a globally consistent representation by estimating the set of rigid motions that best align the local representations using information from the patch overlaps, via group synchronization. A key distinguishing feature of local2global relative to existing work is that patches are trained independently without the need for the often costly parameter synchronization during distributed training. This allows local2global to scale to large-scale industrial applications, where the input graph may not even fit into memory and may be stored in a distributed manner. We apply local2global on data sets of different sizes and show that our approach achieves a good trade-off between scale and accuracy on edge reconstruction and semi-supervised classification. We also consider the downstream task of anomaly detection and show how one can use local2global to highlight anomalies in cybersecurity networks.

cs.LG↗

arXiv2021

DUKweb: Diachronic word representations from the UK Web Archive corpus

Lexical semantic change (detecting shifts in the meaning and usage of words) is an important task for social and cultural studies as well as for Natural Language Processing applications. Diachronic word embeddings (time-sensitive vector representations of words that preserve their meaning) have become the standard resource for this task. However, given the significant computational resources needed for their generation, very few resources exist that make diachronic word embeddings available to the scientific community. In this paper we present DUKweb, a set of large-scale resources designed for the diachronic analysis of contemporary English. DUKweb was created from the JISC UK Web Domain Dataset (1996-2013), a very large archive which collects resources from the Internet Archive that were hosted on domains ending in `.uk'. DUKweb consists of a series word co-occurrence matrices and two types of word embeddings for each year in the JISC UK Web Domain dataset. We show the reuse potential of DUKweb and its quality standards via a case study on word meaning change detection.

cs.CL↗

arXiv2021

Local2Global: Scaling global representation learning on graphs via local training

We propose a decentralised "local2global" approach to graph representation learning, that one can a-priori use to scale any embedding technique. Our local2global approach proceeds by first dividing the input graph into overlapping subgraphs (or "patches") and training local representations for each patch independently. In a second step, we combine the local representations into a globally consistent representation by estimating the set of rigid motions that best align the local representations using information from the patch overlaps, via group synchronization. A key distinguishing feature of local2global relative to existing work is that patches are trained independently without the need for the often costly parameter synchronisation during distributed training. This allows local2global to scale to large-scale industrial applications, where the input graph may not even fit into memory and may be stored in a distributed manner. Preliminary results on medium-scale data sets (up to $\sim$7K nodes and $\sim$200K edges) are promising, with a graph reconstruction performance for local2global that is comparable to that of globally trained embeddings. A thorough evaluation of local2global on large scale data and applications to downstream tasks, such as node classification and link prediction, constitutes ongoing work.

cs.LG↗

arXiv2020

Pull out all the stops: Textual analysis via punctuation sequences

Whether enjoying the lucid prose of a favorite author or slogging through some other writer's cumbersome, heavy-set prattle (full of parentheses, em dashes, compound adjectives, and Oxford commas), readers will notice stylistic signatures not only in word choice and grammar, but also in punctuation itself. Indeed, visual sequences of punctuation from different authors produce marvelously different (and visually striking) sequences. Punctuation is a largely overlooked stylistic feature in "stylometry", the quantitative analysis of written text. In this paper, we examine punctuation sequences in a corpus of literary documents and ask the following questions: Are the properties of such sequences a distinctive feature of different authors? Is it possible to distinguish literary genres based on their punctuation sequences? Do the punctuation styles of authors evolve over time? Are we on to something interesting in trying to do stylometry without words, or are we full of sound and fury (signifying nothing)?

cs.CL↗

arXiv2019

A Framework for the Construction of Generative Models for Mesoscale Structure in Multilayer Networks

Multilayer networks allow one to represent diverse and coupled connectivity patterns --- e.g., time-dependence, multiple subsystems, or both --- that arise in many applications and which are difficult or awkward to incorporate into standard network representations. In the study of multilayer networks, it is important to investigate mesoscale (i.e., intermediate-scale) structures, such as dense sets of nodes known as communities, to discover network features that are not apparent at the microscale or the macroscale. The ill-defined nature of mesoscale structure and its ubiquity in empirical networks make it crucial to develop generative models that can produce the features that one encounters in empirical networks. Key purposes of such generative models include generating synthetic networks with empirical properties of interest, benchmarking mesoscale-detection methods and algorithms, and inferring structure in empirical multilayer networks. In this paper, we introduce a framework for the construction of generative models for mesoscale structures in multilayer networks. Our framework provides a standardized set of generative models, together with an associated set of principles from which they are derived, for studies of mesoscale structures in multilayer networks. It unifies and generalizes many existing models for mesoscale structures in fully-ordered (e.g., temporal) and unordered (e.g., multiplex) multilayer networks. One can also use it to construct generative models for mesoscale structures in partially-ordered multilayer networks (e.g., networks that are both temporal and multiplex). Our framework has the ability to produce many features of empirical multilayer networks, and it explicitly incorporates a user-specified dependency structure between layers.

cs.SI↗

arXiv2019

Core-Periphery Structure in Directed Networks

While studies of meso-scale structures in networks often focus on community structure, core--periphery structures can reveal new insights. This structure typically consists of a well-connected core and a periphery that is well connected to the core but sparsely connected internally. Most studies of core--periphery structure focus on undirected networks. We propose a generalisation of core-periphery structure to directed networks. Our approach yields a family of core-periphery block model formulations in which core and periphery sets are edge-direction dependent. We mainly focus on a particular core--periphery structure consisting of two core sets and two periphery sets which we motivate empirically. To detect this directed core-periphery structure we propose four different methods, with different trade-offs between computational complexity and accuracy. We assess these methods on three benchmarks and compare to four standard methods. On simulated data, the proposed methods match or outperform the standard methods. Applying our methods to three empirical networks -- a political blogs networks, a faculty hiring network, and a trade network -- illustrates that this directed core--periphery structure can offer novel insights about the underlying dataset.

cs.SI↗

arXiv2017

Community detection in temporal multilayer networks, with an application to correlation networks

Networks are a convenient way to represent complex systems of interacting entities. Many networks contain "communities" of nodes that are more densely connected to each other than to nodes in the rest of the network. In this paper, we investigate the detection of communities in temporal networks represented as multilayer networks. As a focal example, we study time-dependent financial-asset correlation networks. We first argue that the use of the "modularity" quality function---which is defined by comparing edge weights in an observed network to expected edge weights in a "null network"---is application-dependent. We differentiate between "null networks" and "null models" in our discussion of modularity maximization, and we highlight that the same null network can correspond to different null models. We then investigate a multilayer modularity-maximization problem to identify communities in temporal networks. Our multilayer analysis only depends on the form of the maximization problem and not on the specific quality function that one chooses. We introduce a diagnostic to measure \emph{persistence} of community structure in a multilayer network partition. We prove several results that describe how the multilayer maximization problem measures a trade-off between static community structure within layers and larger values of persistence across layers. We also discuss some computational issues that the popular "Louvain" heuristic faces with temporal multilayer networks and suggest ways to mitigate them.

physics.soc-ph↗