Year: 2022
Type: dissertation
Abstract: Social Networking Sites (SNS) are the most important way of communication \nnowadays. They have changed how we interact with our friends and family, \nand even how companies target their clients, conduct market analysis and \nmake business decisions. The amount of data that is being generated every \nday is virtually unlimited, and it can be used to conduct social media analyses \nand/or to train Machine Learning (ML) models. However, many handicaps \nneed to be alleviated. SNS data is, typically, unstructured and written in \nnatural language, and it presents misspelled words, contractions, emojis, and \nnew semantic units that sometimes are a heavy burden for learning algorithms. \nA large dataset and multiple preprocessing steps are essential for almost any \nML application in SNS. \nUnfortunately, there is an inherent cost to gather and build labelled databases \n(human effort), and it constitutes a major drawback for low- to mid-budget \nventures. Additionally, many applications may result in social consequences, \nthus they need to be audited. Both objectives fall into the interest of a multidisciplinary \nproject called ª Nutcracker, that aims to detect, track, monitor an \nanalyse radical discourse online. This dissertation is part of the project, and we \npropose in it effortless and interpretable mechanisms to tackle aforementioned \ndisadvantages, using social network’s mechanics as leverage. First, we present \na reasoning mechanism based on similarity between users, that will allow us \nto deduce properties of unknown users, hence reducing the effort required to \nbuild databases. Then, we present a new kind of feature extraction and selection \nmethod whose purpose is to reduce model complexity, thus enhancing \nmodel comprehensibility and transparency. Finally, we study the peculiarities \nof aggregated analysis and, particularly, how well can class prevalence count be \nestimated when working with SNS data. \nOur results show that we are able to build large databases in Twitter with a \nfraction of the effort; that we can train interpretable models as accurate as the \nbaselines but one order of magnitude less complex; and that quantification is \na novel approach that has much to offer to social network analysis, since it \nis able to adjust classification bias. We developed a proof-of-concept tool for \neffortless labelling and continuous user tracking, and we tested the platform by \nproducing four high-quality weak-labelled datasets. The proposed techniques, \nmethodologies and tools have been proven useful for disciplines such as \ncomputational linguistics, political science and cybersecurity. They are being used by members of our team and they have raised the attention of Spanish \nCivil Guard. Applications include building (and working with) supervised \ndatabases (e.g., social network analysis, market analysis, customer service, \nuser profiling...); reaching full transparency in automatic decision-making \nalgorithms (e.g., preemptive account closing, illegal activity tracking, hiring \npolicies...); measuring overall user opinion or sentiment (e.g., during an \nevent like a political debate); studying mental illnesses, detection of epidemic \noutbreaks, targeting customers, profiling brand ambassadors, or determining \nthe impact of organised communities, among many others.
Author: Aparicio, Manuel Francisco
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Awards: G4641073222, G5193324772
OpenAlex ID: W7020044191