The emergence of interpersonal and social trust in online interactions

Prifti, Ylli (2023) The emergence of interpersonal and social trust in online interactions. PhD thesis, Birkbeck, University of London.

Abstract

My PhD work is in the area of extracting and modelling user-created data on the web. In particular, I focussed on locating and extracting user data that ’signals’ the evolution of human, 1-on-1 interactions between participants of large social networks who are forever stranger to each other. The booming of ”Online Social Networks” created an opportunity for social scientists to study social phenomena at a scale unseen before. The vast amount of information combined with computer science techniques led to significant developments in a relatively new field: Computational Social Science. Furthermore, in recent years the Gig Economy and mass adoption of ”business sharing” sites such as Airbnb, Uber, or JustEat drove a new wave of computational social science research into reviews, feedback, and recommendations. All these ingredients of the larger Social Trust have been vastly discussed in the literature, in both the social aspect and computational models of trust. However, some fundamental gaps remain, and there is often confusion about when trust is being expressed and how reviews (or recommendations) relate to social trust. Additionally, the computational trust models found in the literature tend to either be entirely theoretical or focused on a specific data set, thus lacking universal applicability. The latter problem, I believe, was due to a lack of data available to researchers in the early stages of the web. Today, the broader Online Social Networks have matured and consolidated mechanisms for allowing access to data. Access to information is rarely trivial for more specialised and smaller online communities. Yet smaller, focussed platforms are precisely where social trust and interactions could be observed (or not observed) and perhaps acquire a meaning that approaches the social trust social scientists see in in-person interactions. To address this gap, we initially propose and discuss the following research question: ”Is there a meeting point between online interactions and social trust so that the core components of trust are retained? ” We addressed this general open question by working on a computational architecture for data retrieval in social media platforms that can be suitably generalised and re-applied to different platforms. Lastly, as we enjoy the luxury of vast amounts of data that closely represent interpersonal and social trust, we addressed the question of ”what models of trust emerge from data” and ”how do existing models of trust perform with the data available”. I have defined a category of online social networks that retains the core components of social trust, which we call ”Online Social Networks of Needs.” Hence, I have a classification and categorisation mechanism for grouping online social networks of needs by the level of trust necessary for cooperation (aka. the cooperation threshold) and interactions to be triggered among participating cognitive agents. My focus has always been on data acquisition, and I have designed and implemented a system for data retrieval that is easily deployed to social media/social web platforms. A case study of such a system performing in a challenging scenario is further detailed to show the more extensive applicability of such a system for data retrieval and contribution to a scenario of complete distrust, anonymity, and ephemerality of data (such as 4chan.org). Further, studying the granularity of 4Chan data, we discovered that: 1. ephemerality is not sustained, and web archiving sites have a complete view of the ephemeral data [1], 2. we can track sentiment and topic modelling of moderation in 4chan [2], and 3. it is possible to have a live view of the topics and sentiment being discussed in the live board and see how these changes over time We studied the dynamics of high trust interactions [3] and found gender biases [4,5] in care interactions. Another topic related to trust but concerning institutions and media is the ’spillover’ effect between 4Chan and the traditional media. As a premise, 4chan anonymous threads have anticipated important global trends, notably the ”Anonymous” movement. Apart from the US, how do national topics interact with the essentially global discussion that is taking place there? Again, thanks to our extensive data collection/analysis, we sought to determine the level of participation from a selected non-US country, Norway, and the degree to which Norwegian 4chan /pol/ users and domestic news influence each other [6]. We continued the journey by collecting data from eight social networks of needs into the top two high trust demanding categories. Whilst these datasets are made available to researchers [7], we further study emerging networks and their properties and project the online social networks of needs into multiplex graphs by transforming the root links. Finally, we look into the applicability and predictive power of the non-reductionist model of trust proposed by Castelfranchi. We look at total social trust holistically and consider signals to evaluate fluctuations of the social capital influenced by economic and political dynamics and domination of the public discord by conspiracy theories. Summary of contributions 1. the first comprehensive real-time scrape of 4Chan (in literature, only post hoc solutions were available); 2. the application of Castelfranchi’s theoretical model of trust to actual data from online social networks; 3. one of the first studies on the relationship between the institutional (nationwide) press and extremisms on 4Chan; 4. the study of the application of predictive models to heterogeneous multi-source data (not user-created but not very trustable either), and 5. contributing live data scraping expertise into several other publications [8] [9]. Publications [1] Ylli Prifti, Iacopo Pozzana, and Alessandro Provetti. Live monitoring 4chan discussion threads. In 7th Int’l Conference on Computational Social Science, 2021. [2] Y. Prifti I. Pozzana and A. Provetti. On-line page scraping reveals evidence of moderation in 4chan/pol/ anonymous discussion threads. In Proc. of 3rd European Symposium on Societal Challenges in Computational Social Science. ETH Press, 2019. [3] Y. Prifti P. De Meo, I. Pozzana and A. Provetti. The dynamics of recommendation in high-trust personal care services. In 5th Int’l Conference on Computational Social Science (IC2S2), 2019. [4] Y. Prifti P. DeMeo, I. Pozzana and A. Provetti. Finding gender bias in web-based, high-trust interactions. In Proc. of 2nd European Symposium on Societal Challenges in Computational Social Science, GeWISS reports, 2018. [5] Y. Prifti P. De Meo, I. Pozzana and A. Provetti. Gender bias in web-based, high-trust interactions. In 5th Int’l Conference on Computational Social Science (IC2S2), 2019. [6] Alessandro Provetti Iacopo Pozzana, Ylli Prifti and Anders Seyersted Sandbu. Mapping the norwegian 4chan: How conspiracy theories travel the language barriers. In 7th Int’l Conference on Computational Social Science (IC2S2), 2021. [7] Ylli Prifti. 4chan /pol board as a temporary evolution of live threads and posts., July 2021. [8] Paschalis Lagias, George D. Magoulas, Ylli Prifti, and Alessandro Provetti. Predicting seriousness of injury in a traffic accident: A new imbalanced dataset and benchmark. In Lazaros Iliadis, Chrisina Jayne, Anastasios Tefas, and Elias Pimenidis, editors, Engineering Applications of Neural Networks - 23rd International Conference, EAAAI/EANN 2022, Chersonissos, Crete, Greece, June 17-20, 2022, Proceedings, volume 1600 of Communications [9] Andrea Ballatore, A. Pang, Iacopo Pozzana, Ylli Prifti, and Alessandro Provetti. Geo-referencing as a connector between user reviews and urban environment quality. In 5th Int’l Conference on Computational Social Science, 2019.

Metadata

Statistics

DownloadsShow export options

Export as XMLJSONCSV

Activity Overview

6 month trend

131Downloads

6 month trend

206Hits

Additional statistics are available via IRStats2.

Archive Staff Only (login required)

Edit/View Item