Periodic Reporting for period 2 - FARE (FAKE NEWS AND REAL PEOPLE – USING BIG DATA TO UNDERSTAND HUMAN BEHAVIOUR)
Okres sprawozdawczy: 2022-04-01 do 2023-09-30
FARE asks what makes individuals share misinformation, focusing both on fundamental individual factors (such as cognitive bias), and on characteristics of their context (from socioeconomic components to position on social networks), that might make people particularly prone to such beliefs and sharing behavior.
FARE aims at answering four main questions:
1) What are the role(s) that cognitive biases play in FN susceptibility and spread?
2) What are the role(s) that network features play in FN dynamics?
3) How do cognitive biases and network features interact to impact on FN spread?
4) How can we tackle such questions in an ethical and society-focused manner?
This approach is carried out by an interdisciplinary team, bringing together state-of-the-art knowledge in behavioural psychology (to look at individuals), network science and computational social sciences (to study the effects of context), and epidemiology (to help develop and test spreading models). FARE is also developing new strategies, based on distributed computing, to improve current guidelines for the ethical handling of human-related big data and minimize ethical risks of research in the Digital Era.
FARE aims at answering four main questions:
1) What are the role(s) that cognitive biases play in FN susceptibility and spread?
2) What are the role(s) that network features play in FN dynamics?
3) How do cognitive biases and network features interact to impact on FN spread?
4) How can we tackle such questions in an ethical and society-focused manner?
This approach is carried out by an interdisciplinary team, bringing together state-of-the-art knowledge in behavioural psychology (to look at individuals), network science and computational social sciences (to study the effects of context), and epidemiology (to help develop and test spreading models). FARE is also developing new strategies, based on distributed computing, to improve current guidelines for the ethical handling of human-related big data and minimize ethical risks of research in the Digital Era.
FARE requires the establishment of three main datasets and two frameworks (one technical and the other theoretical).
The first dataset is the fake/real news database (yellow, in Figure 1). Information is selected for fact-checking by independent organizations that use different criteria and classification metrics. Therefore, the team has selected more than 300 fact-checkers, collected over 60,000 fact-checked pieces in almost 30 languages, from 2012 to 2023, and standardized their classification as “False”, “True” or “In-between”. This dataset is now being classified by topic and we expect to make it freely available by the end of 2023.
The second step serves to identify the classified fake/real news on a social network to estimate their spread and detect spreaders. From close to 40,000 reviewed items, we have identified more than 10,000 shares, corresponding to close to 3 million different user profiles. We are now in the process of selecting samples and collecting relevant information to complete this second dataset (grey, in Figure 1).
For the third dataset, and to explicitly test our hypothesis, we must identify individual or contextual specificities that may serve as good predictors of belief in misinformation. We are currently testing small pilot surveys and preparing to deploy a full-scale questionnaire, in an experimental context, by the end of the year (blue, in Figure 1).
Regarding the technical framework, we have created a schema to test a proof-of-concept system to analyse the data without crossing the datasets, to minimize ethical risks and protect the privacy of social network users (purple, in figure 1). This is now being implemented and will be deployed in 2025.
Finally, we are using some of the epidemiological models that we developed in the past (and during our efforts to fight the COVID-19 pandemic) to study how misinformation is spreading online (red, in Figure 1). This corresponds to one of our last aims and requires that previous tasks are completed before we can expect results.
The first dataset is the fake/real news database (yellow, in Figure 1). Information is selected for fact-checking by independent organizations that use different criteria and classification metrics. Therefore, the team has selected more than 300 fact-checkers, collected over 60,000 fact-checked pieces in almost 30 languages, from 2012 to 2023, and standardized their classification as “False”, “True” or “In-between”. This dataset is now being classified by topic and we expect to make it freely available by the end of 2023.
The second step serves to identify the classified fake/real news on a social network to estimate their spread and detect spreaders. From close to 40,000 reviewed items, we have identified more than 10,000 shares, corresponding to close to 3 million different user profiles. We are now in the process of selecting samples and collecting relevant information to complete this second dataset (grey, in Figure 1).
For the third dataset, and to explicitly test our hypothesis, we must identify individual or contextual specificities that may serve as good predictors of belief in misinformation. We are currently testing small pilot surveys and preparing to deploy a full-scale questionnaire, in an experimental context, by the end of the year (blue, in Figure 1).
Regarding the technical framework, we have created a schema to test a proof-of-concept system to analyse the data without crossing the datasets, to minimize ethical risks and protect the privacy of social network users (purple, in figure 1). This is now being implemented and will be deployed in 2025.
Finally, we are using some of the epidemiological models that we developed in the past (and during our efforts to fight the COVID-19 pandemic) to study how misinformation is spreading online (red, in Figure 1). This corresponds to one of our last aims and requires that previous tasks are completed before we can expect results.
Besides producing the datasets and frameworks described above, the FARE team has made contributions of relevant scientific impact.
Focusing on overconfidence (as disinformation spreading implies a disconnect between how much one knows and how much one thinks one knows) we developed a new indirect confidence metric, that does not rely on self-reporting or peer comparison, and applied it to four large surveys in Europe and USA, spanning a period of 30 years. These results show a deviation from the traditional view that less knowledgeable individuals have a bigger gap between what they know and what they think they know and have important consequences for identifying individuals more likely to be susceptible to disinformation, particularly when related to scientific knowledge.
The FARE team has also successfully applied to an ERC PoC, the ongoing FARE_Audit project (GA 101100653). This takes advantage of FARE’s fake/real news database and expands our work from social networks to search-engines, to further understand the role that search-engines might have in creating information bubbles. FARE_Audit is a collaborative effort with two NGOs and will lead to an auditing tool and an online interface aimed at journalists and disinformation researchers.
Focusing on overconfidence (as disinformation spreading implies a disconnect between how much one knows and how much one thinks one knows) we developed a new indirect confidence metric, that does not rely on self-reporting or peer comparison, and applied it to four large surveys in Europe and USA, spanning a period of 30 years. These results show a deviation from the traditional view that less knowledgeable individuals have a bigger gap between what they know and what they think they know and have important consequences for identifying individuals more likely to be susceptible to disinformation, particularly when related to scientific knowledge.
The FARE team has also successfully applied to an ERC PoC, the ongoing FARE_Audit project (GA 101100653). This takes advantage of FARE’s fake/real news database and expands our work from social networks to search-engines, to further understand the role that search-engines might have in creating information bubbles. FARE_Audit is a collaborative effort with two NGOs and will lead to an auditing tool and an online interface aimed at journalists and disinformation researchers.