Periodic Reporting for period 1 - Politus (AI-based Data Platform for Fair Social Policies)
Periodo di rendicontazione: 2022-12-01 al 2024-05-31
The politics of contemporary welfare state development in emerging market economies has already been investigated by ERC researchers. Building on this research, the new ERC-funded Politus project will explore the commercial viability of proprietary natural language processing and machine learning technology for AI-based data-driven fair social policymaking. It will address two interconnected problems of opposition political parties and municipalities in Turkey and other emerging markets. The first is the abuse of public social policies with the aid of big data. The second is the government-led information gap that opposition parties and municipalities cannot fill without access to big data. Politus will provide the tools needed to uncover discretionary and clientelist allocation of social policies.
The Politus Project successfully developed and implemented a novel technology to automatically extract public opinion in a privacy-preserving and GDPR-compliant manner, revolutionizing the crisis-ridden public opinion research industry and providing alternative fair social policies for the poor. This innovative technology integrates state-of-the-art AI methods, anonymization techniques, and legal/ethical perspectives to ensure the protection of data owners' privacy while providing valuable insights into public opinion trends.
The Politus Project aimed to address the fundamental challenge of reconciling the need to protect data rights and privacy with the necessity of understanding public opinion. The project sought to create a GDPR-compliant methodology that leverages big data for public opinion research without violating these regulations while using these data and methods to design fair social policies that would ensure the political integrity of citizens.
The Politus Project employed a combination of advanced AI techniques, including natural language processing (NLP), multimodal deep learning, network analysis, and statistical methods, to extract and analyze public opinion from user-generated content on online social platforms. To ensure privacy preservation, the project utilized differential privacy (DP) and local differential privacy (LDP). These methods were adapted to handle the unstructured nature of the data, addressing challenges such as the "vocal user problem" and the need for adaptable and updateable models. To overcome the demographic bias and non-probabilistic nature of the data, the project employed Bayesian multilevel regression with poststratification (MRP). This approach ensured that the data was representative of the population. The project also built advanced models driven by deep learning, network analysis, and amplified learning, enhancing the informativeness and validity of the data.
The Politus Project has employed an interdisciplinary team of mathematicians, computer scientists, sociologists, political scientists, and psychologists. Politus team has worked with a number of stakeholders, including municipalities, political parties, NGOs, universities and corporations:
1. Istanbul Metropolian Municipality
2. Kadıköy Municipality
3. Republican People’s Party
4. Oxfam International and the Refugee Council of Turkey
5. University of Pittsburgh
6. Center for Democracy and Technology, Washington DC, USA.
7. Publicis One
8. Hepsiburada
9. Castrol
10. Arçelik
The project team worked to scale-up Politus Analytics (politusanalytics.com) as an artificial intelligence company established within the scope of the Politus Project. By automatically analyzing digital traces on social media with natural language processing, network analysis and deep learning methods, Politus Analytics conducts public opinion research and provides information about human behaviour at high granularity over time and space, and demography. This includes details of behaviour, such as propensity to purchase, recommend or complaint, stance towards sectors, companies, brands, and consumer characteristics. It also includes voter behaviour and public opinion dynamics such as ideologies, values, beliefs, emotions, political issues, or job approval rates of local and national political leaders. Politus uses MRP to render its data representative and its data pipeline is GDPR compliant. The company share its data and insights with its subscribers through its advanced modular data platform.
The Politus Project aimed to address the fundamental challenge of reconciling the need to protect data rights and privacy with the necessity of understanding public opinion. The project sought to create a GDPR-compliant methodology that leverages big data for public opinion research without violating these regulations while using these data and methods to design fair social policies that would ensure the political integrity of citizens.
The Politus Project employed a combination of advanced AI techniques, including natural language processing (NLP), multimodal deep learning, network analysis, and statistical methods, to extract and analyze public opinion from user-generated content on online social platforms. To ensure privacy preservation, the project utilized differential privacy (DP) and local differential privacy (LDP). These methods were adapted to handle the unstructured nature of the data, addressing challenges such as the "vocal user problem" and the need for adaptable and updateable models. To overcome the demographic bias and non-probabilistic nature of the data, the project employed Bayesian multilevel regression with poststratification (MRP). This approach ensured that the data was representative of the population. The project also built advanced models driven by deep learning, network analysis, and amplified learning, enhancing the informativeness and validity of the data.
The Politus Project has employed an interdisciplinary team of mathematicians, computer scientists, sociologists, political scientists, and psychologists. Politus team has worked with a number of stakeholders, including municipalities, political parties, NGOs, universities and corporations:
1. Istanbul Metropolian Municipality
2. Kadıköy Municipality
3. Republican People’s Party
4. Oxfam International and the Refugee Council of Turkey
5. University of Pittsburgh
6. Center for Democracy and Technology, Washington DC, USA.
7. Publicis One
8. Hepsiburada
9. Castrol
10. Arçelik
The project team worked to scale-up Politus Analytics (politusanalytics.com) as an artificial intelligence company established within the scope of the Politus Project. By automatically analyzing digital traces on social media with natural language processing, network analysis and deep learning methods, Politus Analytics conducts public opinion research and provides information about human behaviour at high granularity over time and space, and demography. This includes details of behaviour, such as propensity to purchase, recommend or complaint, stance towards sectors, companies, brands, and consumer characteristics. It also includes voter behaviour and public opinion dynamics such as ideologies, values, beliefs, emotions, political issues, or job approval rates of local and national political leaders. Politus uses MRP to render its data representative and its data pipeline is GDPR compliant. The company share its data and insights with its subscribers through its advanced modular data platform.
1. The project successfully developed a generalizable method to extract public opinion from online social platforms. This method combines NLP, multimodal deep learning, amplified learning, network analysis, and statistics.
2. The project created a data platform that delivers representative, high-frequency, multilingual, and multi-country longitudinal data on public opinion trends. This platform aggregates and automatically analyzes sensitive digital trace data from online social platforms.
3. The project developed a new technology for privacy-preserving processing of sensitive information using generative AI and advanced anonymization methods, including local differential privacy and privacy-preserving record linkage principles.
4. The platform provides population-level projections on public opinion, such as political ideologies, beliefs, approval/support ratings, and voting preferences. These projections are geolocated, allowing for local and national-level analyses, and disaggregated by key demographic variables.
5. The involvement of social scientists with domain expertise ensured the informativeness and validity of the data, providing in-depth insights into public opinion trends.
6. The project has worked with Istanbul Metropolitan Municipality and the Republication People’s Party to propose alternative social policies for the poor.
2. The project created a data platform that delivers representative, high-frequency, multilingual, and multi-country longitudinal data on public opinion trends. This platform aggregates and automatically analyzes sensitive digital trace data from online social platforms.
3. The project developed a new technology for privacy-preserving processing of sensitive information using generative AI and advanced anonymization methods, including local differential privacy and privacy-preserving record linkage principles.
4. The platform provides population-level projections on public opinion, such as political ideologies, beliefs, approval/support ratings, and voting preferences. These projections are geolocated, allowing for local and national-level analyses, and disaggregated by key demographic variables.
5. The involvement of social scientists with domain expertise ensured the informativeness and validity of the data, providing in-depth insights into public opinion trends.
6. The project has worked with Istanbul Metropolitan Municipality and the Republication People’s Party to propose alternative social policies for the poor.