UN Women continuously aims to address the needs of women and girls and make their work effective and evidence-based. The organization partnered with UN Global Pulse to explore how it could better leverage big data to improve its monitoring and evaluation work. The project involved three case studies that used big data analysis to:
- Understand the reach of UN Women’s social media campaigns to enhance advocacy;
- Monitor online radio in migration zones as part of a Rapid Gender Analysis in Chile;
- Create a guide for monitoring and studying violence against women in politics on social media platforms.
Our impact
UN Women was able to strengthen their applied research by making use of big data and technology. In Chile, radio mining provided information on attitudes towards migrants in local communities and challenges facing migrant women. These findings went into a report that was requested by the Ministry of Women and Gender Equality in Chile and UN Women used the experience to inform the development of strategies on social cohesion programming.
The result in another case was a methodological guide to help country teams conduct studies to expose violence against women in politics on social media. Combining UN Women’s expertise in gender and political participation with Global Pulse’s data science knowledge led to a more structured methodology for such studies and a framework for social media analysis of violence against women. The guide is now being shared across UN Women’s network of offices in the Latin American and Caribbean region for future application.
Our lessons
- Interdisciplinary teams are a must: While data science and engineering teams can bring the technical expertise needed to handle large data streams, they often lack an understanding of gender-related issues or the specifics of the region. Effective teams bring together both types of expertise.
- Quantitative and qualitative methods should be combined: Quantitative approaches can filter down data and identify items of interest in large volumes of information. Qualitative approaches give context, interpretation and meaning to the data. They can also help to improve quantitative methods, e.g. when human analysts label datasets that can then be used to train algorithms.
- Adaptation and refinement will be required: Since data science approaches rarely achieve the intended performance straight away, continuous adaptation and refinement is needed. It is important to get early input and feedback from the teams, and ensure alignment.
- Big data analysis can struggle to identify precise categories: We could, for example, identify radio conversations related to migration but it was difficult to identify conversations related to migration and gender and mental health. Similarly, our tools could pick out hate speech, political discussion and gender in social media but struggled to identify gender-based political hate speech.
- Bias remains a problem: Some people and types of conversation are not represented on social media or radio at all. At the same time, it can be difficult to extract data from relevant channels such as TikTok, making it hard to get a full picture of social media discussions. And each platform/data source is associated with its own distinct biases.