Polpo enables real-time bespoke information for political monitoring

Polpo observed that political analysts and other professionals using political and governmental information spent too much time manually searching for parliamentary information in disparate data sources. The process was expensive, time-consuming and susceptible to human error, and made it difficult to monitor political information relevant to specific individuals and organizations.

Resolving to address this clear and urgent need in the Dutch public sector, the company’s original founders set out to build a platform that provide automated support for parliamentary data searches—and enable professional users to spend more time on strategic decision-making.

Polpo arose from a fruitful collaboration between different areas of expertise, pooled by the company’s co-founders: Johan van Zaanen, Caspar Pille and Desmin Dekker. As Van Zaanen explains: “Not many people have an affinity with technology and linguistics and can apply it to political science. Every AI solution requires a domain expert who is familiar with text, style and standards, building a bridge between content and technology. This requires complementary expertise, often from more than one person.”

Joost Vos, Technical Specialist in NLP and Data Science at IBM, adds, “Even if the domain expert is not primarily tech-oriented, they still need to understand how to build algorithms. Their contribution needs to be translated into a technological architecture that more or less mimics human behavior. It requires partnerships that think outside the box, a different mindset, and a willingness to step outside the well-travelled path and understand each other’s mindset and professional context.”

Starting out with a small team, Polpo sought to aggregate the disparate data sources covering the huge volumes of parliamentary open data in the Netherlands. The aim was to capture and analyze the data stream in near real-time. Compounding the problem, the data was almost entirely comprised of text documents, which are inherently difficult to analyze using conventional software. Polpo therefore looked for solutions using an AI-based approach.

“We built our own landscape and dashboard,” Van Zaanen says. “The IBM technology is intelligent document processing. Input aggregated from online sources is stored in the database, which interacts with IBM Watson® for data enrichment and then displays the results in a dashboard.”

Supported by business partner You-Get and advised by IBM, Polpo used Watson Explorer technology combined with IBM Content Analytics Studio technology (now part of IBM Watson® Discovery) to aggregate disparate data sources and generate AI-based models. The desired end goal was a fully integrated, subscription-based web application for public affairs professionals, with built-in IBM Watson technology.

The biggest challenge at first was figuring out how to “filter out the static,” Van Zaanen explains. “Political terms are very context-sensitive, so words need to be mapped into context. The word ‘energy’, for example, is too ambiguous; education is not the same context as climate, or power companies. We would scan a text and identify context using secondary words. IBM Watson Explorer mimics that process. The smart nature of IBM Watson makes it possible to keep identifying new things; even as some parts evolve, it’s still aligned.”

“Ethical AI and transparency are built-in principles,” Vos adds. “There’s always a human in the loop. It’s not self-propagating.”

“The algorithm is actually part of a solution landscape,” Van Zaanen says. “First we built our own dashboard, and then we used IBM’s own miner with the Watson technology and IBM interface as a point of comparison to see what’s in the data, achieving new insights as inspiration for building features. The front-end is also self-built, designed based on iterative user feedback. Polpo adopted a continuous development model in which client feedback defines the roadmap.”

The Polpo Political Monitoring platform brings together all the relevant political information in one place, applies intelligent document processing through AI, and allows its users to apply smart filtering. By ensuring that users receive accurate, up-to-date and customized information that is relevant to them, Polpo facilitates all professionals who monitor political information by ensuring prompt and comprehensive information on all political developments in the Netherlands.

The company and its platform are still evolving. Van Zaanen explains, “Polpo has an identity and culture of continuous development. The platform is built to constantly accommodate new dashboards and sources: agile by design. As the platform and code grow, it becomes more difficult, but development time isn’t increasing exponentially; we release an update every two weeks. We’re focused on fast and incremental improvement, synthesizing feedback into implementable change.”

“The biggest changes in the model may well be the major topics that shift around elections,” Vos adds. “But it is possible to search for concurrences and extrapolate from the data.”

Having grown from 5 initial co-founders to its current 12-person team, Polpo has moved out of the start-up phase and is still growing rapidly. Even so, the company is quite small to be using such a large-scale IBM product, as Vos explains: “Polpo is relatively unique in applying this technology in the Netherlands, although IBM Watson does support Dutch language solutions. The best results are from start-ups or scale-ups. They achieve more added value than bigger companies, because the adoption curve is faster. Smaller, more innovative, experimental teams are more willing to embrace the added value of a digital service.”