We wanted to build upon the SCB data and product in order to give a friendlier customer-facing side to open data on politics. We know that the historical and live data available is important but both can become redundant to the public if citizens cannot access it to make their own decisions.
We heard from Pierre from Dinriksdag, he was telling us when he makes freedom for information requests the process to access the data he needs is very frustrating. He has to go to a government office, only to find that he will given the documents he requires as scanned images and delivered over USB stick. There should be an easier way to access the data available, right now the barriers to entry are high, too high.
Dinriksdag has the forum and feedback loops build into it's product but we felt adding a layer of data visualisation will help citizens be able to access objective information about their political representatives as easy as they would a newspaper.
What it does
This web app collects data about politicians words and actions in areas the citizens are interested in and allows users to easily visualize the statistics in areas such as education, housing. By comparing politicians words and actions to the actual statistics, it allows users to easily evaluate the credibility of politicians.
How we built it
We divided into three teams. Ioan and Oliver on text analysis, Li and Erlend on visualization and Sophie on web. Using the Microsoft Cognitive Services API and cutting edge AI, we classify the statements of politicians into pre-defined categories. The visualization, built using python Bokeh, uses open data from selected sources in conjunction with the analyzed statements from the politicians. Finally, the web interface is build using the Materialize CSS framework.
Challenges we ran into
We found it difficult to navigate the nuances of politics when designing a workable product, the nature of politics itself and how we traditionally interact with this structure of power meant that we had to reformulate what we aimed to produce.
We deciphered that we could plot the promises and the sentiment analysis of a politician, and how they voted on those promises - but it is never a sure thing that the action they take will result in the outcome they and their constituents desire. Therefore we did not add a summary reporting on the percentages of promises kept. Instead, we offered an organised but summary-less view of the data, a clear signposted continuing timeline of political intent and actions.
Sophie - Using web development at a hackathon for the first time as well as managing the vision for the product, bringing together insights from different sourses.
Oliver - Finding good pre-trained NLP models for the Swedish language.
Li - learning a new Python visualization library.
Erlend - collaborate to find an idea that everyone wanted to work with.
Ioan - Deploying to an AWS Lambda instance.
Accomplishments that we're proud of
Sophie - Applying my learning from my coding boot camp to build the front end part of our product.
Li - Collaborating with my new family; building interactive visualization.
Oliver - Applying my knowledge of NLP to make political and government data more understandable, and winning three pineapples.
Erland - Design a presentation that looks decent for once in my life and keep everyone happy throughout the hackathon.
Ioan - Learning more about NLP, Microsoft Cognitive Services APIs and AWS Serverless
What we learned
Web development with Materialize, implementing AI features with Microsoft Azure Cognitive Services APIs and word embeddings for topic modelling, interactive visualization with Python Bokeh, AWS Serverless deployment.
What's next for Full Stack Of Politics
After going to the conference on Civic tech we heard that it's important to create a feedback loop built into every open data project that is created, the next step for Full Stack of Politics would be to iterate upon, test, and react to feedback given from users.