Our Motivation

Nowadays, political fronts are more hardened than ever, and pigeonholing is widespread. We show that you can belong to several "pigeonholes" at the same time in this spectrum. In this way, we actively help in political decision-making, and do so in an uncomplicated and modern way: Automated, yet transparent. Everyone is political, even if they wouldn't say so about themselves. This doesn't necessarily mean that we regularly express ourselves on current political topics - we also leave traces of our guiding principles and convictions in other ways. Twitter in particular is full of conscious as well as subconscious expression. Chirpanalytica is a tool that can classify these opinions.

The Technology behind Chirpanalytica

Our training dataset consists of about 70,000 tweets of public, German politician accounts (the accounts are found by a Wikidata query). After some preparation, these are stored in a common .csv file together with the factions of the respective politician. In total, our project consists of five parts: the tweet downloader (using Twitter API V2 and a Wikidata query), the neural network training function (scikit-learn), the backend (Python: Waitress), the frontend (HTML/CSS with Bootstrap) and the Twitter bot (Python). The tweets are "cleaned up" and filtered before saving: First, stop words and word endings are removed, also (except for a small selection) ASCII special characters and links are removed; this is how the training function can work best and achieve high accuracy. Then, short texts (< 5 characters) are removed, so individual emojis and words are not saved. We had many minor issues over the roughly two years of development. Bigger ones were annoying symbols in tweets (distortion of word choice), the switch to Twitter API v2 and the development of a mobile optimized frontend. Fortunately, all problems could be solved by the end.

The future for Chirpanalytica

We have future plans to internationalize for the US midterms later this year. An app for the project is already in development. However, thanks to the long development time we had many opportunities to add and test features, so we are happy with the current state.

About us

We are Torben and Paul, to young students from Hamburg, Germany. We began working on Chirpanalytica in 2017 at the age of 16. The project was in the finals for the German competition in artificial intelligence.

Share this project: