A bigger heart represents more people talking about the party.
Welcome! Discover how a next-gen AI computes this information using ULMFit and NLP!
AI election pulse was born as an internal project at Basecodeit to prove the capacity of the latest findings in artificial intelligence and machine learning. In order to do this, we had to implement the most recent papers and implement never-before-used methods in our country (Argentina) for stance analysis on a topic, such as a political party.
Everything starts with a language corpus, which is a word compendium in a language (English), that is used to “train” a computer so it “learns” that language. Then the resulting model is fed with example data, tweets in this particular case, and it’s told in which cases the tweets are in favor or against the subject. The sum of different rounds of training makes the computer be able to adjust it’s base knowledge (the language) to a new model of knowledge (politics) and label each tweet’s stance regarding the political topics. These techniques are named ULMFit and Natural Language Processing or NLP for short.
The biggest difference of this new generation of artificial intelligence with the previous, is that sentiment analysis was made using keywords associated with feelings, such as “happy” or “sad” or “mad”, and they were weighted to determine if the sentiment was more or less intense. During the last decade, this way of quantifying people’s feelings, with the addition of social networks penetration, helped analysts obtain information in real time about a particular topic or brand.
If you’re interested in the hard tech information, we recommend you to read our blog post by Sebastián Balazote.