We Are A Sentiment Thermometer

From Algolit

Type: Algoliterary exploration
Dataset(s): Glove, 1984 by George Orwell, Frankenstein by Mary Shelly
Technique: word embeddings, Scikit Learn supervised machine learning
Developed by: Common Crawl/GloVe, Rob Speer/ConceptNet, Algolit

A language model recounts its story in a metaphorical way. You are guided through a multidimensional world in which artificial intelligences lead explorations, explore landscapes and create maps that allow them to go along paths of predictions.

We are a Sentiment Thermometer is a collective being based on classic supervised machine learning and unsupervised Neural Networks pretrained GloVe word embeddings. They can either judge a sentence on its positive or negative sentiment, or it can guide you through its components and show how they are made, which choices led to their functioning, who developed each of the elements, how each part can be replaced.

Using the collective intelligence of the internet as training data, they show how their scores and judgements are influenced by the data they have been trained with. Our human prejudices and cliches are passed on to machines and induce them with racist and other biases.

Based on a script by Rob Speer: