Several algorithms (algo) work together in the Remesh platform in order for us to understand group at conversational speed.
First, our language embedding algorithm (character level Deep Neural Network) computes a dense embedding vector for each response as it comes in. These vectors are then used to compute similarity between responses.
Next, our dynamic sampling algo chooses pairs of responses for participants to vote on which maximize information gained per vote. It adapts in real time throughout the voting process as it learns.
Then, our ranking algo estimates the mean (~popularity) & variance (~consensus^-1) of participants distribution of opinion about each response.
Lastly, our representation algo identifies a small subset of responses which are the most popular versions of the various concepts found in the full responses corpus.