People sort their emails in folders (or with labels). Currently, they either :
move emails manually from inbox to the correct folder
or setup filters which is lightweight programming (visually or via Sieve depending on the email client/server)
Another solution could be using a classification machine learning algorithm that learns from existing folders and their content and knows where to classify incoming emails
The major benefit is that the cost of classification for human beings is close to zero. They need some initial work to classify manually initially, but then the algorithm learns from this manual classification and gets better over time.
When the human being changes the classification (like add/remove a folder), they can move around emails and the algorithm will re-learns (unlike with filters which have to be fixed manually)
When the human being notices an error, they can fix it manually and the algorithm learns from the mistake (unlike with filters which have to be fixed manually)
When the algorithm doubts, it can ask the human being to make the decision and learn from it
This is a generalization of the use of Machine Learning for spam detection (for whcih they are tons of tutorials)
With code on the client-side observing what the user does, we can learn which folders the user visits most, spend the most time searching emails in, spend the most time reading. Based on this, unsupervised learning clustering could be attempted to try to split a folder into relevant subfolders
People sort their emails in folders (or with labels). Currently, they either :
Another solution could be using a classification machine learning algorithm that learns from existing folders and their content and knows where to classify incoming emails The major benefit is that the cost of classification for human beings is close to zero. They need some initial work to classify manually initially, but then the algorithm learns from this manual classification and gets better over time. When the human being changes the classification (like add/remove a folder), they can move around emails and the algorithm will re-learns (unlike with filters which have to be fixed manually) When the human being notices an error, they can fix it manually and the algorithm learns from the mistake (unlike with filters which have to be fixed manually) When the algorithm doubts, it can ask the human being to make the decision and learn from it
This is a generalization of the use of Machine Learning for spam detection (for whcih they are tons of tutorials)
With code on the client-side observing what the user does, we can learn which folders the user visits most, spend the most time searching emails in, spend the most time reading. Based on this, unsupervised learning clustering could be attempted to try to split a folder into relevant subfolders