How Machine Learning has improved Email experience?

Email is a method to exchange messages between people and communicate. It first started being used in the 1960s in its very

trivial form and has now expanded to be used by billions of people across the globe for a wide number of purposes ranging from

business, academic, marketing or personal use.

Two recent trends that have changed the email experience or due to which the email services have become more personalized

are:

• Increasing use of smartphones
• Growth and use of machine learning techniques

Smartphones provide users with portability, instant file sharing etc which makes it their primary device of use, on the other hand, it gives email providers access to a lot of personal data like location, contacts, photo/videos which can be leveraged to enhance the user experience. However, smartphones come with their set of shortcomings like limited screen size, difficult to use onscreen keyboard, but the recent advancements in machine learning have enabled us to overcome these limitations and provide users with more unique and individual experiences. Few examples of how machine learning has improved the email usage can be:

1. Email Search- Due to smartphone's limited size, as a user what we want when we search for a query is that we get the most relevant search results. To do this we need modeled rankers which can provide results according to different users but to train such rankers we need relevance-labeled training set. However, due to user privacy, email providers cannot use relevance raters to produce labels. So they use the most commonly available and easy to leverage feedback system provided by the users themselves, namely 'Clicks'. Using click logs to get our training data is both interesting and challenging at the same time as there is always a continuous and fresh supply of training data from the user. Also, there is a problem of bias towards those queries which receive more clicks, for example due to already being the top returned results. Different email providers come up with their own solutions to neutralize this bias. The solution that Google came up with for their Gmail application can be read here.
2. Extracting information from appointment, reservation, booking mails- The approach is to learn the templates or format that were used to generate such mails, then extract key information like date, timings, place from the mails and store them and then probably display them as reminders on user's smartphone taking contextual information into account like transit time, example for flight, movie tickets. For example, in Gmail as a user receives his/her flight confirmation mail, Google goes through our mail and extract key points from it and automatically sets a reminder for the appropriate time. You may have noticed something like the following image being self-generated as an Android-Gmail user for a flight reservation that you may have made.
   
3. Smart Reply- This provides a set of short responses to the received emails allowing us to reply to an email in only a few clicks without much typing, thereby giving us a hassle-free and convenient experience. This is done by learning a model of short replies to original emails from the email corpus and then use this trained model to generate automatic replies for new incoming messages. This model gets continually more refined and trained as the user increases the application's usage. The confidence, accuracy and quality of these automated replies also gets significantly better as the model learns more diverse replies and mails. This feature is particularly notable in the Gmail smartphone application.
   

These are a handful of examples of how the growth of machine learning based Information Retrieval and smartphones have enhanced our user experience.

Source:

https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45673.pdf