Kieran Fraser
PhD Candidate @ ADAPT Centre
Trinity College Dublin

Stemming from my research on push-notifications and the potential adverse effects they can have when abused, I am exploring how and if my research can be of use, from a commercial perspective, to enable those who push content to do so by empathetic means. EmPushy hopes to be an empathetic intelligence which can mediate the stream of information pushed at people while also maximising the impact it has for content creators. The idea is that the product will contain 3 commercial services offered to 3 distinct audiences.

1. Websites & Apps - These are the business that are pushing information to their users. Currently, this is done in a very unintelligent or manipulative manner, whereby the information *pings* the user at the discretion of the pusher, usually resulting in instant dismissal of the information by the user. The solution EmPushy offers for this audience, is intelligent delivery, at a time that maximises Click Through Rate but at times that are empathetic to the end-user.
2. Content Creators - These are the journalists and social influencers creating content that is eventually pushed via notification and consumed on various platforms. EmPushy offers this audience a real-time dashboard where they can see what stories are trending with regard push-notifications i.e. which notifications (stories) are being opened and engaged with and which are not. Through Natural Language Processing and Generative Modeling techniques, this service recommends personalised trending content to push to users.
3. People who use technology! - These are everyday users of technologies such as internet browsers and smartphones. Anyone who gets pushed information through technology can benefit from EmPushy as it enables a central point of control to mediate (or stop completely) push-notifications from reaching and distracting you. Push-notifications are delivered, instead, intelligently and with empathy. It also allows people to discover content which they would like to be subscribed to. Previously unknown streams of content, available on your own terms.

• Learn More

Gym-Push is a custom OpenAI Gym environment that simulates real-world push-notification engagements, enabling reinforcement learning (RL) agents to learn how to effectively manage push-notifications on behalf of a user. The idea behind Gym-Push is to benchmark state-of-the-art RL algorithms on the problem of push-notification management so that standard performance metrics can be tracked in the research community. Powering Gym-Push, is a generative model which produces realistic synthetic notification-engagement data derived from a real-world data set. This data is transformed to make up the state, action and rewards of the environment to be used by agents during learning/evaluation.

• Learn More