We use Deep learning to understand purchase impulse on a web session to trigger responses to maximize conversion. DEEPLEARING MODEL FOR PREDICTION OF PURCHASE INTENTION The "predictions" folder contains a CNN (which uses ReLU for activation and Softmax for normalization) which takes the parameters of the session and predicts a probability of purchase (fixing that there are two sets of data, in the data folder - You can run it from your console). CHATBOT (IN PROGRESS) In "chatbot" is the chatbot tool that will trigger the loss of momentum. I took the codes of Github - because - not important is not the code but, the decision to use MN in the place of LSTM (as recommended in this article: https://arxiv.org/abs/1410.3916). DATA COLLECTION IN REAL TIME The data collection of the session, running fashion continuously with JS. If you enter in Boomfix.es you can see the process that runs continuously (records session duration and scroll interactions). WEB APP TO DISPLAY PRODUCT In the APP folder I am putting together the wrapper of the project that I will show you in my local (I use FLASK, for the speed that gives me for prototyping).