I'd like to use Google's TensorFlow (deep learning neural network) to generate new content for me. I have about a 150 emails from a 10 year span. I'd like to use Google's Tensor Flow to generate/predict what these emails may sound like for the next 90 years.
This is an experiment into utilizing neural networks. I understand that the information may be garbled. I do not expect a perfect response. I want something that is developed and coded well.
I want to use the following batches of information:
--> Existing batch of emails which have been condensed into a spreadsheet with certain topics and themes. Freelancer will have access to all emails and their content.
--> RSS feeds from a single newspaper (with a filter for a particular topic/s). Freelancer will help identify good keywords for the particular project and use them to 'inform' the neural network. This will help it generate more accurate content.
--> One wikipedia page listing content related to emails
--> One multi sheet Spreadsheet showing incidents related to emails (not very extensive)
What I need is a piece of software to connect to Google's TensorFlow to generate this content based upon all of this information. As I continue to get more emails (around 1 per month) and the wiki pages and other content sources update I'd like to be able to 'feed' it into Tensor flow to increase its accuracy.
Somebody recently did this with episodes of Friends (http://www.thedailybeast.com/articles/2016/01/20/meet-the-robot-writing-friends-sequels.html)
A good quote from the article was:
"It's quite simple - It works by predicting the next letter to follow a given sequence of letters, and the predictions are determined by what it learned about language."
Feel free to message me, we can discuss this in length.
The freelancer is expected to put together a sensible structure for Tensorflow. His/her work will be reviewed by another coder who is familiar with neural networks/tensor flow.