A project for generating fake omegle conversations, This is a LSTM neural network that is fed with a .txt containing multible Omegle conversations with proggramers, The dataset is hand collected, small and biased, A new automatic data collection system is being developed to sample a large amount of conversations under a wide range of themes, The LSTM is a modifed example from the keras blog and is only a proof of concept, A more complex and original network is also being developed.
To use the pretrained model (updated and improved regularly) just simply run the "Conversation_Gen.py" file, this will output a text file named "generated.txt" in the same directory, to change the length of the generated text (chars), open up the "Conversation_Gen.py" file and change the 5000 in the last line to whatever you desire. You may also delete the premade ".h5" file, after doing this running the "Conversation_Gen.py" file will result in the model being retrained, every epoch 6 txt files are created in organized folders to show the proggress of the training, the txt files names are the diversity the model used, also known as how exploritive the model is when writing text, Amount of training (epochs) can also be altered
I am aware the output is not very good. This is due to the reasons I mentioned above (poor data and model structure). This project is a WIP and just a side project, I hope as time progresses I will be able to remove this disclaimer as it has reached impressive results, Thank you for your time, Gaba