Are there OpenSource neural networks for working with text and communication?

Question

Bots

I want to find an opportunity to use a self-learning neural network that I could adapt for chatting.
Are there such projects?
I saw something from Microsoft, but there was photo recognition.

Answer 1

The neural network does not work with text. It works with data, and what kind of data is how you teach it.
You need an OpenSource framework to build neural networks. Next, you build a neural network, and train.
Neural networks are not self-learning and are very limited. They need to be trained from outside.
They need to be trained under controlled conditions, and further fixed, to prohibit development. That's what they usually do.
It is such trained and fixed networks that are of commercial value, being a finished product. Unfixed networks are feared like fire - they tend to throw out surprises.
The fact is that the development of neural networks spontaneously, and without control, leads to chaos.
Under controlled conditions, development is tightly controlled, which is why the network is trained in the right direction.
In combat conditions, it is difficult or impossible to create an adequate control system, so surprises are not uncommon.
Or, if a developing neural network is so needed, you need to build a dynamic control system over it, which is more difficult, which will decide what is good and what is bad.
The difficulty lies precisely in determining what is good and what is bad, having a random data set - after all, the network must be trained in the process of work.
There are tasks in which it is easy to determine the result of the work of a neural network - whether it is good or bad. There are also those where it is difficult, if not impossible, to do this - it requires a whole trained neural network just to control another neural network.
There is a third option, a compromise - data processing by a fixed combat network, and in the background data analysis, further training of a combat network clone on them, clarification of success criteria (including with human participation - there is nowhere to rush), testing, comparing results with combat network, and, if everything is cool - in the next service period, the replacement of the combat network with a more developed clone. And so in the cycle.
It's safer this way - a growing network will never go into combat mode until it meets the reliability requirements, and a reliable network that is fixed from changes is doing the work, giving a stable result without surprises.
In general, the technology is primitive: evolution is evolution, you simply determine the criteria for success, by which you separate successful variations from unsuitable ones, and then fix positive changes, and discard negative ones.
You can google teaching methods - there are several of them, and they are all old.

Answer 2

There are chats with bots like www.siteheart.com/ru/features - but I'm not sure if they use neural networks. But perhaps this option is right for you?
Neural networks, in my opinion, are an interesting technology, but still too far from building bots for communication. Yes, and the resource intensity is probably prohibitive. What makes their use impossible in solving everyday business tasks related to chatting.