These are many chatbot platforms available on the web which are programmed for various purposes. There is a chatbot to help dementia patients, there’s one for the insomniacs and there are chatbots for numerous websites which handle the initial conversation with a visitor of the website.
Google’s team used 40 billion words which include 341 GB of text data including social media chats to train Meena using seq2seq model. Now seq2seq is a modified model of Google’s Transformer. It is a neural network that compares words in a paragraph with each other to understand the relationship between them.
Google claims that Meena has 1.7 times more model capacity. The team trained it on 8.5 times more data than one of the best chatbot which is available – OpenAI’s GTP-2. The researchers also developed a new metric system which they are calling Sensibleness and Specificity Average (SSA) to measure how sensible or specific a conversation response is. Image: Google
