Recurrent Neural Networks for Text Generation

Among the leading practical developments in the use of computer neural networks is the improvement of language generative text models by training on materials from classical literary texts in natural language. A promising direction is recurrent neural networks (RNNs), in which the connections of elements are specified by directed sequences and preliminary automated configuration (training). Improving language generative text models for automated determination of sentiment (emotional semantics) of source texts can be achieved through advanced methods of training a recurrent neural network: the use of internal memory allows processing changing sequences of relationships of arbitrary length. The work explores methods for using RNN and LSTM neural networks to solve one of the most interesting and complex problems of text generation – improving the contextual adequacy, semantic richness, and emotional coloring of search results based on user requests. Examples of creating a neuron and building a neural network with hidden layers are given. Recurrent neural networks, their structure, and areas of application are considered in detail. An example of creating a classic RNN for analyzing the sentiment (emotional coloring) of a text is presented. Particular attention is paid to one of the modifications of RNN, the LSTM (long-short-term memory) model, which is best suited for solving the problem of generating text with increased contextual adequacy, semantic richness, and emotional coloring based on user requests.