News text generation with adversarial deep learning

Sammanfattning: In this work we carry out a thorough analysis of applying a specific field within machine learning called generative adversarial networks, to the art of natural language generation; more specifically we generate news text articles in an automated fashion. To do this, we experimented with a few different architectures and representations of text, evaluated the results and used the information retrieved from the results, to create a model that should give the best result. For evaluation, we used perplexity and human evaluation. We also looked at the token distribution to see which model captures the texts most successfully. We show that it is possible to use generative adversarial networks to generate sequences of tokens that resemble natural language, but this does not yet reach the quality of human-written text. Further hyperparameter tuning and using a narrower-subjected corpus could improve the output.