Using AI for Categorisation of TV Shows on Embedded Platforms


In the course of a master thesis, we investigated, how well the genre of TV shows can be automatically determined with the help of AI, based on their description texts. We used Keras as an abstraction layer for AI frameworks and trained a variety of neural networks with Tensorflow. After comparing different approaches and selecting the most suitable neural network, it was fine-tuned with hyper-parameter tuning.

The resulting neural network can be used to automatically categorise a large part of the TV shows with a very high degree of accuracy. The achieved performance is comparable to that of humans or even exceeds it.

Through further optimisations, the performance was increased to such a level, that the AI-based categorisation could be used even on an embedded platform. To demonstrate this the neural network was ported to the BCM97271T reference platform using TensorFlow Lite.