Using a Markov Model and LSTM Neural Network to generate text
Chose an NLP problem, not restricted to the following:
- Reconstruction
- Document Classification
- Token Classification
- Language Modeling
- Machine Translation
Identify or construct a solution based on a generative probabilistic language model. Describe the model in detail and develop a solution using parameter inference (and/or decoding).
Identify or construct a solution based on a discriminative neural network. Describe the network structure in detail and develop a solution using parameter inference (and/or decoding).
Train and apply both approaches to real data acquired legally. Evaluate the results qualitatively and quantitatively. Highlight situations where each approach performs well or poorly. Any unusual/unexpected results require explanation.
Train and apply both approaches to synthetic data acquired legally. Evaluate the results qualitatively and quantitatively. Highlight situations where each approach performs well or poorly. Any unusual/unexpected results require explanation.
Discuss pros and cons of the two approaches. Consider:
- quality/correctness
- data, time, and computational requirements
- interpretability