Author profiling algorithms are indeed now-days efficient, but freqeuntly rely on (very) large sets of features. with this task I tried to investigate (as work for my intro to programming class) if a similar classification problem, that identifying the author of a speech, using (word2vec) embedding. To do so I implemented a Bilateral LSMTM. The classification is on 43 united stated elected presidents and is done based on their discourses.
- nltk
- random
- logging
- numpy
- gensim
- keras
- sklearn
- pandas_ml
- matplotlib
The contained files are: the network script, the zip of the U.S. presidential speeches as well as the already prepared President - Year - sentence corpus. This last two were scripted by Patrizio Bellan (patriziobellan86) for a previous project.