Notes

begin{itemize}
item 3 methods: LDA, LDA SVM and LDA W2V
item textbf{ LDA :} Pair topics with families in one-to-one manner
begin{itemize}
item Create lda model with train data,
item Assign topic that has the most probability to each sequence,
item Investigate all sequences in a family,
item Find the most frequent topic in a family,
item Assign such topics to each family
item Use this pairs to assign families to test data
end{itemize}
item textbf{ LDA SVM:} Classify with feature vectors
begin{itemize}
item Create lda model with train data,
item Create feature vectors with topic probability distribution,
item Feature vector size is the topic size (e.g. 20, 100, 200)
item Train SVM model with training sequences
item Create feature vectors of test data according to lda model.
end{itemize}
item textbf{ LDA W2V:} Classify with feature vectors that uses Word2Vec
begin{itemize}
item Create lda model with train data,
item Create feature vectors with topic probability distribution and Asgari W2V data,
item Feature vector size is the topic size (e.g. 20, 100, 200) + 100 (W2V word embeddings)
item Train SVM model with training sequences
item Create feature vectors of test data according to lda model and Asgari word embedding.
end{itemize}
end{itemize}