├── README.md
├── Topic_Modeling_BERT+LDA.ipynb
├── dev-deployment.png
└── model-explanation.png


/README.md:
--------------------------------------------------------------------------------
 1 | # Topic-Modeling-BERT-LDA
 2 | Topic modeling with BERT, LDA and Clustering. Latent Dirichlet Allocation(LDA) probabilistic topic assignment and pre-trained sentence embeddings from BERT/RoBERTa.
 3 | 
 4 | 
 5 | Model explanation
 6 | LDA for probabilistic topic assignment vector.
 7 | BERT for sentence embedding vector.
 8 | Concatenated both LDA and BERT vector with a weight hyperparameter to balance the relative importance of information from each source.
 9 | Use autoencoders to learn a lower dimensional latent space representation of the concatenated vector.(Why auto encoders why not pca. PCA is linear transformation of vectors and AE are non linear.
10 | The assumption is that the concatenated vector should have manifold shape in higher dimensional space.
11 | Used clustering on the latent space representations to get topics.
12 | ![](model-explanation.png)
13 | 
14 | Data pipeline(From development to deployment)
15 | ![](dev-deployment.png)
16 | 


--------------------------------------------------------------------------------
/dev-deployment.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/AravindR7/Topic-Modeling-BERT-LDA/cbb7e4777d677330c64a84e4ab9be547424c5a0b/dev-deployment.png


--------------------------------------------------------------------------------
/model-explanation.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/AravindR7/Topic-Modeling-BERT-LDA/cbb7e4777d677330c64a84e4ab9be547424c5a0b/model-explanation.png


--------------------------------------------------------------------------------