This project is based on Deep Learning and NLP (Team Project)
keywords : unstructured text document, context, semantic relationship
DataSet(labeled : Text , Topic) :
-
Manually 600 unstructured Texts collected from wikipedia .
-
Topics were manually extracted .
Description :
Today, and thanks to the Web, the use of information technology has led to an explosion in the volume of data exchanged between individuals, companies, etc. and even led to a profound change in all aspects economic, political and social. This growth constant of heterogeneous data volume leads to greater complexity in information retrieval and knowledge extraction. It’s in that context that we’re talking about. We’re talking about extraction. information from textual documents. More specifically, we we are interested in extracting relationships between named entities and their context classification. Our objective is to develop an application whose realization is made in two parts: the first part is devoted to the improvement of extraction of context from an unstructured text document. This step is based on “deep learning”.
Steps:
1- Extract Keywors(important words) from unstructured text documents : used deveral keyword extraction tools :
TF-IDF/KeyBert/LDA/TextRank - chosed the best tool comparing to the giving
F1-score results for each one.
2- Training our model using :
2-1 Word2Vec to vectorize our sentence and create sentence embeddings
2-2 Training our Model using LSTM
2-3 Training a 2nd Model using GRU
2-3 chosing the best architecture according to the giving accuracy in our case ==> LSTM
3- Extracting the context using Markov chains (trained on our Dataset)/LSTM Model/Python code
4- Compare the givin context with the one figured in our dataset using ROUGE matric.
Read More about ROUGE metric : https://medium.com/nlplanet/two-minutes-nlp-learn-the-rouge-metric-by-examples-f179cc285499