The project was to do Extravtive and Abstrative Text Summarization for articles. The Abstractive Text Summarization was based on Page Rank algorithm and Eucledian Distance/Cosine Similarity

The Abstractive Text Summarization was bsed on Encoder Decoder Architecture and was developed on Tensorflow. The project required NLP Pre Processing tasks(Noise Removal, Text Cleaning, Contraction Removal, HTML Tags removal etc). The tex was tokenised into sentences and then into words using Keras and NLTK. The application was built with word embeddings and attention layers

Below was a sample output from the application:

Review: influenster quaker sent cookie try looked delicious healthy never know delighted good great tasting love Original

Ground Truth summary: love this cookie

Predicted summary: quaker soft baked oatmeal cookie

The project was to build a Knowledge Graph from a set of PDF documents and extraxt information from the Knowledge Graph. The project was based on Natural Language Processing concepts and required to process the documents, extract the entity pairs(NER) and their relationships and build the knowledge graph.

Further, exract the graph for an entity, a relationship and build a question answer application.

Example:

Q.How many tests has Marshall played?

A: Reply from graph: 81 Tests

The project was to build a Neural Style Transfer application. The source was two separate images and the application extracts feature vectore from the first image and style vector from second image and builds and image merging the two.

The project was based on CNN architecture

The Machine Learning application was built for a Food Delivery customer to predict demand for orders for next 3 months on a weekly basis. Historical data for previous 145 weeks was provided.

The project POC was developed with following activities:

1. Data Cleansing

2. Descripive & Inferential Statistics

3. Null Value Treatment

4. Data Visualization

5. Feature Engineering

6. Test of Seasonality and Stationarity

7. Data Normalization

8. Dimentionality Reduction

The model was build on multiple algorithms: Light GBM, Catboost, XG Boost and ARIMA. Finally, the predicted values from the 4 algorithms were stacked to get the final output