Webselect

Complex Networks, Web Crawling, Advertisements, Genetic Algorithms, Python

Supervisors

Prof. Uttam Sarkar, MIS, IIM Calcutta and Prof. Agam Gupta, IIM Rohtak

Duration

Mar – May 2016

Demo Link

webselect.herokuapp.com

Designed a tool to select the best subset of websites to maximise the reach of advertisements, within budget and demographic limits, from a graph of websites by scraping traffic information from Alexa. Genetic Algorithm used to optimize the selection problem as the original problem is NP-Hard.

Image source: The Internet Map

LTTKGP

Complex Network, Facebook, Social recommender, Music

Duration

Aug – Nov 2018

Demo Link

lttkgpreco.herokuapp.com

From the music sharing activities in our school facebook group, called Listen to This KGP, we built a bipartite network of songs and users, and using Glove vectors to encode the song tags, we built this recommender system to suggest a personalized playlist to every person in the group.

Tweet Situationality

NLP, Twitter, Classification, Feature Engineering, Disaster

Supervisor

Prof. Niloy Ganguly, IIT Kharagpur

Duration

Nov 2015 – Jan 2016

Paper

ACM TWEB

A classifier to separate disaster related tweets into Situational and Non Situational Classes, using sentiment detection, dependency graphs and linear patterns for feature engineering. Work showcased at IBM Demo Day 2016. Also part of a situational summarizer project published in ACM Transactions on the Web.

Savitr

Twitter, Geolocation, Dash, Python, MongoDB, NLP, ML, Feature extraction

Supervisor

Prof. Saptarshi Ghosh, IIT Kharagpur

Duration

Jul 2017 – Jan 2018

Web Demo

http://savitr.herokuapp.com

Paper

WWW-SMERP 2018

Aa system that leverages the information posted on Twitter to monitor and analyse emergency situations. We infer the locations mentioned in the microblog text, in an unsupervised fashion and display it on a map-based interface. The system is designed for efficient performance, achieving an F-score of 0.79, and is approximately two orders of magnitude faster than other available tools for location extraction.

SMA

Finance, Market prediction, ML

Supervisor

Prof. Abhijeet Chandra, IIT Kharagpur

Duration

Jan 2017

Project to crawl social media (Facebook/Twitter) and top financial news websites to find the top posts and news articles that can affect share prices the most. This was used in the demo built for Inter IIT Tech Meet 2017 where Won the Silver Medal at the Stock Market Analysis Event.

Facebook Ad Transparency

Facebook, Propublica, Advertisements, Behavior Patterns, Transparency, Controlled experiments

Supervisor

Prof. Alan Mislove, Northeastern University

Duration

May - Jul 2018

Study of advertiser behavior and targeting patterns on Facebook by using the ad reach information obtained from Facebook’s ad transparency feature and the personal targeting dataset from Propublica’s Facebook ad dataset, aided with controlled ad placement experiments.

Targeted Commenting

NLP, Deep Learning, Web Design

Duration

Jun – Oct 2018

Paper

Paper

Most of the user comments on online news articles are relevant to only particular sections of an article. In this work, we build a system which can automatically classify and display comments against relevant sections or paragraphs of an article. To implement that, we develop a deep neural network based mechanism to find comments relevant to any section and a paragraph wise commenting interface to showcase them. Such a data driven commenting system can help news websites to further increase reader engagement.

Madad

Android, Geolocation, SOS

Duration

Feb 2016

Link

Github

Security is increasingly important in the modern world today, and thanks to the penetration of smartphones and technology we can use them to solve a large number of problems. We developed an android application to allow people facing any kind of peril to signal for help to their preferred contacts and also to people nearby. This app won the Gold Medal in the Inter IIT Tech Meet, Software Development Event. I was responsible for the App frontend and rest api integration.

Molecule2Vec

Doc2Vec, Deep Learning, Regression, Chemistry

Supervisor

Prof. Debasis Sarkar, IIT Kharagpur

Duration

2017 - 2018

Converted 3D Molecules to a Vector Space Model using Doc2Vec. Using the trained molecule vectors and the solubility data, I trained ML and Deep Learning Algorithms to predict solubility within an error of 0.3 g/litre. Almost 65% of the solubility variance was explained by the structure vectors exclusively. This work was accempted and presented at the EUCHEMS 2018 conference by the Royal Society of Chemistry.