Academic | Shikhar Agrawal

Internship Management Portal

Sun, 01 Nov 2020 10:39:09 +0000

Human Detection

Sun, 01 Nov 2020 10:36:43 +0000

Improved accuracy in detecting humans in low-resolution thermal image using combinations of Histogram of Oriented Gradients.

Vectorization Diagnostics

Sun, 01 Nov 2020 10:08:41 +0000

Analyzed the feedback mechanism of LLVM optimizers and proposed ideas to make LLVM on par with ICC.

Load Balancer

Sun, 01 Nov 2020 10:06:38 +0000

Load balancing is defined as the methodical and efficient distribution of network or application traffic across multiple servers in a server farm. Each load balancer sits between client devices and backend servers, receiving and then distributing incoming requests to any available server capable of fulfilling them.

This is a simple load balancer implemented in Python.

Fake News Detection

Sat, 31 Oct 2020 13:29:10 +0000

The spread of rumour related to an event on a social media platform affects the dissemination of true information. Identifying rumour from multiple post becomes a challenging task as the rumour is similar to actual information. Rumour detection can be seen as a text classification problem.

PHEME dataset for Rumour Detection and Veracity Classification -
Kochkina, Elena; Liakata, Maria; Zubiaga, Arkaitz (2018): PHEME dataset for Rumour Detection and Veracity Classification. figshare. Dataset. https://doi.org/10.6084/m9.figshare.6392078.v1

Further based on the 9 events in the dataset more data on rumours was collected from Twitter. Classified the tweets following a rumour by its stance into one of supporting, denying, questioning or commenting. A tweet agreeing with the source tweet is labeled as supporting, a tweet disagreeing with the source tweet is labeled as denying and so on. Also, a tweet disagreeing to a denying tweet is labeled as suporting. A probablistic framework such as Hawkes Processes (HP) has been used in the modelling.

Finding Active Expert Users in CQA System

Sat, 31 Oct 2020 13:27:05 +0000

Community Question Answering (CQA) system forms crucial part of the internet. Websites such as stack exchange, Quora, Reddit, etc helps lot of people. Experts in numerous fields are available to answer these questions. I still remember back in the early days of Quora, answers related to wikipedia were answered by Jimmy Wales (Founder of Wikipedia) himself. Given the large number of questions and users, it is important that each question is routed to the most suitable person. This ensures that the question is answered appropriately and within time.

The first challenge is to find expert users in the system. This is achived using topic-modelling techniques such as Latent Dirichlet Allocation (LDA). Just finding the expert users won’t work if the users are not active. The second part deals with finding the active users using time-series analysis. This can be extended to suggest the most appropriate time to route questions to a specific user.

Share IIT

Sat, 31 Oct 2020 13:19:43 +0000

Web Crawler

Wed, 28 Oct 2020 07:53:04 +0000

A simple application in Haskell that lists all the web-links from an input domain.