As social media has become widespread, it is increasingly used as a way of gathering and conveying information about news and controversial issues. To prevent false information/rumours from spreading or verify them immediately, there is a need for rumour classification system. This work focuses on stance classification towards rumours and rumour verification on Twitter. It utilizes a deep learning model on stance classification and supervised learning models on rumour verification.
Information-Centric Networking is still an incomplete paradigm which introduces a large variety of new topics and approaches over the traditional networking. Since it is a relatively new concept which promises for easier and faster data access many aspects of it have been studied intensely in recent years. Nonetheless, load balancing is one of the less focused but high potency, open-ended areas. The aim of this project is to present an alternative technique to load balancing in the domain of pre-suggested routing method LSCR (Link State Content Routing).
In this project, we provide efficient, parallelized inference algorithms for Allocation Model which has a close relationship with Nonnegative Matrix Factorization. NMF is the problem of writing a nonnegative matrix, X, as the multiplication two nonnegative factor matrices, W and H. Bayesian analysis of NMF models show us that (M x N) X matrix is implicitly decomposed into a hidden (M x N x K) tensor S. Allocation Model starts from this idea, and analyzes hidden tensor S more explicitly.
The aim of our project is to predict home country of social media users who do not specify it explicitly. To achieve the goal, we used location information of users who already specified it and approach the problem in a probabilistic manner. In this project, we focused on Twitter users, however, the scope may be widened to any other social media platform. The problem can be easily extended to other prediction problems with similar missing information structure.
As the number of computer programs that need the gradient of functions increase, (such as machine learning programs that use gradient descent algorithms) efficient derivation of gradients problem has risen. Many machine learning / deep learning algorithms use derivations at each step. So the efficiency of the derivation matters a lot for those programs. There are three main ways to differentiate a function. On this project, we implemented automatic differentiation which is the most suitable one of these methods for our purpose of use.
The use of multiple controllers in SDN can be due to performance improvements, scalability, and reliability. That is, traffic can be processed much faster and a fault in one controller can be hidden by another one with multiple controller usage .
To take advantage of multiple controllers, one can distribute the load on the controllers in such a smart way that SDN does not suffer from high traffic and consequently latency, low bandwidth, and even worse, crashes.
Grocery shopping is a daily activity for most people. Not only do people buy what they need but probably also what is promoted at the time. Super markets promote some products daily, or weekly. The problem is that they promote the same products to all people. Some people might like the products whereas some might not. What is lacking is a personalized promotions so that super markets will promote products based on user’s shopping history and history of people with similar history.
Understanding the need of the infant from his/her cry is a major topic which is studied by psychologists and medical doctors at the beginning. After the technological improvement, with the help of data analysis and data processing, this topic has got a big interest and be considered in many researches. In this project, we developed an approach to find a good accuracy of recognizing and classifying a newborn's cry into 8 categories which are 'Hunger, Need of Burping, Belly Pain, Discomfort, Tiredness, Loneliness, Cold/Hot and Scared' .
