About Me

I’m a Data Scientist at Fidelity Investments, working in personalization and recommender systems. My passion lies with making sense of people through data. It generally manifests itself through my interest in Recommender Systems, but I welcome all opportunities for analyzing the data that people produce.

Before that, I was a Computer Science Masters student at NYU Courant Institute of Mathematical Sciences (class of 2019), with a passion for Machine Learning and Data Science. Before that, I was studying BS. in Computer Engineering and doing an Economics minor at Bogazici University in Turkey, with one semester spent at the University of Queensland in Australia as an exchange student.

Before starting at Fidelity, I interned at Hifi building Recommender Systems in order to personalize your music needs. Before that, I was working part time at NYU IT as a Software Developer. In my free time, I like listening to different kinds of music, reading fantasy novels (not counting my obsession with Haruki Murakami), and playing games.

Projects

Listed below are my projects revolving around Machine Learning/Data Science.

cu2rec: GPU Accelerated Matrix Factorization for Recommender Systems

Recommender Systems · Machine Learning · CUDA · C++


· Built (in CUDA) a Matrix Factorization library optimized for Recommender Systems
· Implemented parallel Stochastic Gradient Descent optimized for GPUs
· Reached error metrics similar to the best sequential versions while being 10x faster

Books2Rec: Machine Learning meets Reading

Recommender Systems · Machine Learning · Latent Factor Models · Python


· Built a hybrid Recommender System, using Goodreads book ratings and book features
· Used SVD and Autoencoders to achieve a RMSE (Root Mean Squared Error) of 0.843
· Available live at books2rec.me

Visualizing the Rental Housing Crisis in US

Data Science · Data Visualization · RShiny


· Visualized the increase in rents with respect to income levels in major metropolitan areas of the US
· Produced an RShiny app that showcases the differences between the metropolitan areas
· Available live at dorukkilitcioglu.shinyapps.io/RentBurden/

Relation Extraction using Deep Learning

Natural Language Processing · Recurrent Neural Networks · Machine Learning · Tensorflow · Python


· Conducted a research on extracting 19 different kinds of relations between entities
· Read & implemented ideas from multiple different research papers
· Achieved 49% F1-score using CNNs and 51% F1-score using Bi-LSTMs on ACE 2004 dataset

Financial Analysis using Machine Learning Methods

Machine Learning · Hidden Markov Model · Finance · Natural Language Processing · Python


· Conducted machine learning based analysis on various stock prices & estimated future prices
· Collected and annotated relevant articles on stocks
· Obtained 54% accuracy (over baseline 50%) with a Hidden Markov Model variant with sentiment analysis

Deep Recurrent Neural Networks for microRNA Target Prediction

Bioinformatics · Deep Learning · Machine Learning · Keras · Python


· Implemented a research paper on Deep RNN model for miRNA target prediction
· Simplified the model while maintaining 94% accuracy and 96% F1-score
· Showed the importance of gathering negative samples

A Monte Carlo Algorithm for Cold Start Recommendation

Recommender Systems · Monte Carlo Methods · Machine Learning · Python


· Implemented a research paper on collaborative filtering based Monte Carlo Algorithm for cold-start recommendation
· Decreased MAE (Mean Absolute Error) by 1.8% by using better transition priors and verified results using MovieLens database

Content-based similar movie recommendation engine

Recommender Systems · Natural Language Processing · Machine Learning · NLTK · Python

· Implemented a content-based filtering approach to movie recommendation, using movie plots, paper by Fleischman et al.
· Improved the user-rated similarities by 3.5% (verified using A/B testing) using Named Entity Recognition to simplify names and converting plots into topic vectors using Latent Dirichlet Allocation

doc-annotate: A minimalistic tool for annotating documents

Data Collection · Data Annotation · Flask · Python

· Basic web tool for annotating the sentiment of articles
· Allows for choosing between 3 levels of sentiments (positive, negative, neutral) and 3 levels of relevance (relevant, irrelevant, neutral)
· Built using Flask & MongoDB

Contact Me

Any questions or comments? Email me below.