directorman9 Goto Github PK

cross_collection_opinion_mining

This repository contains code and data sets I used in my Master's thesis "Cross-collection aspect based opinion mining using topic models". The document can be found at http://research.sabanciuniv.edu/36615/1/10209464_HemedHamisiKaporo.pdf.

directorman9.github.io

A personal website

dynamic-topic-modeling

Models the hiring trend of world organisations. We use the keywords searched and topics discussed over the years to determine the In demand jobs and skills as a function of time and location.

google-maps-scraping

This repository shares results of a google-map scraping task. The goal was to scrap information about schools in the carribean. The information included categorty of the school i.e. daycare, university, sports school etc; name, physical address, web adsress, and contact information of the school. The result is a list of 2000 schools and their required information

gym-minigrid-games

This notebook trains an agent to navigate a maze and reach a desired destination. It uses the Gym-MiniGrid's fourRoom-v0 environment as the maze. The agent is trained by using reiforcement learning's vanilla policy gradient (REINFORCE) algorithm.

labeling

An application to help students label my data.

lawsuits-summarizer

# lawsuits-summarizer A lawsuit is long (5-100 pages) document detailing a legal case. The goal of this project is to train a summarization model and expose an API endpoint such that given a lawsuit document returns its short (less than 500 words) summary.

optical-character-recognition

The notebook in this repository uses pytesseract to extract text from a pdf document. The script can be used to automate text acquisition from a large body of printed resources such as books. The acquired text can then be used for dowstream tasks, such as training language models, topic models, document summarization etc

sequence-to-sequence

This project trains a sequence to sequence model. The model takes in a paragraph of product description and retuns a suitable product name. The approach can be used to match similar products in a store.

timeseries-forecasting

An LSTM for weather forecasting. Predicts the condition of 14 weather indicators (temperature, wind, humidity, pressure etc) for the next 2 weeks with high accuracy

transformer-for-text-classification

A custom state of the art transformer for text classification. It classifies long text (500-1000 words) hotel reviews) into binary sentiment classes i.e. positive or negative.

vimeo-index