Naassih Gopee

Naassih Gopee

Technical Co-founder at Inpleo

Contact Me

About Me

I studied computer science at Carnegie Mellon University. I minored in machine learning and completed a college honors thesis titled: Applying Recurrent Neural Network to Arabic Named Entity Recognition.

While in CMU, I worked as a research assistant for the robotics lab under Prof. Majd Sakr. The research project involved developing a bilingual robot receptionist capable of interacting with students in both English and Arabic. I also spent my senior year at CMU working on applying Recurrent Neural Network to NLP tasks under Prof. Kemal Oflazer and co-advised by Prof. Bhiksha Raj, Prof. William Cohen & Prof. Houda Bouamor . The project tackles the problem of classifying Named Entity Recognition using Long Short Term Memory neural network.

More recently, I co-founded Inpleo – a B2B reverse auction SaaS platform. As the technical lead, my role at Inpleo is to develop the backend. The developments include designing the system infrastructure, developing the machine learning engine and the data analytics dashboard.

Latest Projects/ Publications

Applying Recurrent Neural Network for Arabic Named Entity Recognition

CMU SCS Honors Thesis

project name
New

Named Entity Recognition (NER) (also known as entity identification) is a subtask of information extraction that seeks to locate and classify elements in text into predefined categories such as the names of persons, organizations, locations, etc. NER plays an important role in many Natural Language Processing (NLP) problems, such as Machine Translation as it helps improve the performance of algorithms that solve these problems.

In this work, we plan to tackle Arabic NE recognition and classification with an approach using Long Short Term Memory (LSTM) neural networks. We use LSTMs’ ability to memorize long term dependencies to train a model for Arabic NE recognition, on a training dataset. The model is then used to predict the NEs for a sample of Arabic sentences in our test set. We tested our system on a pilot dataset. In its current version, it achieves a word level accuracy of 85%. More recently we trained our model on the more standard ACE 2007 dataset and achieved an F1 score of 57.54 for detecting boundaries and 53.31 for categorizing the named entity. However, adding part-of-speech as a feature reduced our performance. Overall, LSTM seems to be a promising model for Arabic NER. We plan to compare it with different existing baselines trained on other dataset. We also plan to identify an optimal feature set in order to study its impact on the accuracy of our predictor.

View the Report
project name

Classifying CIFAR-10 Images Using Unsupervised Feature & Ensemble Learning

We performed the image classification task on the CIFAR-10 dataset, where each image belongs to one of the ten distinct classes. The classes are mutually exclusive and are mostly objects and animals. The images are small (32 x 32 pixels), of uniform size and shape, and RGB coloured. We implemented and proposed an image preprocessing framework to learn and extract the salient features of the images. The method was demonstrated to increase the classification performance significantly. We testified such claim and saw a considerable improvement of more than 15% from the baseline (i.e., without preprocessing). We further experimented with various parameters and settings of the proposed method to tune the preprocessing frameworks. We also experimented with a variety of linear classifiers on the preprocessed images. We found out that a simple SVM classifier with linear kernel performs the best. We finally experimented with ensemble learning by combining a linear SVM with a multinomial logistic regression. The ensemble learning marginally improved on the simple linear SVM at a high computational cost.
Open Source

Find out more

project name

Contextual Spellchecker To Improve Human-Robot Interaction

This research focuses on developing a contextual spellchecker to improve the correctness of input queries to the multi-lingual, cross-cultural robot. Queries that have fewer misspellings will improve the robot’s ability to answer them and in turn improve the human-robot interaction. We focus on developing a language model based contextual spell-checker to correct misspellings and increase the query-hit rate of the robot. Our test bed is a bi-lingual, cross-cultural robot receptionist, Hala, deployed at the Carnegie Mellon Qatar reception.
Best Student Poster Award Qatar Foundation Annual Research Conference

Find out more

Work Experience

Technical Co-Founder - Inpleo (2015 - Present)

Inpleo is a cloud-based reverse-auction platform that changes traditional Request For Proposals (RFPs) by automating the bidding process and eliminating conventional purchasing dilemmas.

Pittsburgh Post-Gazette Carnegie Mellon News Qatar Tribune

Technical Lead - Oola Sports (2016 - 2018)

At Oola I deal with anything that has to do with the technical aspect of the company from developing the webstore to growth hacking. As a growth hacker, I focus on implementing data-driven marketing across marketing channels and product development to identify the most effective and efficient ways to grow the business.

NLP Group Research Assistant - Carnegie Mellon University (2015 - 2016)

The initial work focused on uderstanding Recursive Neural Network for semantic compositionality.

Applied Long Short Term Memory Neural Network to Arabic Named Entity Recognition (NER). On top of doing NER, other features such as Part of Speech (POS) were added in a deep learning fashion to try to improve the performance of the LSTM model.

Robotics Lab Research Assistant - Carnegie Mellon University (2012 - 2014)

Developed ETL for raw data and help developed statistical dashboard for log analysis of Hala (Bi-lingual robot receptionist).

Developed contextual spellchecker to improve human-robot interaction.

Developed API for adding robot functionalities. Modules for weather, location and translation query were also implemented.

Co-founder & Lead Developer - Ebs3.org (2008 - 2009)

Developed first e-learning portal in Mauritius for secondary schools. The idea was to help augment student learning by allowing them to have access to extra learning materials written by teachers through the e-learning portal.

Other Projects

Roomie Open Source

Roomie finds you an apartment to sublet with other people based on your preferences. Roomie, with its rank algorithm, provides you with the best apartments based on your work location, budget, number of rooms and number of roommates. Roomie is a web application that had been developed during the CarnegieApps Hackathon 16’.

View on GitHub

Video CDN Source Available Upon Request

In this project we explored how video content distribution networks (CDNs) work. In particular, we implemented adaptive bitrate selection, DNS load balancing, and pieces of OSPF.

Find Out More

Congestion Control with Bittorrent Source Available Upon Request

We implemented a BitTorrent-like file transfer application. The application runs on top of UDP with implemention of reliability and congestion control protocol (similar to TCP) for the application. The application is able to simultaneously download different parts, called “chunks,” of a file from different servers.

Find Out More

More projects