EDUCATION Ph.D. Computer Science
Florida Atlantic University,
Boca Raton, FL
GPA: 4.0
M.S. Computer Science
Science and Research University,
Tehran
GPA: 3.8
B.S. Computer Science
Azad University,
Tehran
GPA: 3.6
Similarity Measurement
- Proposed a new similarity measurement to eliminate the problem of Cosine Similarity in high dimensional data.
- Researched K-means, PCA, Symmetric NMF, Normalized cuts, and SVM to compare the performance of our measure with that of the cosine similarity.
- End result shows based on ANOVA and Tukey’s test results, our measure outperforms cosine similarity (https://journalofbigdata.springeropen.com/articles/10.1186/s40537-017-0083-6). Sentiment Analysis
- Conducted Sentiment analysis on financial data to evaluate the impact of investor sentiment on stock market price.
- Investigated Doc2Vec (using Genism), long short term memory: LSTM (using Theano) and Convolutional Neural Network: CNN (using Tensorflow).
- Predicted investor’s sentiment to ~31% of transactions with more than 90% accuracy. Predicting Top Authors
- Examined the word usage of investors who can predict stock prices correctly.
- Studied Doc2Vec and Convolutional Neural Network (CNN).
- Predicted top authors to ~26% of transactions with more than 80% accuracy. JM Family Enterprises, Inc.
Data Scientist
2017-Present
Florida Atlantic University
Python, R, Machine Learning
2015-Present
Recommendation System Based on User Behavior
- Predicted the likeliness that a new user would buy a product based on the behavior of other users. Used Locality Sensitive Hashing and Cosine Similarity to design and implement a collaborative filtering engine.
- The engine is able to make recommendations to ~32% of transitions with more than 85% accuracy.
Recommendation System Based on Product Features
- Probed Singular Value Decomposition to recommend products to a new customer based on a product-feature matrix.
- Recommended products to ~29% of new users with more than 80% accuracy using content- based filtering.
Predicting Claim Amount
- The goal was to predict claim payments based on the characteristics of the customers’ vehicle. Investigated Gradient Boosted Decision Trees to predict the amount of a claim and utilized the Random Forest Information Gain, to obtain the most relevant features.
- Predicted the claim amount for ~30% of transactions with less than 10% root mean square error.
Finding Common Problems
- Assisted the business department in finding common product part failures by exploring association rules to discover the relation between products and their features.
- Used measures of significance and interest, including support, confidence, and lift, to select interesting rules based on business criteria from the set of all possible rules. WORK EXPERIENCE
Sahar Sohangir
ac5fex@r.postjobfree.com
Software Architecture
Auction theory
Matlab
2010-2011
Conducted Research on Service Assignment Methods
- Conducted research on service-oriented architecture to effectively assign service providers to service consumers.
- Investigated auction algorithms and game theory. Implemented service assignments based on auction algorithms in Matlab.
Azad University
Lecturer
2011-2013
MACHINE LEARNING
AND DATA MINING
ALGORITHM
EXPERIENCE
TECHNICAL EXPERTISE
PUBLICATIONS
Freelance Technician
Senior Software Engineer
2007-2009
- Deep Learning for Financial Sentiment Analysis. Knowledge Discovery and data Mining
(KDD) 2016. Sahar Sohangir, Dingding Wang, Anna Pomeranets.
- Document Understanding Using Improved Sqrt Cosine Similarity. International Conference on Semantic Computing (ICSC) 2017. Sahar Sohangir, Dingding Wang.
- Update Summarization using Semi-Supervised Learning Based on Hellinger Distance. Conference of Information and knowledge Management (CIKM) 2015. Dingding Wang, Sahar Sohangir, Tao Li.
- Finding Expert Authors in Financial Forum Using Deep Learning Methods (ICSC) 2018. Sahar Sohangir, Dingding Wang.
- Financial Sentiment Lexicon Analysis (ICSC) 2018. Sahar Sohangir, Nicholas Petty, Dingding Wang.
- A new Method for service Binding in service Oriented Architecture. Advance in Information Science and Service Science 2010. Sahar Sohangir and MirAli Seyyedi.
- A Service Binding Method using Forward Auction. International Journal of Information Processing and Management 2011. Sahar Sohangir and MirAli Seyyedi. Proficient in Programming and Scripting Languages: Python, R, MATLAB, Weka Java, C/C++, C#
MySQL Unix Shell Scripts
- Random Forest, Boosted Trees, Support Vector Machine, Logistic Regression, Naïve Bayes, Feature Selection, and Dimensionality Reduction
- Deep Learning and Neural Networks
- Bagging and Boosting
- Ensemble and Stacking Methods
- Natural Language Processing
Adjunct Professor
- Taught Database Systems, Data Structure, Analysis of Algorithms, and programming in C++ at Azad University.
- Managed a freelance team to implement Oracle Server Database through development and extension of logical and physical data models.
- Evaluated business performance utilizing different key performance indicators. Matin System Rahnegar
Software Engineer
2006-2008
- Designed and implemented Integrated Software Management systems for various companies.
- Implemented an online test for new users to measure their speed and ability to work with our ISM system.