Armin Tabari

adh0in@r.postjobfree.com j 980-***-**** j LinkedIn: armintabari j Github: armintabari

**** **** *** *** *, San Francisco, CA 94122

EXPERIENCE

Text Analytics Lab, UNCC Charlotte, NA

Graduate Assistant Aug. 2015 - Present

Emotion Detection Using GRU, Attention and BERT: Created a Recurrent Neural Networks model to detect discrete emotions in tweets. The model reached 36% increase in F-measure compared to published results by using bidirectional GRU and attention mechanism to create better representation for text.To compare the model to transformers, I used HuggingFace to fine-tune the BERT model for emotion detection in tweets. The results showed slight improvement compared to the combination of GRU and attention network.

Emotional Word Embedding: Changed pre-trained word vector spaces to include emotional information acquired from Plutchik s emotional model without losing original information in the model. This can be especially useful as features in neural network-based systems. This resulted in a vector space with higher performance in detecting emotion relations by 20% compared to standard embedding models, and higher performance in emotion detection model using neural networks.

Sentiment Analysis: Used NLP and deep neural networks (CNN, LSTM) to perform sentiment analysis of stock market tweets with 92% accuracy. The result is a public dataset of labeled stock-related tweets over three years.

Topological Data Analysis: Used entity extraction to detect and extract character appearances in 19th century novels in order to generate topological features. We showed that authors could be detected based on these features with average of 77% accuracy.

Extracting Emotion and Blame in Social Media: We used flash surveys, dispersed via social media, during and after a high-impact event to generate training data and labels for emotion and blame frames in social media. To classify social media text as the event unfolds, we used Logistic Regression to extract features and a Random Forest as the classifier. We achieved 91% accuracy.

Daystar ITC Tehran, Iran

Senior Software Designer April 2008 - Nov. 2014

Artomim Customized T-shirt Shopping: Lead a team that designed and implemented the database and backend of the website.

Nasan Hotel Room Search and Registration Services: Lead the team for database design, and part of the backend programming group.

ADDITIONAL EXPERIENCE & ACHIEVEMENTS

Publications: Eleven papers in proceedings and workshops of different conferences such as: ACL, ECML, and SemEval and on Arxiv. The manuscripts are available on my Google Scholar.

Instructor and Teaching Assistant: I was instructor of record for three semesters for Logic and Algorithms course. I was teaching assistant for various undergraduate and graduate courses including Data structures and algorithms, and Software Systems Design and Implementation.

Fellowship: Awarded summer funding from the UNC Charlotte Graduate School after a competitive application and review process. The Graduate Student Summer Fellowship (GSSF) proposal was titled: “Refining Word Embeddings to Capture Emotional Information of Words”.

EDUCATION

University of North Carolina at Charlotte Charlotte, NC Ph.D. Computer Science 2020

Thesis: Detecting Emotions in Textual Data, Focusing on Text Representation University of North Carolina at Charlotte Charlotte, NC Masters in Computer Science; GPA: 3.9/4.0 Aug 2015 - 2017 Sharif University of Technology Tehran, Iran

Bachelor of Technology in Computer Science & Engineering; GPA: 3.4/4.0 Aug 2003 - 2009 SKILLS

Conceptual skills: Machine learning, Neural Networks, Natural Language Processing

Languages: Python, R, SQL, Java

Libraries: TensorFlow, Keras, Scikit-Learn, Numpy, Pandas, Jupyter

Technologies: GitHub, AWS, HPC

Contact this candidate