REZA NOURALIZADEH GANJI

 Google Scholar | ï LinkedIn |  ORCiD | [ ResearchGate

# rezang52@gmail.com | Ñ rezang.github.io

EDUCATION

Master of Artificial Intelligence 2020 – 2023

K. N. T. University of Technology Tehran, Iran

•

Notable Courses: Natural Language Processing, Neural Networks, Recommender Systems, Information Re-

trieval, Evolutionary Computation

GPA: (18.32/20 – 3.88/4)

Bachelor of Computer (Software) Engineering 2017 – 2020

Shomal University Amol, Iran

•

Notable Courses: Machine Learning, Artificial Intelligence, Algorithm Design, Data Structures, Formal Lan-

guages and Automata Theory, Engineering Probability and Statistics

•

Thesis: A machine learning-based model for spam detection on mobile phone short message service (SMS);

under supervision of Dr. Hamidreza Koohi `

•

•

Ganji, R.N. and Tohidi, N. (2025). Sentiment Analysis of Short and Incomplete Text using Transformers and

Attention Mechanism.

PAMR: Persian Abstract Meaning Representation Corpus 2 Published

•

Tohidi, N., Dadkhah, C., Ganji, R.N., Sadr, E.G. and Elmi, H., 2024. PAMR: Persian Abstract Meaning Rep-

resentation Corpus. ACM Transactions on Asian and Low-Resource Language Information Processing, 23(3),

pp.1-20.

Improving Sentiment Classification for Hotel Recommender System 2 Published

AI Researcher — Supervisor: Dr. Chitra Dadkhah K. N. T. University of Technology

Study: Advanced Sentiment Polarity Detection for Short and Incomplete Texts 2022 – 2025

•

Motivation: The presence of short and incomplete text, such as tweets, with misspellings, grammatical errors,

and a lack of context could cause critical challenges for sentiment analysis and traditional NLP models.

•

Methodology: For precise classification, a novel 3-phase system was architected to handle noisy texts. It auto-

corrects data, uses autoencoder nets for denoising, and fuses intermediate features extracted from RoBERTa

LLM.

•

•

Motivation: The lack of essential semantic resources, like the AMR corpus for low-resource languages like

Persian, can restrict advanced NLP research like semantic parsing and text generation.

•

Methodology: To develop the AMR corpora, 1,020 sentences were annotated by modifying guidelines for

unique Persian features. The data augmentation was performed to generate 888 synthetic sentences from the

corpus.

•

Findings: Co-developed and released the first Persian AMR corpus. Its use in data augmentation boosted a

sentiment analysis model’s F1-score and accuracy by 12%. The research was published in an ACM journal.

Methodology: For precise recommendation, an RS was proposed to benefit from sentiment analysis. It utilized

a T5 model for data balancing via augmentation, and a cross-lingual XLM-RoBERTa classifier with an attention

mechanism used model’s hidden states.

•

Findings: It Achieved an 89% F1-score on the TripAdvisor data and surpassed baselines by 5%. Its efficient

integrated architecture cuts inference time by over 60% compared to the baseline. This research was published

in the CYS journal.

RESEARCH INTERESTS

•

Credential ID: LCKQELFDBRYW

Deep Learning Specialization 2 Coursera

•

The five courses in this specialization educate students how to design, develop, and optimise CNNs, RNNs,

LSTMs, and Transformers utilising Dropout, BatchNorm, Xavier/He initialization, and other approaches.

•

Credential ID: K8PGAYP9BUZC

CONFERENCES & PRESENTATIONS

How do search engines use machine learning methods? May 2019

Shomal University Artificial Intelligence’s Research Seminar

TECHNICAL SKILLS

Programming: Skilled in Python, Familiar with: PHP, HTML, CSS

Deep Learning: Transformers, Attention mechanisms, Large Language Models (LLMs), Recurrent Neural Network

(RNN), Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Auto Encoders

Machine Learning: Clustering, Decision Tree, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), En-

semble Models, Logistic Regression

Math/Theory: Linear Algebra, Probability & Statistics, Multivariate Calculus, Optimization Methods

AI Packages: Pytorch, Numpy, Pandas, Matplotlib, WandB, PLotly, Scikit-learn