AI Projects & Publications

Here is a list of the research projects (in AI) that I have published:

  • Transformer as a Regression Model for Fine-Grained Scoring of Textual Semantic Relations

This paper and the accompanying code were utilized in the 18th International Workshop on Semantic Evaluation (SemEval-2024) for the task of automatically detecting the degree of semantic relatedness between pairs of sentences.

paper code
  • A Transformer-Based Approach to Detect Machine Generated Text

This paper and the respective code were utilized in the 18th International Workshop on Semantic Evaluation (SemEval-2024) for the task of detecting machine-generated text.

paper code
  • Persian Formality Style Transfer

This project was done as my Master's thesis which involved a model trained in an unsupervised fashion to convert informal Persian into its formal equivalent.

huggingface

  • Persian Paraphrase Generation

A T5 model for generating paraphrases of Persian sentences. Trained on the Tapaco dataset.

huggingface

  • Supervised Sentiment Analysis on Tweets

A class project to classify a sample data-set of 1200 tweets to either positive or negative class, based on the type of words used in those tweet.

code

  • Unsupervised News Article Clustering

This project aims to cluster a news data-set to one of the four classes which are: Sports, Economics, Politics, and Culture using K-Means & Mixture Models in Python. The data-set is crawled from Tasnim News Agency and the crawler used for this task is written in Julia language.

code

  • Product Review Classification

In this project I used deep learning on a data-set from Digikala, an Iranian E-commerce to correctly label the review of customers over a range of products and whether they recommend it or not, or if they are unsure about it.

code slides
  • Supervised Movie Classification

The goal of this project is to classify the genre of movies by their plots. 4000 documents were used to train the current model. Data-set used for this task was obtained from Kaggle.

code

  • Neural Speech Synthesis with Transformer Network

This project was done for the Speech Processing course which used Mozila's Common Voices dataset (due to the lack of available audio in Persian) as the source to train a transformer network to synthesize Persian speech.

original paper