xAI for Dota 2

Multiplayer Online Battle Arena (MOBA) games, such as Dota 2, have become increasingly popular, with competitions that attract millions of spectators (e.g., The International). The large number of professional matches have generated huge amounts of Dota 2 data, which raises the opportunity for applying data analytics and machine learning techniques to obtain interesting insights from the game.
This project is the first step towards understanding which factors are associated with victory prediction models for Dota 2. The goal is to build explainable prediction models for team victory in Dota 2. I further analyze how the factors vary across different types of matches: blowouts (matches where a team outscores the opponent by a large margin or extremely fast) and regular matches.

Source-code repository:

Dataset:

Paper associated with project: (Published at the 16th AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment - AIIDE 2020)

• Period of development: January 2020 - present


• Tags: Explainable AI, Predictive analytics, MOBA games

Game Review Sentiment

Several sentiment analysis techniques have been proposed in recent years, however the performance of existing popular techniques is still far from acceptable. This is more common when those techniques are applied in a domain different from the domain they were originally trained and built, such as training a sentiment classifier on Twitter data and applying on game data.
This project aims at better understanding how existing sentiment analysis techniques perform on game reviews. I rely on a large-scale dataset (approximately 12M reviews from Steam) to evaluate the techniques and identify the root causes of wrong classifications. I then provide recommendations for developers on what they need to prioritize to improve their sentiment analysis classification pipeline. Finally, I evaluated the performance of per-genre models (building different sentiment classifier models for different game genres).

Source-code repository: (repository is currently private due to the review process of the work. It will be available soon!)

Paper associated with project: (paper's access is currently restricted due to the blind review process of the work. It will be available soon!)

• Period of development: January 2019 - present


• Tags: Natural Language Processing, Sentiment Analysis, Game Reviews, Steam

Real-time Analytics

Real-time analytics is an exciting topic as it allows us to perform a series of different analysis on real-time data and act quickly. This project is a prototype application of a real-time news collector and analyzer. Currently, the application is able to collect the latest Donald Trump's tweets and Donald Trump-related news from CNN. The application also performs sentiment analysis on his tweets.
Link to the app

Source-code repository:

• Period of development: August 2018


• Tags: Data Analytics, Sentiment Analysis

Code Change Evolution

Code changes are performed differently depending on the platform of the application: mobile (e.g., Android) or non-mobile (e.g., desktop and web). However, the patterns of evolution of code changes in those platforms are still unknown. This project aimed at investigating the frequency and the co-evolution of different types of code changes in mobile and non-mobile platforms. I investigated 3 types of changes:

• source code (changes made to source code files)
• build (changes made to build configuration files)
• test (changes made to test files)

I collected and analyzed 500K commits from GitHub repositories. For the analysis, I built explanatory regression models to explain the factors that influence the frequency of changes and used a frequent itemset algorithm (Apriori) to reveal the co-occurring types of changes.

Source-code repository:

Paper associated with project: (Published at the 35th International Conference on Software Maintenance and Evolution - ICSME 2019)

• Period of development: July 2018 - December 2018


• Tags: Software Analytics, Statiscal Modelling, Data Mining