My Thesis for the Diploma in Computer Engineering & Informatics Department, at the University of Patras.
Sentiment analysis, or opinion mining, is the process according to which a system has to identify the author’s feeling - opinion expressed in his text, by deciding whether the aforementioned text is positive, negative or neutral. A common use of this technology is to discover how people feel about a particular topic.
The purpose of this project was to implement an extensible, scalable and reliable computer system which performs sentiment analysis on user textual data mined from sources around the internet like social media, forums, etc. It is comprised of two discrete subsystems.
This is a web app that enables users to graphically create and manage requests for sentiment analysis. It was created using the .NET Framework libraries (EntityFramework, MCV, etc).
This is the main subsystem that performs all of the project's core funcionality, by acting as a BackEnd Server in order to fulfil the requests received by the SentiMeter app:
- Firstly, it mines texts from the selected sources.
- Secondly, it analyses each one of them and classifies it as positive, negative or neutral.
- Lastly, it calculates results and sends them back to the app.
In order to achieve the desired aforementioned principles, this system was designed as a cloud system application using the Microsoft Azure Service Fabric Platform (PaaS), as well as other important and basic programming patterns, techniques and concepts such us:
- Cloud Programming
- Microservices architectures
- Actor Model
- RESTful Services
- Dependency injection
- Modular Hierarchies
Therefore, the system can easily be extended to support :
- additional mining sources like social media (Facebook, Instagram) and forums
- the implementation of other sentiment analysis algorithms, for each different source, combination etc, making easy to observe their results and test their efficiency
- additional features like results caching, more specific search criteria, live observation of the progress of users' sentimental trends and opinions, etc
