Informal Text Sentiment Analysis System

In the realm of digital communication, particularly social media, there exists a plethora of informal texts — be it a tweet on Twitter or a status on Weibo. This vast sea of data holds valuable insights on the sentiments and emotions of users towards a multitude of subjects, from products to news and more. Tapping into these sentiments has always been a challenging yet rewarding endeavor. It offers benefits like tailoring product recommendations, predicting societal reactions, and even forecasting financial market dynamics.

However, the nature of such social texts sets them apart from conventional language structures. They are inherently informal, often brevity-laden, and ridden with noise in the form of slang, emojis, and unconventional grammar. These characteristics demand the development of specialized machine learning systems to decode and derive sentiments from such texts effectively.

This project aims to embark on this challenge. Here, we design a sequence labeling system tailored for informal texts. Grounding our approach in the principles of Hidden Markov Model (HMM) that we've studied, we aspire to lay the foundational stone for a more intricate, intelligent sentiment analysis system in the future.

For this project, we delve deep into two unique datasets: RU and ES. These datasets, containing labeled and unlabeled data, serve as the backbone for training our model. The labeling system identifies tokens with sentiments - positive, negative, and neutral - by designating them as "Beginning" or "Inside" of the sentiment entity. The "Outside" label is reserved for tokens that lie outside any sentiment context.

In essence, this endeavor is not merely about creating a sentiment analysis tool. It's about understanding the language of social media, respecting its nuances, and extracting invaluable sentiments that can help businesses and individuals make informed decisions.

For more information regarding the project, please read more here.

Setup & Execution

1. Clone the Repository

   git clone https://github.com/verneylmavt/hmm-text_analysis-py.git

2. Navigate to the Project Directory

   For ES Dataset:

   cd hmm-text_analysis-py/ES

   For RU Dataset:

   cd hmm-text_analysis-py/RU

3. Execute The Jupyter Notebook / Python Files

   PLEASE REFER TO JUPYTER NOTEBOOK FOR FULL DOCUMENTATION

   • Jupyter

     For ES Dataset:

     jupyter notebook es_hmm_dev_test.ipynb

     For RU Dataset:

     jupyter notebook ru_hmm_dev_test.ipynb

     Once inside Jupyter, go to the "Cell" menu and click "Run All".

   • Python

     For ES Dataset:

     python es_hmm_dev_test.py

     For RU Dataset:

     python ru_hmm_dev_test.py

     If all goes as planned, "Everything Executed👍" will greet you in the terminal.

   ---

   Heads Up: There are no infinite loops in the code, it takes ~10s to execute

4. Test F-Score

   After executing the code, the program will generate following files:

   • dev.p1.out
   • dev.p2.out
   • dev.p3.2nd.out
   • dev.p3.8th.out
   • dev.p4.out
   • test.p4.out

   Now, you can check the F-Score via:

   python evalResult.py dev.out <FILENAME_HERE>

Dependencies

Note: Ensure that in your working environment, you have all the required dependencies. If it is not installed, you can get it using:

pip install jupyter
pip install numpy
pip install pandas

Or alternatively:

pip install -r requirements.txt

Contributors

This project was made possible thanks to the hard work and dedication of the following team members:

• Dorishetti Kaushik Varma
• Elvern Neylmav Tanny
• Harini Parthasarathy

Kudos to all contributors for their invaluable insights and dedication.