Authors:
- Sotiris Loizidis (sotirisloizidis@gmail.com)
- Marcos Antonios Charalambous (mchara01@cs.ucy.ac.ucy)
Twitter has become an important communication channel in times of emergency. The ubiquitousness of smartphones enables people to announce an emergency they are observing in real-time. Because of this, more agencies are interested in programmatically monitoring Twitter (i.e. disaster relief organizations and news agencies).
But it’s not always clear whether a person’s words are actually talking about a legitimate disaster.
In this project, we started by inspecting our dataset to gain any useful information about what kind of text we are facing by the use of plots during the EDA analysis phase and then proceeded to check a plethora of relevant text processing models and subsequently choosing the best and ultimately proceed in the best possible prediction.
More specifically, what we achieved is building a machine learning model that predicts which Tweets are about real disasters(predict 1) and which ones aren’t(predict 0). For the project's purpose we where given access to a dataset of 10,000 tweets that were hand classified.
Evaluations are done using F1 between the predicted and expected answers.
Each sample in the train and test set has the following information:
- The id of a tweet
- The text of a tweet
- A keyword from that tweet (may be blank)
- The location the tweet was sent from (may be blank)
train.csv: Dataset used for model building which contains the above columns.
- python disasters.py
- python parameter_tuning_logistic_regression.py
- python ngrams_feature_selection.py
- python submit.py (If sumbitting to the competition is desired)
Warning: There are many non-default modules installed in this project. Please install anaconda first and then install any other missing module using pip install needed to successfully execute the files in this project.