A text mining challenge from course INF582.
Please check out the github version of our repository for a tidier display of this README.md file !
The structure of the directories should be kept as it is in our project:
- data/
- *.csv (put the node_information.csv file here)
- *.txt (testing_set.txt and training_set.txt)
- featureEngineering/
- abstractFeatures/
- graphArticleFeatures/
- graphAuthorsFeatures/
- journalFeatures/
- lsaFeatures/
- originalFeatures/
- submissions/
- *.csv (all the submission.csv files will be exported here)
- report/
Note that in each folder Features/ a folder output/ should be included.
We compared each features individually on the same cross validation training-testing-set, using the same regressor: RandomForestRegressor with random_state set to 42.
Bear in mind that these are feature sets, and not a single feature. For instance, the graphAuthors feature set is composed of 7 features ("meanACiteB_col", "maxACiteB_col","AOut_col", "BIn_col","ACiteAMean_col", "ACiteASum_col","BOut_col")
| Feature Set | Individual F1 score |
|---|---|
| 'lsa' | 0.576185 |
| 'original' | 0.811745 |
| 'graphAuthors' | 0.879161 |
| 'graphArticles' | 0.992431 |
| 'journal' | 0.611100 |
| 'similarity' | 0.778112 |
We compared these classifiers, and obtained the respective performance
| Algorithm | F1 score |
|---|---|
| Gradient Boosting | 0.9 |
| Random Forest Regressor | 0.8 |
| Logistic Regression | 0.9 |
To prevent overfitting, we did cross validation on a sample of the training set of approximately the same size For each classifier, explain the procedure that was followed to tackle parameter tuning and prevent overfitting.