- The purpose of this project is to see if it is possible to identify users based solely on their search histories.
- I collected the search data of 11 friends and family (it's important to us that their data stayed confidential, for this reason I didn't upload any of the data, just the python notebooks. Any data I used as an example in those notebooks are mine personally)
- I then processed this data into a cleaned data frame, and created a model to predict a user given a search term.
- dataPreprocessing.ipynb - This file contains all preprocessing for users search histories, from html to cleaned data frame for each user.
- model.ipynb - This file shows the process of turing preprocessed data in to a model.
The best balanced accuracy I was able to get was ~%72 using logistic regression and bag of words classification, please see 3.3 in my report for more details.
- Python 3.9.9
- Python NLTK 3
- scikit-learn 1.0.2
- Beautiful soup 4.8.1
- pandas 1.3.5
- numpy 1.21.5
- Other critical packages seen in my notebooks