Project 1: Hotel Recommender

That project is a simple recommender based on data from hotels. Created for Recommender Systems Class conducted by @PiotrZiolo

Requirements:

1. Create (and activate) a new environment with Python 3.8 with conda.

   • Linux or Mac:

     conda create --name recommender python=3.8
     source activate recommender

   • Windows:

     conda create --name recommender python=3.8 
     activate recommender

2. Install jupyter notebook

   pip install notebook 

3. Install all needed packages

   pip install numpy pandas matplotlib seaborn ipython sklearn hyperopt

4. Start Jupyter Notebook

   • project_1_data_preparation.ipynb contains the process of preprocessing data
   • project_1_recommender_and_evaluation contains recommender

Algorithms used:

• To create user features I used one-hot encoding from all values that appeared in every column.
• To create item features I used pandas.get_dummies() which did basically the same as the method I used to create user features.
• To get negative interactions I took random all item_ids and random sample from user_ids with size of item_ids list. Then i iterated over item_ids and checked if there is a row in dataframe with that item_id and user_id.
• I've decided to use LinearRegressionCBUIRecommender, because it worked for me the best and got tuned relatively quick. It even beat Amazon recommender.

Validation results:

Recommender	HR@1	HR@3	HR@5	HR@10	NGCG@1	NGCG@3	NGCG@5	NGCG@10
LinearRegressionCBUIRecommender	0.049219	0.130007	0.174134	0.233198	0.049219	0.094502	0.113016	0.131827