Video game recommendation system

Note: The app is currently not live but can be run by cloning the repository, installing the requirements in requirements.txt and running

streamlit run NSRecommender_app.py

End-to-end project involving web scraping, clustering analysis and web app development.

This is an end-to-end project where I built a web app that, given an input Nintendo Switch game, gives you a set of recommended video games using unsupervised learning techniques and natural language processing on the games' gameplay or plot.

The app is accessible from the following link: https://hectoramirez.github.io/Recommender-app.html

Files

• Switch_games.ipynb: Web scraping Wikipedia articles using Beutiful Soup. The full dataset is generated using the code in this notebook.

  First, I scrap the tables in

  • https://en.wikipedia.org/wiki/List_of_Nintendo_Switch_games_(A–L)
  • https://en.wikipedia.org/wiki/List_of_Nintendo_Switch_games_(M–Z)#Games_list_(M–Z)

where I select only the games with links to their wiki pages. Then, for each game, I access their pages and scrap the paragraphs contained in the Gameplay or Plot sections or both if exist.

Finally, I clean the text and drop null entries.

• Game_similarity.ipynb: Here the texts are processed by tokenizing and stemming them, and are vectorized using NLTK's TF-IDF vectorizer.

  From the TF-IDF matrix, the similarity distances between the texts are computed by substracting the cosine of vectors from 1.

  Finally, recommendations are queried using the matrix: once a game is selected, the top 5 closest games are returned.

• NSRecommender_app.py: Streamlit web app. The app is hosted in an Amazon Web Services EC2 instance, accessible from here.

Contact

For comments or questions, drop me a message at vanhramirez@gmail.com.