Anime-Database-and-Recommendation-System

Capstone

Introduction

• Business problems:

  • Created my own dataset by combining the scraped data (14,000 rows) from https://www.anime-planet.com/ and data from kaggle.com including over 4 thousand lines of animes information and over 3 million rating data(over 70 thousand users and over 4 thousand animes)
  • Built my own database system and recommendation system
  • Conducted an interface using HTML and WT Form and connected it with the database using Python Flask in Sublime Text
  • Gave users some recommendation animes and some overview analysis

• Home Page

• Features:

1. Accurate Search
   • Search an anime information by anime id
   • Search an anime information by any key words of its title
   • Resulted by all the information for all qualified animes
2. Scope of the search(Sort)
   • Filtrated by some categorical variables such as on going(True or False), season of realease and so on
   • Sorted by numerical variables such as start year, rating and so on
   • Ordered by (sacending or descending) following by the sorted by option
   • Resulted by all the information for all qualified animes

3. Analyze
   • Entered the excat anime id and selected a feild (genre, type, duration and so on) for analyzing
   • Resulted by the information of this anime and a graph of this feild following
   • Example:
4. Recommendation System
   • Trained severval algorithms and chose a suitable one
   • Tune hyper parameter for this model
   • Build a recommendation system
   • Gave a survy to the user to rate
   • Retrained this model by adding these new rating
   • Resulted by a list of recommendation animes
   • Survy:
     
   • Recommendation List:
     

Techniques

• Notebook link: https://github.com/JRRRRRRR/Anime-Database-and-Recommendation-System/blob/main/final.ipynb
• Data collecting, Data Scraping, Data cleaning, Exploratory data analysis, recommenation system and Visualization
• Python, Jupyter Notebook, HTML

App Instruction

Welcome to try this app yourself. Here are some instructions and tips.

• Download clean datasets from data link in data file, and all python folder
• Launch the app.py using terminal or other software
• Copy the website link(mostly is http://127.0.0.1:5000/) showing on the output and paste it on your browser

Conclusion

• In the database system, I totally got three dataset. However, two of them are having a lot of repeated features. At last, I use the dataset from https://www.anime-planet.com/ as the basic one and added more features from the other dataset. Also, I scraped more features in this website and added into it.
• In the recommendation system, I had tried some algorithms(including KNNBasic(), KNNBaseline(), KNNWithMeans(), SVD(), SVDpp(), NMF(), SlopeOne(), CoClustering()) in this recommendation system. At last, CoClustering algorithm gave me a best outcome with RMSE 2.8484734984857 and MAE 2.04126179225725 in a rating scale of 10. What's more, this algorithm gave me the shortest training time which is around 42 seconds.

Recommendation

• For users:

  • If you have no ideas which anime to watch, you can try Sort or recommendation features.
    • Sort feature can filtrate you prefer scope or just sorting the animes by want watch variables and recommendation features will give you some recommendation animes basing on your rating of the survy animes.
  • If you have some ideas about what you want to watch, you can use search feature.
    • Search animes by key words to see any other related animes

• For analysts:

  • To pass graphs and charts to the front desk, I found two methods. One is saving them on the local machine and supply the path to the front desk. The other one is encode them into base64.
  • It is a pretty real case in our daily life. The training time is much longer than we excepted because of the huge dataset. I suggest than we can slightly dope down our training size base on the dataset size.

Future Work

• Algorithm: Try more algorithms from other packages
• Hyper Parameter Tuning: tune more parameters of the models and use random search before grid search
• Exploratory Data Analysis: Exploratory more characteristics and more details in the analyze feature
• Interface: Learn more about the html or other tools to make the UI more nicer

Summary

This is the last project for me in Flatiron School. I tried to apply more of the skills I've learned here. Some analysis maybe not so comprehensive due to the limited time. However, I reviewed many knowledge and also learned more new skills. It is a big project than before and more practical. After completing this capstone, I feel more fonfident for the road to data science.