/Recommendation-System

Primary LanguageJupyter Notebook

What is this repository?

This is a repository that I had practiced Instagram Crowling for sentiment analysis.

What kind of activity contains?

  • Instagram Crawling : Practice Python Crowling using BeautifulSoup and Selenium
  • Melon Crawling : Get 4000 songs from Melon Chart
  • Sentiment Analysis for single sentence (sentiment_analysis.ipynb) => apply machine learning model (LightGBM, Logistic Regression)
  • Sentiment Analysis for several sentence(same as dialogue) (sentiment_model_practice.ipynb) => using LSTM model
  • Recommendation System using sentiment analysis (참빛설계 학습모델 테스트 - 멜론 Top 100)

What I used?

Programming Language

  • Python Jupyter Notebook

Library

  • Selenium
  • BeautifulSoup
  • Numpy
  • Pandas
  • Keras
  • Konlpy
  • Tf-IDF

Data

  • AI Hub Data
  • Melon Song Information Data

Modeling

Model Name Accuracy
Logistic Regression 17.36%
Random Forest 17.69%
LightGBM 18.04%
LSTM 68.05%

How to Recommend?

Since the accuracy was not high enough, cosine similarity was used to make a better recommendation.

  • sentiments are predicted using a model learned from the user's SNS posts and collected songs.
  • extract only the same of the predicted emotions from SNS and songs.
  • compare the extracted cosine similarity and recommend the top 5 songs with the highest similarity.