Skip to content

Latest commit

 

History

History
37 lines (28 loc) · 2.15 KB

README.md

File metadata and controls

37 lines (28 loc) · 2.15 KB

Coursera Data Mining Capstone Project

This repository contains all the codes, data and reports that I wrote for this capstone project of Coursera Data Mining Specialisation.

There are one final report and five reports corresponding to six tasks (the 4th and 5th tasks are combined).

  1. Task I -- Data Exploration
  • topic mining among all the yelp reviews for restaurants, and insights from the results
  • topic comparison, and the difference analysis
  • data distribution (rating, review number trend, etc.) of reviews
  1. Task II -- Cuisine Clustering
  • Cuisine clustering by similarity
  • Clustering result comparison between using different features / results of feature extraction (TF-IDF, LDA)
  • Clustering result comparison between different clustering algorithms (K-means and aglomerative)
  1. Task III -- Dish Recognition
  • Dish recogntion using three different algorithms, comparison and analysis
    • ToPMine, unsupervised frequent pattern-based phrase mining algorithm
    • SegPhrase, dish mining using external knowledge base, and many other addon features (phrasal segmentation) for quality phrase mining
    • word2vec, dish mining based on word association
  1. Task IV & V -- Popular Dish and Restaurant Recommendation
  • Rank dishes by ratings, sentiment scores of reviews
  • Restaurant recommendation when given dish name, ranked by similar features
  • Visualisation of the above two results
  1. Task VI -- Hygiene Prediction
  • Predict hygiene condition of restaurants based on known features (location, reviews, ratings, tags, etc.)
  • Algorithms used: ensemble algorithms (random forest, xgboost) and regression on top of them
  1. Final Report
  • A summary of these six tasks
  • Raised some insights of the data set (reviews) that are not mentioned in the earlier tasks
    • what customers may care, which might be helpful to restaurant owners
    • how does the review topic distribute for those 1. frequent yelp users, 2. return customers