YelpBusinessesRecommender

Description:

• Develop a Recommendation System that suggests businesses to users based on the location, stars/ratings, category, reviews and business hours.
• Build on Apache Spark and Recommendation System.

Dataset:

The json has been converted into csv files using the sample code provided by the Yelp at https://github.com/Yelp/dataset-examples. The converted CSV can't be loaded into Github due to large size.

Pre-Processing:

1. Converting all files from json to csv for easier reading and utlizing the existing resources. [ref: https://github.com/Yelp/dataset-examples]

2. Loading business data. Dropping variables ['type','postal_code','latitude','longitude','review_count','attributes','address','is_open','hours','stars']

3. Filtered data to contain only state 'NC'

4. Loading review data. keeping only business id, user id, stars, funny, cool, and useful votes.

5. Merging both datasets by business id.

6. Updating value of the rating 'stars' by the formula (definiately it includes bias):

if stars>=3
        stars=stars+(0.1*funny+.6*useful+0.3*cool)/(funny+useful+cool+1)
else
        stars=stars-(0.1*funny+.6*useful+0.3*cool)/(funny+useful+cool+1)

Recommendation + Evaluation:

• Use apache spark to store and transform data on parallel (sc variable)
• Use Alternative Least Square as collaborative algorithm to recommend products
• Apply RMSE for evaluation of the model

Files:

|_ Work: where all code resides.
|_ Documents: All the documents provided by the professor.
|_ Sample Data: The filtered data files and sample initial files.

├── Work\
|   ├── ReadData.py            
|   ├── business_recommender.ipynb
├── Documents\
|   ├── Business Questions.txt
|   ├── FinalProjectProblemsProposal.pdf
├── Sample_Data\
|   ├── business.txt          
|   ├── checkin.txt
|   ├── merged_BR3.csv        
|   ├── tip.txt

Works Folder Details:

• The ReadData_AG.py is used for reading csv files, merging data, reducing dimensionality of data and data modification like modifying the rating. After completing data pre-processing, the final dataframe is stored in file to avoid processing the code again and again. Take a look at the path folder of data file.
• Before running evaluate.pynb make sure Spark is working properly on your machine. One will require Jupyter Notebook to run it. From terminal invoke the command IPYTHON_OPTS="notebook" $SPARK_HOME/bin/pyspark. Using the run button, the code can be executed.

Contributors:

• Anshuman Goel (agoel5@ncsu.edu)
• Siddharth Heggadahalli (ssheggad@ncsu.edu)
• Huy Tu (hqtu@ncsu.edu)