Amazon review data served through PostgreSQL

Setup

CAUTION: This is only tested wit Book reviews.

1. Download dataset from: http://deepyeti.ucsd.edu/jianmo/amazon/index.html
2. Put dataset and meta file into ./import_job/datasets directory named:
   • data.json.gz
   • meta.json.gz
3. Run start.sh
4. Run ./import_job/run.sh
5. Check if import was successful at
   • http://localhost:8080/?pgsql=db&username=postgres&db=dataset
   • Username: postgres
   • Password: postgres

How to query the data

We can consider each reviewer as a partition in our Federated Learning experiment. To get all partition IDs run:

SELECT DISTINCT data_json.values->>'reviewerID'
FROM data_json;

To get all reviews for a single reviewerID you might run

SELECT *
FROM data_json
WHERE data_json.values->>'reviewerID' = 'AGUJ97UTICV4W';

SQL Functions

To optimize some queries a few materialized views and SQL functions are created.

Get reviews by author ID:

Id range with the book reviews dataset is from 1 to 15362619.

SELECT * FROM reviews_by_author_id(1);

WARNING: The numerical author id / string reviewerID mapping is not stable and even with the same base dataset you might end up with a different numerical id / reviewerID mapping after recreating the database. They are stable between queries though.

Get authors with min number of reviews:

SELECT * FROM authors_by_min_num_reviews(10);