Event Recommendation Systems
Business Design
- To design a personalization based event recommendation systems for event search.
General Instruction
- Design a web service with
RESTful APIs
in Java to handle HTTP requests and responses - Frontend: an interactive web page with
AJAX
technology implemented withHTML
,CSS
andJavaScript
. The Event Recommendation Website realizes three main functions:- Search events around users
- Favorite events they like and also delete events they don’t like anymore
- Get recommendation of events around based on their favorite history and distance to where events will be hold
- Backend: use
Java
to process logic request, and some supports are as below:- Built with both relational database and NoSQL database (
MySQL
andMongoDB
) to support data storage from users and items searched in TicketMaster API - Design content-based recommendation algorithm for event recommendation
- Built with both relational database and NoSQL database (
- Deploy website server on
Amazon EC2
: Event Recommendation System - Analyze website traffic both online and offline with ELK (
ElasticSearch
,Logstash
andKibana
) andMapReduce
in MongoDB
Infrastructure Design
- 3-tier architecture
- Presentation tier: HTML, CSS, JavaScript
- Data tier: MySQL, MongoDB
- Logic tier: Java
- Local and remote development environment
Local development environment
Remote development environment
API Design
- Logic tier(Java Servlet to RPC)
- Search
- searchItems
- Ticketmaster API
- parse and clean data, saveItems
- return response
- History
- get, set, delete favorite items
- query database
- return response
- Recommendation
- recommendItems
- get favorite history
- search similar events, sorting
- return response
- Login
- GET: check if the session is logged in
- POST: verify the user name and password, set session time and marked as logged in
- query database to verify
- return response
- Logout
- GET: invalid the session if exists and redirect to
index.html
- POST: the same as GET
- return response
- GET: invalid the session if exists and redirect to
- Register
- Set a new user into users table/collection in database
- return response
- Search
APIs design in logic tier
- TicketMasterAPI
Official Doc - Discovery API - Recommendation Algorithms design
- Content-based Recommendation: find categories from item profile from a user’s favorite, and recommend the similar items with same categories.
- Present recommended items with ranking based on distance (geolocation of users)
Process of recommend request
Database Design
- MySQL
- users - store user information.
- items - store item information.
- category - store item-category relationship
- history - store user favorite history
MySQL database design
- MongoDB
- users - store user information and favorite history. = (users + history)
- items - store item information and item-category relationship. = (items + category)
- logs – store log information
Implementation Details
- Design pattern
- Builder pattern:
Item.java
- When convert events from TicketMasterAPI to java Items, use builder pattern to freely add fields.
- Factory pattern:
ExternalAPIFactory.java
,DBConnectionFactory.java
ExternalAPIFactory.java
: support multiple function like recommendation of event, restaurant, news, jobs… just link to different public API like TicketMasterAPI. Improve extension ability.DBConnectionFactory.java
: support multiple database like MySQL and MongoDB. Improve extension ability.
- Singleton pattern:
MySQLConnection.java
,MongoDBConnection.java
- Only create specific number of instance of database, and the class can control the instance itself, and give the global access to outerclass
- Builder pattern:
User Behavior Analysis
- Online (ElasticSearch, Logstash, Kibana)
- Use Logstash to fetch log (in NoSQL-like form), then store data in ElasticSearch, finally use Kibana to analyze the data in ElasticSearch, getting some tables and graphs like APIs use, request status, geolocation of visitors, etc
Remote development environment
- Offline (MapReduce in MongoDB)
- Copy-paste some logs from Tomcat server
- Purify log data and store in MongoDB
- Do
mapreduce()
in MongoDB - Get a list of timebucket-count in descending order of count, then find the peak time of website traffic