Reputation: 121
I am planning to implement product recommendation in my eCommerce site using neo4j graph database .
Recommendation will be based on User action on a product. Actions will be
- Product View ,
- Rating ,
- Read book
- Download book ,
- Purchase ,
- Add to card ,
- Review ,
- Share
- Some more action applicable to our site.
The graph structure will be
User (Node )
Product ( Node )
Action ( Relationship between User and Product node )
Later I will add social relationship between the User nodes .
I found different recommendation methods and algorithms from my initial analysis from internet . Following are the list which is categorized based on my understanding . Some of term might be incorrect or redundant or wrong categorization ( Correct me if I am wrong ).
- Item-Item similarity
- k-nearest neighbors (k-NN) algorithm
- Pearson correlation coefficient.
- User-User similarity
- Matrix Factorization
- Singular Value Decomposition (SVD)
- Restricted Boltzmann Machines (RBM)
- Non-Negative Matrix Factorization ( NNMF )
- Latent factor analysis
- Co-visitation analysis
- Latent topic analysis
- Cluster model
- Association rule
- Bi-gram matrix association rule
- Ensembles
My problem is to identify which all methods are applicable in my eCommerce site and can be solved using neo4j graph database ( Based on the above model ).
Upvotes: 4
Views: 1326
Reputation: 45043
Your question is more about data science rather than how to implement something. Then I point you to the Data Science StackExchange.
If you want to implement your recommendation engine for eCommerce I highly recommend to use GraphAware Reco. Which is Framework for creating Recommendation Engines top of Neo4j.
Here is the scaffold for basic Recommendation Engine based on GraphAware Reco - https://github.com/graphaware/recommendations-meetup
If your application is based on PHP you can use GraphAware reco4php
Upvotes: 1