Towards Modeling and Analyzing Transaction Data Stored as Directed Big Graphs in NoSQL DBMS for Products Recommendation

Patterns can be mined from transaction data and used for suggesting items to customers. Customers may have habits of purchasing items in sequence or repetitively. Recommendations can also be given to customers using this pattern. Other facts, certain items are bought seasonally. The period of a season can be short. For these kinds of items, the recommendations should be given in near real time. The aims of this research are designing and implementing: (1) Directed graphs that represent the sequence of items based on the purchasing time both for individual as well as the whole customers. (2) Algorithms for updating graphs based on the newest transactions; (3) Algorithms for analyzing graphs to provide recommended items in near real time. As the graphs will grow into big graphs with complex edges while fast graph query functions are needed, the designs are implemented on the graphbased DBMS, Neo4j, that support complex graphs and fast graphs query. By using indexes in Neo4j, the experiments results using synthetic data demonstrate that the algorithms run efficiently and support near real time product recommendation.