Recommendation / Personalization
Technological Specification
Improvement of Sockets recommendation engine
Recommendation was once analogical technology which depends on recommender’s knowledge. Sockets have long time challenged to automate recommendation by big data analyzing and statistic algorithm. These days our recommendation engine has improved into hybrid entertainment engine which is based on original emotional metadata and collaborating filtering recommender engine.
General recommendation environment “emerge”
Sockets original general recommendation environment “emerge” is the whole compiled knowledge of recommendation from our long time experience.
It is able to mixture emotional metadata analyzing, action analysis from service log, uniqueness rule-based, which cannot be derived from contents based collaborative filtering, especially in entertainment. It also improves by machine learning.
Emerge has four recommend style by analyzing and classifies by “what information is needed for recommendation”, “what is the recommendation rule”, and each perspective is broke down into by what attribute is the recommended item determined, and finally mixed to fit the service feature. Especially Bayesian network probabilistic inference algorithm is an effective method to analyze the environment, time and means of the contents.
Original recommendation engine which human and machine are well-balanced
Idealistic recommendation result cannot be realized by big data (log) analysis. Metadata which enables human sense machine readable, careers of entertainment services log analysis, and most important is creator and experts knowledge. Well-balanced of automation and human experiences is able to offer the best recommendation.
Features of Sockets recommendation engine
1.Man-powered emotional metadata, ten years performance of music/video major service operation
Efficiency and improvement through static data/ training data which has domestic operation performance
2.Recommendation engine which is highly customizable and flexible/h4>
Highly customizable in the process of source code layer, which is essential due to complicated structure compared to Western countries.
3.Emotional metadata which suites Japanese sense
Especially focused on Japanese data interpretation by expert team.
4.Mixture design which is well-balanced compounded of multiple recommendation method
We can deal with light user, deep user cold start by selecting appropriate recommendation module for service requirement and user preference.
5.Cross-data collecting
Data collecting and organization is available by general architect, through possessing music, video, books and others. We realize seamless and overall business.
