
Gustos Guide
Our collaborative filtering, recommendation engine is designed to offer flexibility both in implementation and technology requirements. The full power of the product can be used by any site for free with our StarRater Java applet or installed on-site with custom user interfaces. The product is divided into two primary, scaleable components: the Prediction Server and the Relation Server.
Gustos Prediction Server
For sites that wish to use standard CGI form interfaces or develop their own Java user interface, the Gustos Prediction Server allows full customization and integration with existing web sites. The Prediction Server includes the Gustos Recommendation Engine, Ratings Server, Web Server and an easy to use, Application Programming Interface (API). A number of tools are also provided to make integration with existing applications fast and easy. Typically, a single Pentium class PC is dedicated to the Gustos Prediction Server process; however, faster machines or multiple servers may be installed as needed to meet desired performance levels.
Our Prediction Servers work in conjunction with on-site or central Gustos Relation Servers. Advantages of using central servers include reduced cost and increased accuracy. The Prediction Server asynchronously communicates with the Relation Server to transfer user ratings and to get updates of the Prediction Server relation cache. All other functions are performed locally in real-time so performance can be guaranteed and scaled up as needed.
Accuracy is increased by centralizing the data since more information is known about each user. And with more ratings, the resultant relation cache is more accurate so recommendations are better for users. The Web site further benefits from the ability to display user ratings logged from other web sites. Finally, the need to 'build' a database of ratings prior to getting accurate results is largely eliminated since this data already exists in our central servers.
Gustos Relation Server
Gustos Software also offers complete private systems for sites that require local storage and processing of collaborative data or have other special needs. The purpose of the Relation Server is to find statistically significant relationships in the master database.
The Relation Server examines the data on a continuous basis and produces an optimized relation cache which is used by the Prediction Servers to service user requests for recommendations and predictions. In addition, the Releation Server keeps a master datanase of user ratings and item records that are logged by individual Prediction Servers.
The database structure is open and JDBC compliant.
Platform Support
The products are written in Java and are available for any system with a Java JDK. Currently, the products have been tested under the following operating systems:
- Sun Solaris 2.5
- Windows NT
- Linux
For questions or more information, send mail to: info@gustos.com.