Docket #: S08-085
Random Alpha Pagerank (RAPr)
Stanford engineers have developed a patented algorithm that improves search results from ranking the objects of a database when viewed as a graph (e.g. a web graph). This system, Random Alpha Pagerank (RAPr), computes the importance of pages in a web graph using a random variable as a teleportation coefficient (compared to the standard PageRank algorithm which assumes a constant coefficient). This approach uncovers new characteristics that can be used to provide more relevant results for web searches, web spam detection, or gene/protein classification. It can also be used to derive more meaningful measures of importance by incorporating user behavior or domain specific knowledge.
Stage of Research
The inventors have completed a study that shows spam ranking with RAPr has a meaningful improvement in the performance of web-spam detection. They have also demonstrated that the model is valid for measured user behavior on the web.
Applications
- Web searches - may reveal patterns in user behavior, which may suggest advertising strategies
- Web spam detection
- Gene/protein classification - may be useful in identifying genes that are sensitive to perturbations in the Markov model
Advantages
- More relevant results - this model provides new features to uncover characteristics of a web graph that can be used to provide better search results
- More capable modeling - the model is flexible and lets you incorporate either user behavior or “anti”-user behavior to investigate items that users will probably see or users will probably not see.
Publications
- Random Alpha PageRank Internet Math. Volume 6, Number 2 (2009), 189-236.
Patents
- Published Application: 20090276389
- Published Application: 20150220534
- Issued: 8,972,329 (USA)
Similar Technologies
-
WiDeo: A Motion Tracing Camera using WiFi Signals S14-059WiDeo: A Motion Tracing Camera using WiFi Signals
-
Improved Anomaly Detection Using Adversarially Learned Inference S19-208Improved Anomaly Detection Using Adversarially Learned Inference
-
Reconfiguration of Tabular Data for Discovery of Deep Interaction Features and its Applications in Analysis of Multidimensional Data S22-041Reconfiguration of Tabular Data for Discovery of Deep Interaction Features and its Applications in Analysis of Multidimensional Data