Abstract | ||
---|---|---|
We describe efficient methods to score structured hypotheses from threat detection technologies that fuse evidence from massive data streams to detect threat phenomena. The strongly object-oriented threat cas e representation summarizes only key object attribute s. Pairing of hypothesized and reference cases exploit s a directed acyclic case type graph to minimize case comparisons. Because case pairing is expensive, we expediently a void it where possible. One global pairing operation suffi ces to develop: • Count-based metrics (precision, recall, F-value) th at generalize the traditional versions to object-orien ted versions that accommodate inexact matching over structured hypotheses with weighted attributes; • Area under the object-oriented precision-recall cur ve; • Cost-based metrics that address timely incremental evidence processing; • Statistical significance of computed scores. Many software parameters support customized experimentation. |
Year | Venue | Keywords |
---|---|---|
2006 | AAAI Fall Symposium: Capturing and Using Patterns for Evidence Detection | statistical significance,object oriented |
Field | DocType | Citations |
Graph,Data mining,Data stream mining,Computer science,Pairing,Exploit,Software,Artificial intelligence,Recall,Machine learning,Object Attribute | Conference | 1 |
PageRank | References | Authors |
0.75 | 1 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Robert C. Schrag | 1 | 325 | 26.58 |
Masami Takikawa | 2 | 23 | 4.25 |