Method of locating related items in a geometric space for data mining
A method for locating related items in a geometric space transforms relationships among items to geometric locations. The method locates items in the geometric space so that the distance between items corresponds to the degree of relatedness. The method facilitates communication of the structure of the relationships among the items. The method is especially beneficial for communicating databases with many items, and with non-regular relationship patterns. Examples of such databases include databases containing items such as scientific papers or patents, related by citations or keywords. A computer system adapted for practice of the present invention can include a processor, a storage subsystem, a display device, and computer software to direct the location and display of the entities. The method comprises assigning numeric values as a measure of similarity between each pairing of items. A matrix is constructed, based on the numeric values. The eigenvectors and eigenvalues of the matrix are determined. Each item is located in the geometric space at coordinates determined from the eigenvectors and eigenvalues. Proper construction of the matrix and proper determination of coordinates from eigenvectors can ensure that distance between items in the geometric space is representative of the numeric value measure of the items' similarity. 12 figs.
- Sponsoring Organization:
- USDOE; USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC04-94AL85000
- Assignee:
- Sandia Corp., Albuquerque, NM (United States)
- Patent Number(s):
- US 5930784; A
- Application Number:
- PPN: US 8-918701
- OSTI ID:
- 6163778
- Resource Relation:
- Patent File Date: 21 Aug 1997
- Country of Publication:
- United States
- Language:
- English
Similar Records
Method using a density field for locating related items for data mining
ORNL_AISD-Ex: Quantum chemical prediction of UV/Vis absorption spectra for over 10 million organic molecules