Name
Abstract | ||
|---|---|---|
The rationale and methodology for r etrieval based on the relative locations of documents within a geometric information space are introduced. Results from category A routing and filtering experiments in TREC-5 are discussed. The techniques used are related to the vector space model, latent semantic indexing, and other methods that rely on statistical qualities of texts to assess document relatedness. Results show some promise, but additional research is needed to determine the extent to which retrieval may be improved over existing approaches. I NTRODUCTION Sp atial models for IR are well-known, with almost 30 years of refinement and variation. The primary points of departure of this work from many of the efforts utilizing the vector space model (VSM) or derivatives are: 1. Relations among terms are measured. This is contrary to te basic VS M, in which term vectors are mutually unrelated (orthogonal); and 2. Only a relatively small subset of available terms are utilized. |
Year | Venue | Keywords |
|---|---|---|
1996 | TREC | vector space model,latent semantic indexing |
Field | DocType | Citations |
Data mining,Latent semantic indexing,Vector space,Information retrieval,Computer science,Filter (signal processing),Information space,Vector space model,Correlation analysis | Conference | 1 |
PageRank | References | Authors |
1.16 | 4 | 1 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Gregory B. Newby | 1 | 220 | 32.13 |