Name
Abstract | ||
|---|---|---|
The main goal for the Information Space system for TREC9 was early precision. To facilitate this, an emphasis was placed on seeking matches from only the TITLE, H1, H2 and H3 tags in the Web (wt10G) and large Web (wt100) document collections. Ranking of documents was based on a combination of Boolean union sets, term weights, and principal components analysis (PCA). Very large sparse cooccurrence matrices were created for term weighting and PCA. The Information Space system is part of a larger general software package called IRTools. The Information Space (IS) main Web task entry for this year focused only on tags in the <TITLE>, <H1>, <H2>, and <H3> tags in the datasets. This was intended to facilitate early precision, by matching the short TREC topic title or title plus description statements to terms in these tags. The submission for the main Web task was 6 days late, and therefore not judged (although it was counted by NIST as an "official" run). Post hoc analysis of some queries indicate that if results were judged, they probably would not have been substantially better than the non-judged results found in the conference proceedings. IS also made an entry to the large Web task or VLC. The 100GB VLC (w100) was processed similarly to the main Web task, by focusing only on terms in the same set of tags (title, h1, h2 and h3). Because this run was also submitted late, by nearly 2 weeks, it was not judged. Due to the small number of official VLC submissions and small number of judged documents, no useful recall or precision statistics are available. This paper will present an overview of the procedure used to index and retrieve from the wt10g and w100 datasets, followed by a brief discussion of the large co-occurrence matrices generated. Then, system-based and relevance-based performance outcomes are discussed. It is concluded that query expansion did not serve well to facilitate early high precision. Furthermore, a lack of sophisticated term weighting also hurt results. |
Year | Venue | Keywords |
|---|---|---|
2000 | TREC | query expansion,principal component analysis |
Field | DocType | Citations |
Data mining,Information retrieval,Query expansion,Computer science,Information space,Vector space model,Term Discrimination | Conference | 4 |
PageRank | References | Authors |
1.00 | 1 | 1 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Gregory B. Newby | 1 | 220 | 32.13 |