6.1 Parametric and zone indexes

Digital documents come with metadata. Metadata includes fields (e.g. creation date, file format). Field values are part of a finite set. Each field has one parametric index.

Hence, query processing: merge(standard inverted index, parametric index)

Zones are like fields, but its contents are not part of a finite set, but free text (e.g. document title, abstracts). Numerous approaches to indexing zones:

1. Build separate inverted index for each zone
   • (dict) william.abstract -> (postings) {11, 121, 1441}
   • william.title -> {2, 4, 8, 16}
2. Encode zone in which a term occurs in the postings (favourable)
   • (dict) william -> (postings) {(2.author,2.title), 3.author, 4.title}

The second approach is favourable because it allows for weighted zone scoring.

6.1.1 Weighted Zone Scoring

Method

• Consider a pair $(q, d)$ where $q$ is some Boolean query and $d$ is some document within a collection. Let each document in the collection have $l$ zones.
• Let $z_i$ be the $i$-th zone of document $d$
• Let $g_1, \dots, g_l \in [0,1]$ such that $\sum_{i=1}^{l} g_i=1$.
• Each zone $z_i$ will have a Boolean score $s_i$ denoting a match with $q$. This match is based on some Boolean function mapping the presence of query terms in a zone to either $0$ or $1$.
• Then, the weighted zone score for the document $d$ is $\sum^{l}_{i=1}{g_i s_i}$

Example Implementation

ZONESCORE(q1, q2)
	float scores[N] = [0]  // array with score entry for each doc, initialised to 0
	constant g[l]  // contains the respective zone weights
	p1 <- postings(q1)
	p2 <- postings(q2)
	// p1 and p2 are initialised to point to the beginning of their postings	
	
	while p1 != NIL and p2 != NIL
	do if docID(p1) = docID(p2)
		then scores[docID(p1)] <- WEIGHTEDZONE(p1, p2, g)
			p1 <- next(p1)
			p2 <- next(p2)
		else if docID(p1) < docID(p2)
			then p1 <- next(p1)
			else p2 <- next(p2)
	return scores

In this implementation, the query is a two-term query with query terms q1 and q2, and the Boolean function is AND: 1 if q1 and q2 are in a zone, and 0 otherwise.

WEIGHTEDZONE function is assumed to calculate the weighted zone score based on postings and weight.