Supporting Text Retrieval by Typographical Term Weighting - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 206

Table 5. Comparison of author guidelines of GI and ACM

GI Guidelines ACM Guidelines

Title:
Helvetica, 16
p

Title:
Times, 14 p,
bold

Abstract:
Times, 9 p

Abstract:
Times, 8 pt

Section:
Times, 12 p,
bold

SECTION:
Helvetica, 10 p, up-
perc.

Subsection:
Times, 10 p, bold

Subsection:
Helvetica, 10 p

Continuous Text:
Times, 10 p

Continuous Text:
Times, 10 p

Footnotes:
Times, 8 p

Footnotes:
Times, 8 p

References:
Times, 9 p

References:
Times, 8 p

1. The absolute approach to typographical weighting was already introduced in Section 3 using
HTML tags as an example. Similar to Table 4, a weighting table can also be set up for the general
typographical weighting of text passages in other document formats like PDF. Whenever a com-
mon style guideline exists, this approach has the advantage that typographical weighting can be
adapted to the common style guideline.
2. Global relative approaches to typographical weighting of documents were introduced in the sec-
ond section, again using HTML documents as an example. All of these approaches evaluate each
combination of styles based on quality improvement with respect to a weighing function. They are
applicable only if the evaluated documents are based on a common style guideline. Nevertheless,
they are not limited to HTML, but can be used for PDF documents and other formats as well.
3. The document-relative approach evaluates each style combination occurring in a document based
on frequency. Less frequent style combinations get stronger weights than more frequent style com-
binations. This is motivated by the assumption that the most frequently used style combination,
i.e., normally the continuous text, has the smallest weight. This approach is appropriate mainly for
heterogeneous document collections, but in some cases leads to an overestimation of text passages
like footnotes.
4. The mixed approach is based on a combination of the document-relative approach with the ob-
servation that character size plays an important role in the weighting of text passages. We favor
the mixed approach in the following, since it appears to be most appropriate for heterogeneous
document collections which typically can be found in document management systems.