Abstract: Marrying Relevance and Genre Rankings: an Exploratory Study

Marrying Relevance and Genre Rankings: an Exploratory Study

by Pavel Braslavski

In: Genres on the Web Computational Models and Empirical Studies
Alexander Mehler, Serge Sharoff and Marina Santini
Text, Speech and Language Technology
Volume 42, 2011, DOI: 10.1007/978-90-481-9178-9

Abstract

In this chapter, we discuss different options for using genre-related information inWeb search. We conduct an experiment on merging genre-related and text-relevance rankings using a reference Web collection. A method for automatic extraction of formality score akin to readability score using canonical discriminant analysis applied to a sample of genres with decreasing formality is proposed. Effects of aggregating genre-related and text relevance rankings are considered. Evaluation of the results shows moderate positive effects. Findings suggest that further research is needed on implicit use of genre-related information in Web search.

1 Introduction
Recent years have shown a growing interest to automatic genre analysis ofWeb documents, especially in the context of Web search. As the amount of indexed documents grows, the specification of a few keywords is not enough to describe user information need. Many studies suggest looking at document genre as an additional non-topical retrieval criterion. The output of a genre classifier could be used inWeb search both explicitly and implicitly. Explicit use implies at least three possibilities. First, a focused (`vertical’) search engine (SE) over documents belonging to a certain genre could be built. Second, the user can be given an opportunity to specify the desired genre in the query. Finally, the search engine results page (SERP) can be improved by enriching snippets with genre labels 2 or grouping the documents of the same genre together. However, all three options bring up issues.
If we look at successful vertical search services such as scientific paper search3, blog search4, news search engines5, or product search and comparison services6 we notice that the task of gathering (or filtering out) content for services does not require especially sophisticated methods. Either the contributors are highly interested in providing their content to the service (scientific papers authors/publishers, on-line merchants), or the content is concentrated on several host sites in a certain form (like blog services, RSS feeds), or it can be found on the Web using simple surface features with high precision and satisfactory recall (e.g. scientific papers on authors’ homepages).

[Continue reading excerpts here or download PDF from here]

Leave a Reply

Your email address will not be published. Required fields are marked *

*