Analysing a modified ranking algorithm for exploratory search
Ladda ner
Typ
Examensarbete för masterexamen
Program
Engineering mathematics and computational science (MPENM), MSc
Publicerad
2020
Författare
Fällman, Markus
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
Exploratory Search is a small emerging field within Information Retrieval, studying a
type of searching called exploratory searching. This type of search is directed towards
learning and investigating, and has recently started to draw attention. However, the field
of Exploratory Search struggles with its methodology. A central problem is the difficulty
to measure improvements due to that exploratory searching by definition lacks precise
goals. New tools and ideas are therefore often evaluated with user studies. By focusing
on describing how tools and ideas work, researchers can avoid the difficulty and contribute
to the field. Such an indirect approach allows formulating measures that can be applied
to ranked lists, which, in turn, allow using simulations with many benefits. This study
showcases the approach.
The aim is to determine if a ranking algorithm modification influence the formation
of groups in lists of ranked articles returned from an academic search engine. The data are
generated by simulated searches and a Linear Mixed Model is used for the analysis. The
main covariates represent how the ranking of a standard ranking algorithm is weighted
together with the ranking according to two new criteria. The response variable consists
of scores on how tightly connected the ranked articles are, with the importance of links
decreasing with the depth, and comes from the application of a measure developed in the
thesis.
The main result is that the level of interconnectedness between high ranking articles
can be clearly and statistically significantly influenced by the modification, although the
influence varies with the randomly generated queries. While more research is needed, this
might be useful for controlling the articles interconnectedness when constructing a search
engine. On a different level, the thesis shows how the indirect approach can be applied,
that it enables using simulations, and it indicates that the approach can produce results
interesting for exploratory searching.
Beskrivning
Ämne/nyckelord
exploratory search, ranking algorithm, rank biased measure, citation expansion, linear mixed model.