Information Technology Reference
In-Depth Information
•
In what respect are these learning-to-rank algorithms similar and in which aspects
do they differ? What are the strengths and weaknesses of each algorithm?
•
Empirically speaking, which of those many learning-to-rank algorithms performs
the best?
•
Theoretically speaking, is ranking a new machine learning problem, or can it
be simply reduced to existing machine learning problems? What are the unique
theoretical issues for ranking that should be investigated?
•
Are there many remaining issues regarding learning to rank to study in the fu-
ture?
The above questions have been brought to the attention of the information re-
trieval and machine learning communities in a variety of contexts, especially during
recent years. The aim of this topic is to answer these questions. Needless to say, the
comprehensive understanding of the task of ranking in information retrieval is the
first step to find the right answers. Therefore, we will first give a brief review of
ranking in information retrieval, and then formalize the problem of learning to rank
so as to set the stage for the upcoming detailed discussions.
1.2 Ranking in Information Retrieval
In this section, we will briefly review representative ranking models in the literature
of information retrieval, and introduce how these models are evaluated.
1.2.1 Conventional Ranking Models
In the literature of information retrieval, many ranking models have been proposed
[
2
]. They can be roughly categorized as relevance ranking models and importance
ranking models.
1.2.1.1 Relevance Ranking Models
The goal of a relevance ranking model is to produce a ranked list of documents ac-
cording to the relevance between these documents and the query. Although not nec-
essary, for ease of implementation, the relevance ranking model usually takes each
individual document as an input, and computes a score measuring the matching be-
tween the document and the query. Then all the documents are sorted in descending
order of their scores.
The early relevance ranking models retrieve documents based on the occurrences
of the query terms in the documents. Examples include the
Boolean model
[
2
]. Ba-
sically these models can predict whether a document is relevant to the query or not,
but cannot predict the degree of relevance.
