Information Technology Reference
In-Depth Information
Chapter 14
Applications of Learning to Rank
Abstract
In this chapter, we introduce some applications of learning to rank. The
major purpose is to demonstrate how to use an existing learning-to-rank algorithm
to solve a real ranking problem. In particular, we will take question answering, mul-
timedia retrieval, text summarization, online advertising, etc. as examples, for illus-
tration. One will see from these examples that the key step is to extract effective
features for the objects to be ranked by considering the unique properties of the ap-
plication, and to prepare a set of training data. Then it becomes straightforward to
train a ranking model from the data and use it for ranking new objects.
14.1 Overview
Up to this chapter, we have mainly used document retrieval as an example to in-
troduce different aspects of learning to rank. As mentioned in the beginning of the
topic, learning-to-rank technologies have also been used in several other applica-
tions. In this chapter, we will introduce some of these applications.
Basically, in order to use learning-to-rank technologies in an application, one
needs to proceed as follows. The very first step is to construct a training set. The
second step is to extract effective features to represent the objects to be ranked, and
the third step is to select one of the existing learning-to-rank methods, or to develop
a new learning-to-rank method, to learn the ranking model from the training data.
After that, this model will be used to rank unseen objects in the test phase.
In the remainder of this chapter, we will show how learning-to-rank technolo-
gies have been successfully applied in question answering [
1
,
14
-
16
], multimedia
retrieval [
17
,
18
], text summarization [
10
], and online advertising [
5
,
9
]. Please note
that this is by no means a comprehensive list of the applications of learning to rank.
14.2 Question Answering
Question answering (QA) is an important problem in information retrieval, which
differs from document retrieval. The task of question answering is to automatically
answer a question posed in natural language. Due to this difference, QA is regarded
