This paper explores how to predict query difficulty for contextual image retrieval. We reformulate the problem as the task of predicting how difficult to represent a query as images. We propose to use machine learning algorithms to learn the query difficulty prediction models based on the characteristics of the query words as well as the query context. More specifically, we focus on noun word/phrase queries and propose four features based on several assumptions. We created an evaluation data set by hand and compare several machine learning algorithms on the prediction task. Our preliminary experimental results show the effectiveness of our proposed features and the stable performance using different classification models.
[1]
George A. Miller,et al.
Using a Semantic Concordance for Sense Identification
,
1994,
HLT.
[2]
Russell N. Carney,et al.
Pictorial Illustrations Still Improve Students' Learning from Text
,
2002
.
[3]
George A. Miller,et al.
WordNet: A Lexical Database for English
,
1995,
HLT.
[4]
Hwee Tou Ng,et al.
An Empirical Evaluation of Knowledge Sources and Learning Algorithms for Word Sense Disambiguation
,
2002,
EMNLP.
[5]
Christopher D. Manning,et al.
Enriching the Knowledge Sources Used in a Maximum Entropy Part-of-Speech Tagger
,
2000,
EMNLP.