An Empirical Study of Automatic Summarisation

This research aimed to improve search engine result
summaries to help internet users make quick and
accurate relevance judgements. Term Order and Query
Term Order(QTO) algorithms were developed in order to
produce better search engine result summaries. Six
sentence weighting schemes were constructed in
different weighting components combinations with the
aim of comparing the effectiveness between QTO and
Query Term Frequency (QTF).
The literature on automatic summarisation evaluation
is classified into intrinsically motivated Gold
Standard/Baseline and Subjective Scoring evaluations,
and extrinsically motivated Task Based evaluation.
In order to triangulate evidence of the usefulness of
the QTO algorithm, Document Understanding Conference
data was used for intrinsic evaluation and
online English web data was used for both intrinsic
and extrinsic evaluations.
The QTO summarisation system was compared against
that of Google. Representativeness, Judgeability and
Scanning-Speed were the three tasks in the
evaluations, and the summary quality was derived from
the measurements of the three tasks.

Autorentext

I started my research in Automatic Summarisation with the
Information Retrieval Research Group in the University of
Sunderland. Then I was involved in Semantic Web research in the
University of Southampton. Therefore, my research interests have
been developed to explore web knowledge using Natural Language
and Semantic Web techniques.

Klappentext

This research aimed to improve search engine result
summaries to help internet users make quick and
accurate relevance judgements. Term Order and Query
Term Order(QTO) algorithms were developed in order to
produce better search engine result summaries. Six
sentence weighting schemes were constructed in
different weighting components combinations with the
aim of comparing the effectiveness between QTO and
Query Term Frequency (QTF).
The literature on automatic summarisation evaluation
is classified into intrinsically motivated Gold
Standard/Baseline and Subjective Scoring evaluations,
and extrinsically motivated Task Based evaluation.
In order to triangulate evidence of the usefulness of
the QTO algorithm, Document Understanding Conference
data was used for intrinsic evaluation and
online English web data was used for both intrinsic
and extrinsic evaluations.
The QTO summarisation system was compared against
that of Google. Representativeness, Judgeability and
Scanning-Speed were the three tasks in the
evaluations, and the summary quality was derived from
the measurements of the three tasks.