The document discusses how to build an automatic title generator by investigating title structure and common patterns in titles. It analyzes over 10,000 computer science article titles from arXiv.org to determine average length, common sentence structures, nouns, adjectives and punctuation. The analysis found the average title length is 8.91 words and common structures include noun phrase verb noun phrase or noun phrase preposition noun. Frequent nouns included image, evaluation, study and algorithms. Questions in titles lead to more downloads but less citations.

1. Killer Titles
Using Entitlement: A tool for promoting discussion around the
relationships between a research paper's content, citations,
abstract and title
Dr Viccy Adams - Leverhulme Trust Writer in Residence at the School of
Informatics, Edinburgh University
Clare Llewellyn - PhD Student, School of Informatics, Edinburgh University

2. Why is a title important?
• Signpost
• Hook
• Name

5. How to Build an
Automatic Title
Generator?
• Investigate general title structure
• Investigate titles in the Computer Science domain
• Build a rule based text generator (or borrow one)
• Use a grading system to score title produced

6. Title Investigation
3 main types of title:
Declarative
"Adipose gene expression prior to wait loss can differentiate
and weakly predict dietary response"
Descriptive / neutral
"A worldwide phylogeography for the human X chromosome"
Interrogative
"Is there rearrangement hotspots for the human genome?"
Most common type of title is
descriptive

7. General Facts:
• Titles which are questions lead to more downloads but
less citations of the papers
• Longer titles generally mean less downloads (slightly)
• Titles with colons tend to be longer which leads to fewer
downloads and citations*
o (but the number of words and the use of colons
increased 1981-2001)
• The use of a colon and acronym lead to more citations
• Including a country name means less citations
• Humour means less citations, but more enjoyment -
because of a lower perception of credibility
• Pleasantness means more citations
• Google gives higher weighting to title words but not
keywords

8. Investigation: Computer
Science
Based on 10, 000 article titles in the Computer Science Category from
arXiv.org
Statistical and part of speech analysis to this text
Average length of title = 8.91 words

9. Common Sentence Structure
Noun Phrase Verb Noun Phrase
Noun Phrase Preposition Noun
Determiner Noun Preposition Noun Phrase
Verb Noun Phrase Preposition Noun Phrase
Adjective Noun Preposition Noun Phrase
Common Nouns
Image, evaluation, study, time, search, logic, network, design, method,
algorithm
Common Adjectives
Parallel, intelligent, global, large, polynomial, new, complex, local,
distributed
Common Punctuation
! 10
? 139
. 9955

16. Dr Viccy Adams: vsadams@gmail.com
Clare Llewellyn: C.A.llewellyn@sms.ed.ac.uk