Unpacking Altmetric Donuts: Content Analysis of Tweets to Scholarly Journal Articles
1. Unpacking Altmetric Donuts:
Shinji Mine (Mie University,Japan)
Content Analysis of Tweets
to Scholarly Journal Articles
The 3rd Altmetric Conference, 28-29 September 2016
2. Introduction
• The widespread popularity of altmetrics in scholarly
communication
• the meaning of altmetric score remains unclear.
• The necessity of contextual and content analysis for
altmetrics (Priem 2014, Bornmann 2016)
• Although previous qualitative studies focused on some
aspects of twitter (Sugimoto et. al, submitted), no attempt
has been made to deal with both tweet’s content and
tweeter’s profile in a study.
• This study analyzes both content of tweets to articles and
tweeter’s profile to investigate the meaning of altmetrics
(Twitter).
3. Research Questions
1.Who tweets about scholarly journal articles?
2.What exactly are they tweeting about?
3.In terms of content and profile, does the number of
altmetrics score make a difference, especially between
Altmetric Top 100 articles and random sample articles?
4. Method: Process of this study
Altmetric
2014 Top 100
Article, Review, Letter
in WoS 2014
Altmetric API
Article, Review, Letter
with DOI
Twitter API
Article, Review, Letter
with twitter post
Random Sampled
Article, Review, Letter
Content Analysis
105,087
10,000 1,082
1,082
358,373
1,656,832
560,663
1,279,503
100
No. of docs
100
Stratified sampling
5. Content Coding
Facet 1: Bibliographic Information
1.Read descriptions in each tweet
2.Add 0/1 value
Author Name,Article Title,Journal Title, Other Source Title, URL
Facet 2: Excerption
Abstract, Body Text (BT), Figure, Other Source BT
Facet 3: Sharing/Introduction/Presentation
Pointer, Summary, Related Information
Facet 4: Comment
Positive, Negative, Neutral, Critiques, Question,As-a-Authority
Facet 5: Others
Keywords, Others
Profile Coding
Facet 6: Tweeter’s Profile 1.Read descriptions in each tweet
2.Google Search
3.Add 0/1 value
Academics, Non Academics
Tweet Type Coding
Original Tweet, Retweet
Add 0/1 value
* Twitter API data
6. Tweeter’s Profile & Tweet Type
• In both groups, most of tweet were by non academics and retweet.
• In random articles, there were more academic tweeters & original
tweets than top 100 articles.
Top 100
Random 100
0% 25% 50% 75% 100%
69.9%
84.7%
30.1%
15.3%
Academics
(n=10,000)
(n=1,082)
Non Academics
Top 100
Random 100
0% 25% 50% 75% 100%
52.1%
64.5%
47.9%
35.6%
Original Retweet(n=10,000)
(n=1,082)
7. Academics
Non Academics
0% 25% 50% 75% 100%
65.3%
58.2%
34.7%
41.8%
Academics
Non Academics
0% 25% 50% 75% 100%
51.7%
52.4%
48.3%
47.6%
Tweet Type by Profile
• In top 100 articles, tweet by academics included more original tweets.
• Random 100 articles showed similar trends between academics and no
academics.
Original Tweet Retweet
Random 100
Top 100
(n=1,605)
(n=8,395)
(n=323)
(n=758)
8. 0% 10% 20% 30% 40% 50% 60%
Author
Article Title
Journal Title
Other Title
URL
Abstract
Body Text(BT)
Figure
Other BT
Other Title
Pointer
Summary
Related Information
Positive
Negative
Neutral
Critiques
Question
As-a-Authority
Keywords
Others
Author
Article Title
Journal Title
Other Title
URL
Abstract
Body Text(BT)
Figure
Other BT
Other Title
Top 100
Random 100
(n=10,000)
(n=1,082)
Random 100 ’s original tweet
Top 100 ’s original tweet
Retweet
•Original tweets in top 100 articles tended to
include more comments and “critiques” tweets.
Possible reasons are that these articles attracted
much wider attentions in-and-outside
academia and are more controversial.
•Random 100 articles’ original tweets tended
to be “article’s title”, “short expression of
article”, or “keywords”.
These articles may be less controversial than
top 100 articles.
Proportion of contents between Top and Random 100
9. 0%10%20%30%40%50%60% 0% 10% 20% 30% 40% 50% 60%
Author
Article Title
Journal Title
Other Title
URL
Abstract
Body Text(BT)
Figure
Other BT
Other Title
Pointer
Summary
Related Information
Positive
Negative
Neutral
Critiques
Question
As-a-Authority
Keywords
Others
Random 100Top 100
Academics
Non Academics
(n=1,605)
(n=8,395)
(n=323)
(n=758)
Non academic’s original tweet
Academics ’s original tweet
Author
Article Title
Journal Title
Other Title
URL
Abstract
Body Text(BT)
Figure
Other BT
Other Title
Pointer
Summary
RandoTop 100 Academics
Non Academics
(n=1,605)
(n=8,395)
(n=323)
(n=758)
A
Article
Journa
Othe
Retweet
•Academics: original tweets on “pointer”, “summary”,
“positive“, “critiques” than non academics.
•Non academics: more “article title”, “journal title”,
“abstracts” and “keywords” tweet.
•Most of all “summary” by non academics were
original tweet.
These articles may be more “esoteric” and failed to
get social attentions, the publishers of the articles
seemed to tweet on articles in their own journals.
•Academics: more “bib
info” and “critiques”.
More excerpted from
“body text” but most of
all were retweet.
•Non academics: Less
original tweets in most of
categories. But the
differences were not so
large.
Relationship between Tweeter’s profile and Tweet’s contents
10. Conclusion
•Content of tweets became more diverse than that of previous
study (Thelwall et. al, 2013).
•Two groups with different altmetrics profile were not identical
in terms of 1) type and content of tweet and 2) tweeter’s
profile. However, the effect of removing retweets is not
negligible.
•In most of categories, retweets accounted for all tweets
regardless of tweeter’s profile. If engagement and impact
matters, position of retweets in calculating the score of
altmetrics should be carefully considered.
11. Acknowledgements
•This work was supported by JSPS KAKENHI Grant Numbers
26330364(PI:Shinji Mine, Mie University),
26280121(PI:Keiko Kurata, Keio University).
• I would like to thank
Mamiko Matsubayashi at University of Tsukuba,
Catherine Williams & Amy Rees at Altmetric.com
for their kind help and data provision.