A content evaluation of the proceedings of the 4th all Africa conference on animal agriculture(AACAA) an unsupervised leximancer™ analysis

A CONTENT EVALUATION OF THE PROCEEDINGS OF THE 4th ALL AFRICA CONFERENCE ON ANIMAL AGRICULTURE (AACAA):

AN UNSUPERVISED LEXIMANCER™ ANALYSIS

5th AACAA, Addis Ababa: Ethiopia, 25-28 October 2010

Percy Madzivhandila and Garry Griffith

Presentation Outline

• Introduction• Gaps and an Opportunity• Aim• Research Questions• Data and Method• Results• Brief Discussion• Conclusions• Study Limitations

Introduction

• AACAA allow researchers to further develop and showcase their work

• After the conference is over, very little is known about lessons learned at the aggregate level

• Available information management and knowledge generation technologies are not used in this context – i.e., Animal Agriculture & conference proceedings

Gaps and an Opportunity

• Generation of knowledge [from the proceedings] can be daunting and it is often difficult to synthesize and validate (Hearst 1999, 2003; Smith and Humphreys 2006): – Magnitude and diversity of contributed papers– The limitation of a person’s reading speed ability– Lack the resources (or knowledge about such tools) to analyse

large text data objectively – Problem with human subjectivity when analyzing text data

• Therefore, unsupervised text data mining methods are entering knowledge generation lexicon (Lin and Richi 2007; Smith 2000, 2003)

Aim

• To identify and analyse recurring issues raised to address the 4th AACAA theme

– The theme was: The role of biotechnology in animal agriculture to address poverty in Africa: Opportunities and challenges.

Research Questions

• What messages emerged (i.e., lessons) from the 4th AACAA conference proceedings?

• What is the potential of Leximancer™ in establishing credible conference messages and/or lessons?

Data

• The entire proceedings from the 4th AACAA (Arusha, Tanzania)– Edited by Rege et al. (2006)

Method• Approach: Unsupervised text data mining technique

– No predetermined categories were imposed on the data through coding

• Computer software used: Leximancer™ (see www.leximancer.com)

1. Editing of concepts– It is possible within the software to delete, combine or add new

concepts, However:• The decision was made to retain only those concepts identified by

the software. • But the concepts that are similar semantically (i.e., in singular and

plural forms) were merged

2. Data analysis– Concept maps were augmented by using a three slide bars

embedded in the software to adjust theme size, concept points size and rotation

Results

The entire analysis allowed us to illustrate five types of information :

• The central theme(s)• The main concepts (set at 20% concept size)• The thematic group of concepts that

demonstrate similarity (i.e., clusters)• Frequency within which these concepts occur

(i.e., ranking)• Frequency of co-occurring concepts

Results 1: Underpinning Themes

Results 2: Main Concepts

Results 3: Thematic & Concepts Map

Results 4: Ranked Concepts#

Concept Absolute

Count Relative Count

1 animals 771 100% 2 livestock 573 74.3% 3 production 516 66.9% 4 research 452 58.6% 5 countries 364 47.2% 6 biotechnology 350 45.3% 7 development 346 44.8% 8 milk 339 43.9% 9 genetic 330 42.8% 10 level 315 40.8% 11 systems 305 39.5% 12 breeds 295 38.2% 13 Africa 280 36.3% 14 cattle 274 35.5% 15 products 269 34.8% 16 farmers 264 34.2% 17 Proceedings 255 33% 18 disease 238 30.8% 19 health 229 29.7% 20 goat 228 29.5% 21 information 215 27.8% 22 food 214 27.7% 23 study 205 26.5% 24 African 198 25.6% 25 resources 191 24.7%

Results 5: Co-occurrence of ConceptsAnimal(s)

Concept Absolute Count Relative Count 1 livestock 218 28.2% 2 production 178 23% 3 genetic 149 19.3% 4 health 147 19% 5 research 130 16.8% 6 biotechnology 128 16.6% 7 disease 109 14.1% 8 countries 101 13% 9 development 96 12.4% 10 products 93 12% 11 resources 91 11.8% 12 breeds 85 11% 13 systems 84 10.8% 14 food 76 9.8% 15 farmers 75 9.7% 16 cattle 75 9.7% 17 Africa 74 9.5% 18 milk 70 9% 19 potential 69 8.9% 20 indigenous 66 8.5% 21 level 66 8.5% 22 breeding 63 8.1% 23 agriculture 57 7.3% 24 human 56 7.2% 25 National 54 7.0%

Results 5: Co-occurrence of ConceptsMilk

Concept Absolute Count Relative Count 1 production 103 30.3% 2 goat 86 25.3% 3 animals 70 20.6% 4 dairy 59 17.4% 5 meat 56 16.5% 6 protein 52 15.3% 7 level 51 15% 8 products 47 13.8% 9 cows 43 12.6% 10 cattle 39 11.5% 11 study 37 10.9% 12 livestock 36 10.6% 13 systems 35 10.3% 14 quality 34 10% 15 farmers 32 9.4% 16 high 30 8.8% 17 breeds 29 8.5% 18 Africa 29 8.5% 19 control 25 7.3% 20 genetic 23 6.7% 21 human 22 6.4% 22 countries 21 6.1% 23 health 20 5.8% 24 small 20 5.8% 25 food 20 5.8%

Concept Absolute

Count Relative Count

1 animals 128 36.5% 2 research 122 34.8% 3 livestock 110 31.4% 4 countries 82 23.4% 5 development 77 22% 6 production 68 19.4% 7 products 67 19.1% 8 application 57 16.2% 9 health 56 16% 10 Africa 51 14.5% 11 food 47 13.4% 12 national 45 12.8% 13 agricultural 42 12% 14 genetic 42 12% 15 poor 41 11.7% 16 agriculture 41 11.7% 17 potential 40 11.4% 18 developing 40 11.4% 19 resources 39 11.1% 20 poverty 38 10.8% 20 science 36 10.2% 21 human 35 10% 22 issues 32 9.1% 23 international 31 8.8% 24 African 29 8.2% 25 risk 29 8.2%

Results 5: Co-occurrence of ConceptsBiotechnology

Discussion• Leximancer™ Version 2.25 (2005) was able to deal with large

amounts of unstructured text data – It extracted previously unknown, useful and important

concepts and categorized the information that might be missed by manual methods

• It easily enabled us to provide information about the results of the content analysis in a number of ways– E.g., superimposing of thematic circles enriched the viewing of

the initial analytic description

• It enabled us to discover information within the document objectively without being influenced by human biases

Conclusion

• This study results can provide a most valued learning environment for policy makers, decision makers and practitioners– A message at the aggregate level nexus

• We posit that Leximancer™ has an established role in the content analysis of large and diverse text data

Study Limitations

• Analysis does not purport to be a complete analysis of the proceedings – The results reported are by no means

exhaustive

• The study does not claim to represent the intentions of the authors of the proceedings' papers

Thank You

References• Hearst, M. (1999), 'Untangling Text Data Mining'. <

http://people.ischool.berkeley.edu/~hearst/papers/acl99/acl99-tdm.html>, accessed 18 May 2010.• --- (2003), 'What is Text Mining?'. <http://www.ischool.berkeley.edu/~hearst/textmining.html.>,

accessed 18 May 2010.• Leximancer (2005), 'Leximancer Version 2.25 Manual'. <<

http://www.leximancer.com/documents/Leximancer2_Manual.pdf.>>, accessed 12 May 2010.• Lin, C. and Richi, N. (2007), 'A Case Study of Failure Mode Analysis with Text Mining Methods',

in K.-L. Ong, W. Li; and J. Gao (eds.), 2nd International Workshop on Integrating Artificial Intelligence and Data Mining (AIDM) (Gold Coast, Queensland: CRPIT), 49-60.

• Smith, A. E. (2000), 'Machine Mapping of Document Collections: the Leximancer System', The 5th Australasian Document Computing Symposium (Sunshine Coast, Queensland: Australasian Document Computing Symposium).

• --- (2003), 'Automatic extraction of semantic networks from text using Leximancer', The North American Chapter of the Association for Computational Linguistics (HLT-NAACL) (Edmonton, ACL: HLT-NAACL), 23-24.

• Smith, A.E. and Humphreys, M.R. (2006), 'Evaluation of unsupervised semantic mapping of natural language with Leximancer concept mapping', Behaviour Research Methods, 38 (2), 262-79.

• Rege, J.E.O.; Nyamu A.M. and Sendali, G. (eds) (2005), ‘The role of biotechnology in animal agriculture to address poverty in Africa: Opportunities and challenges’, Proceedings of the 4th All Africa Conference on Animal agriculture and the 31st Annual Meeting of the Tanzania Society for Animal Production (Arusha:AACAA)