1. RaJoLink: Creative Knowledge Discovery by Literature Outlier Detection
Ingrid Petrič
University of Nova Gorica
Bojan Cestnik
Temida, Ljubljana and Jozef Stefan Institute, Ljubljana
Nada Lavrač
Jozef Stefan Institute, Ljubljana and University of Nova Gorica
Tanja Urbančič
University of Nova Gorica and Jozef Stefan Institute, Ljubljana

2. Overview Motivation: Focus domain: Knowledge discovery support tools:
to present a method that supports knowledge discovering by connecting information from different contexts in a new way
Focus domain:
biomedical research
autism, a spectrum of pervasive developmental disorders
Knowledge discovery support tools:
RaJoLink (Petrič and Cestnik, 2007)
OntoGen (Fortuna et al., 2006)
Knowledge sources:
MEDLINE (
MeSH (

3. Swanson’s model of literature-based discovery (Swanson, 1986)
Literature about Literature about
magnesium (A) migraine (C)
(Bi)

4. Closed vs. open discovery process (Weeber et al., 2001)

5. Combined open and closed discovery in RaJoLink (Petrič et al., 2009)
Open discovery (generation of hypothesis)
Identifying rare terms r
Finding joint terms a
Closed discovery (hypothesis testing)
Searching for linking terms b

6. The RaJoLink method: open discovery
Literature R3
Literature R1
Joint term A
Literature R2
Rare term R1
Rare term R2
Rare term R3
Literature C

7. The RaJoLink method: closed discovery
Literature A
Joint term A
Linking term B1
Linking term B2
Linking term B3
Literature C

8. The RaJoLink method’s procedures (Petrič et al., 2009)

9. Steps of the RaJoLink method – step Ra
Input
Action
Human involvement
Output Ra
Set of records about domain of interest (about phenomenon C)
1.1
Extraction of texts
1.2
Data preprocessing
1.3
Identification of rare terms
1.4
Terms filtering
Indication of interesting rare terms
Rare terms C_r1, C_r2,…C_rp

10. Step Ra

11. Steps of the RaJoLink method – step Jo
Input
Action
Human
involvement
Output Jo
Sets of records about C_r1, C_r2,…,
C_rp
2.1 Extraction of texts
2.2 Data preprocessing
2.3 Search for joint terms
Selection of a
significant
joint term
Joint term a

12. Step Jo

13. Steps of the RaJoLink method – step Link
Input
Action
Human
involvement
Output
Link
Joint set of records about a and articles about c
3.1 Extraction of texts
3.2 Data preprocessing
3.3 Identification of content related A and C records
3.4 Search for linking terms b
Selection of
meaningful
linking terms
Linking terms B1, B2,…Br

14. Step Link

15. Step Link - alternative

16. Conclusions Open discovery:
RaJoLink represents a more interdisciplinary approach to hypotheses generation that bridges the overspecialization in the sciences. We provide connections between biomedical literature by analysis and explanation of rare terms.
Closed discovery:
With the combination of outlier detection and high frequency analysis approach we demonstrated that outlying documents could be used as a heuristic guidance to speed-up the search for the linking terms and alleviate the burden on the expert when hypotheses have to be tested.
Recent experiments:
Detection of published evidence of autism findings that coincide with specific calcineurin and NF-kappaB observations (Petrič et al., 2007, Urbančič et al., 2007).
The gold standard evaluation: RaJoLink led to the Swanson’s relation of magnesium with migraine and to other three discoveries important for migraine.

17. Future work
Automated identification of semantic variants such as abbreviations, acronyms and synonyms.
Handling the language specifics for Slovenian and other languages.
Providing visualizations of results.
Implementing the similarity measure between documents in the Link step.

18. RaJoLink - references
Petrič, I.; Urbančič, T.; Cestnik, B. Discovering hidden knowledge from biomedical literature. Informatica 31(1):15-20 (2007).
Petrič, I.; Urbančič, T.; Cestnik, B. Literature mining: potential for gaining hidden knowledge from biomedical articles, In: Bohanec, M.; Gams, M.; Rajkovič, V.; Urbančič, T.; Bernik, M.; Mladenić, D. et al., editors. IS Proceedings of the 9th International multi-conference Information Society; Ljubljana, Slovenia (2006).
Petrič, I.; Urbančič, T.; Cestnik, B.; Macedoni-Lukšič, M. Literature mining method RaJoLink for uncovering relations between biomedical concepts. Journal of Biomedical Informatics 42(2): (2009).
Urbančič, T.; Petrič, I.; Cestnik, B.; Macedoni-Lukšič, M. Literature mining: towards better understanding of autism. In: Bellazzi R; Abu-Hanna A; Hunter J, editors. AIME Proceedings of the 11th Conference on Artificial Intelligence in Medicine in Europe; Amsterdam, The Netherlands (2007).