3. What is a Web 2.0?
O’reilly, T. (2005). What is web 2.0.
Briefintroduction
● A huge collection of applications
from Internet and the World Wide
Web.
● Blogs, wikis, video sharing services,
and social media websites such as
Facebook, Youtube, Twitter,
Instagram and MySpace, which
focus on collaboration rather than
simple content delivery.
● It was introduced by the O'Reilly
Media Web 2.0 conference in 2004.
3
4. What is a Web 2.0?
Briefintroduction
● People are the center of
the web 2.0
4
6. What is a Web 2.0?
Briefintroduction
Long tail problem on Web 2.0
Bebensee, T., Helms, R., & Spruit, M. (2012). Exploring Web 2.0 applications as a mean of bolstering up knowledge
management. Leading issues in social knowledge management, 1, 22.
6
7. What is a Web 2.0?
Briefintroduction
Top impact factor
Decreasing impact
Time
Impact
Long tail problem on Web 2.0
7
8. What is a Web 2.0?
Briefintroduction
Increase impact: AI methods
Time
Impact
Long tail problem on Web 2.0
8
11. What is a Web 3.0?
Briefintroduction
The traditional web - A web of documents11
12. What is a Web 3.0?
Briefintroduction
Mahoney, L. M. and Tang, T. (2016) The Future of Social Media, in Strategic Social Media: From Marketing to Social
Change, John Wiley & Sons, Inc., Hoboken, NJ, USA. doi: 10.1002/9781119370680.ch20
The semantic web - A web of human and machine
readable content employing linked data12
13. What is a Web 3.0?
Briefintroduction
RDF
Example FOAF example
<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-
ns#"
xmlns:foaf="http://xmlns.com/foaf/0.1/">
<foaf:Person>
<foaf:name>Joe Lambda</foaf:name>
<foaf:mbox rdf:resource="mailto:joe@example.org"/>
</foaf:Person>
</rdf:RDF>
13
14. What is eHealth?
Briefintroduction
“eHealth is an emerging field in the intersection of medical
informatics, public health and business, referring to health
services and information delivered or enhanced through the
Internet and related technologies. … a state-of-mind, a way
of thinking, an attitude, and a commitment for networked,
global thinking, to improve healthcare locally, regionally,
and worldwide by using information and communication
technology.”
Eysenbach, G. (2001). What is e-health? Journal of Medical Internet Research, 3(2), e20.
http://doi.org/10.2196/jmir.3.2.e20
14
16. What is a recommender system?
Briefintroduction
● Serve the “right” item to users.
● Recommender systems reduce information overload by
estimating relevance.
Jannach, D., Zanker, M., Felfernig, A., & Friedrich, G. (2010). An introduction to recommender systems. In 25th
ACM Symposium on Applied Computing (p. 139).
16
17. What is a recommender system?
Briefintroduction
● Information Overload.
● User Experience.
● Revenues.
Recommender systems help in addressing the information
overload problem by retrieving the information desired by the
user based on his or similar users' preferences and interests.
17
18. What is a recommender system?
Briefintroduction
18
20. Background
Introduction on Recommender Systems
Fox S, Jones S. The social life of health information. Washington, DC: Pew Internet & American Life Project.
2009;2009-12
● Information and communication technologies (ICT) provide
new ways of searching and gathering health information.
Health consumers have access to a vast amount of
different kinds of resources which are disseminated
through the Word Wide Web.
20
21. Background
Introduction on Recommender Systems
Fernandez-Luque L, Karlsen R, Melton GB. HealthTrust: A Social Network Approach for Retrieving Online Health
Videos. Journal of Medical Internet Research. 31 de enero de 2012;14(1):e22.
● People are demanding accurate and trustworthy health
information. Search engines are in charge of this task and
many studies analyze how these search engines determine
health information to be trustworthy.
21
22. Background
Introduction on Recommender Systems
Fernandez-Luque L, Karlsen R, Vognild LK. Challenges and Opportunities of Using Recommender Systems for
Personalized Health Education. MIE. 2009. p. 903-7.
● What users demand is trusted information selected
according to their user profiles.
● “Health recommenders” find trustworthy health information
and adapt it to the user profile.
● It can be obtained from their Personal Health Record, in a
process that contributes the empowerment of the patient.
22
23. Background
Introduction on Recommender Systems
● Recommender methods depend mainly on
○ Domain.
○ Items to search from health information.
○ User profile.
○ Context.
23
28. Background
Basics methods on recommender systems
Sanchez-Bocanegra, C. L., Sanchez-Laguna, F., & Sevillano, J. L. (2015). Introduction on health recommender
systems. Data Mining in Clinical Medicine, 131-146.
● Collaborative approach
Users Correlation
28
29. Background
Basics methods on recommender systems
● Collaborative approach
User/Item Item 1 Item 2 ... Item N
User 1
User 2
....
a) Integer value.
b) like/dislike.
User M
Userscorrelation
29
30. ○ “If users shared the same interests in the past, they
would have similar tastes”.
■ How do users find others with similar taste?
Background
Basics methods on recommender systems
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
Recommend
items
item 1 item 2 ... item N
User 1
Collaborative
Filtering
● Collaborative approach.
30
31. Background
Basics methods on recommender systems
Adomavicius G, Tuzhilin A. Toward the next generation of recommender systems: a survey of the state-of-the
art and possible extensions. IEEE Transactions on Knowledge and Data Engineering. 2005;17(6):734–49.
○ Pearson’s correlation is used only in a few subsets of
domains. It measures the user's nearest.
○ Cosine similarity measure where users are considered
as vectors in a n-dimensional space.
○ Adjusted Cosine measure as subtract the average
rating behavior of the user.
○ Spearman's rank correlation coefficient statistical
dependence between the ranking of two variables.
○ Mean square difference measure the difference
between the estimator and what is estimated.
● Collaborative approach.
31
32. Background
Basics methods on recommender systems
○ Inverse user frequency reduces the relative importance
of those cases that receive universal agreement.
○ Significance weighting where two users may be highly
correlated based on too few items.
○ Case amplification gives more weight to highly similar
users through an amplification factor.
● Collaborative approach.
32
33. Background
Basics methods on recommender systems
○ TF-IDF Vector
■ Encode items as vectors with the dimension of the
number of terms.
■ TF: Term Frequency, how often the term appears in
the item.
■ IDF: Inverse Document Frequency, reduce the
weight of terms that appear too often.
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
● Collaborative approach
33
34. Background
Basics methods on recommender systems
○ K-nearest neighbor (kNN)
■ Similarity-based retrieval is based on two main
measures: item similarities and user likes/dislikes on
previous items.
○ The Rocchio method
■ Allows a user to rate the items (as documents), and
these ratings are incorporated into the user’s profile.
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
Suchal J, Návrat P. Full Text Search Engine as Scalable k-Nearest Neighbor Recommendation System. In: Bramer M,
editor. Artificial Intelligence in Theory and Practice III [Internet]. Springer Berlin Heidelberg; 2010.
Pazzani MJ, Billsus D. Content-Based Recommendation Systems. In: Brusilovsky P, Kobsa A, Nejdl W, editors. The
Adaptive Web. Springer Berlin Heidelberg; 2007. p. 325–41
● Collaborative approach.
34
35. Background
Basics methods on recommender systems
○ Item-based nearest neighbor filtering: Defines the
similarity between items. There is an offline processing
that reduces subsequent real time recommendations.
○ Slope One: precomputes the average difference
between the ratings of different users on one item.
● Collaborative approach.
35
36. Background
Basics methods on recommender systems
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
Lops P, Gemmis M de, Semeraro G. Content-based Recommender Systems: State of the Art and Trends. En: Ricci F,
Rokach L, Shapira B, Kantor PB, editores. Recommender Systems Handbook [Internet]. Springer US; 2011.
○ Item properties and user profiles as the essence of the
recommendation.
○ Only learned models are used to make
recommendations online, so the system needs regular
offline phases in order to improve its algorithms and
make better recommendations each time.
● Collaborative approach.
36
38. Background
Basics methods on recommender systems
● Content-based approach
Item Size Height Prize
Item 1 X Y Z
Item 2 A B C
... ... … ...
Item X X’ Y’ Z’
38
39. Background
Basics methods on recommender systems
○ Probabilistic methods
■ Approach similar of a classification task, labeling the
items according to previous user’s ratings.
○ Machine learning
■ Created to separate relevant and non-relevant
items.
■ Techniques such as clustering, decision trees,
neural networks, etc.
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
● Content-based approach
39
40. Background
Basics methods on recommender systems
Sanchez-Bocanegra, C. L., Sanchez-Laguna, F., & Sevillano, J. L. (2015). Introduction on health recommender systems.
Data Mining in Clinical Medicine, 131-146.
● Content-based approach → Knowledge based
40
41. Background
Basics methods on recommender systems
○ The user may be asked to change the requirements
○ Constraint-based
■ Recommenders search for a set of items that fulfil
the recommendation rules (or requirements)
○ Case-based
■ Try to find items similar to the user’s requirements
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
Barry Smyth, Case-Based Recommendation. In P. Brusilovsky, A. Kobsa, and W. Nejdl (Eds.): The Adaptive Web, LNCS
4321, pp. 342–376, 2007.
● Content-based approach → Knowledge based
41
42. Background
Further methods on recommender systems
○ Finds useful patterns in a transaction dataset.
○ The association rules are following the form:
■ X → Y where X and Y are two disjoint subsets of all
available items that satisfy constraints on measures
of the significance and interestingness.
■ Represented in an iteration item matrix, each
intersection cell contains a boolean value: True
means presence and False means absence.
Hipp J, Güntzer U, Nakhaeizadeh G. Algorithms for association rule mining — a general survey and comparison.
SIGKDD Explor News. 58-64.
# Transaccion Item1 Item 2 ... Item N
1 True False True
2 False True False
...
n True True False
● Matrix-factorization/latent factor models
42
43. Background
Further methods on recommender systems
○ SVD: Simple Value Decomposition - algebraically
reduces the dimension of the associated matrices.
○ LSA: Latent Semantic Analysis - discovers the latent
factors and reduces the size of these matrices by
merging semantically similar element.
○ LSI: Latent Semantic Indexing - retrieves relevant
documents even if it does not contain many words of
the user's query.
Koren Y, Bell R, Volinsky C. Matrix Factorization Techniques for Recommender Systems. Computer. 2009;42(8):30-7.
● Matrix-factorization/latent factor models
43
44. Background
Further methods on recommender systems
○ Gathering the precise meaning of one term and
determining relationships between them.
○ A term is made of words that represents an item.
○ Either terms or relations help the system to improve the
recommendations they make.
○ The Resource Description Format (RDF) is a standard
that represents information modeled as a "graph".
Hipp J, Güntzer U, Nakhaeizadeh G. Algorithms for association rule mining — a general survey and comparison.
SIGKDD Explor News. 58-64.
● Ontologies and semantic approach
44
45. Background
Hybrid methods
● Hybrid recommender systems combine two or more
recommendation approaches to gain better performance.
Most commonly, collaborative filtering is combined with
some mentioned techniques in order to minimise their
respective weaknesses.
Naime Ranjbar Kermany, Sasan H. Alizadeh, A hybrid multi-criteria recommender system using ontology and neuro-fuzzy
techniques, Electronic Commerce Research and Applications, Volume 21, 2017, Pages 50-64, ISSN 1567-4223,
http://dx.doi.org/10.1016/j.elerap.2016.12.005.
45
46. Background
Challenges on recommender systems
● Implicit and explicit ratings.
○ Localization/likes-dislikes
● Data sparsity.
○ “If two users suffer from the same disease they could
be considered similar even if they have not rated the
same item similarly”
● Cold-start.
○ New user o item?
● Serendipity/overspecialization.
○ Some recommenders rely on the similarity of items.
● Latency.
○ The system is unable to select recently added items.
Zanker M, Felfernig A, Friedrich G. Recommender Systems: An Introduction. Cambridge University Press; 2011.
46
47. Background
Health Recommender Systems
● Why?
○ More than 70% of the Internet users have searched for
health information.
Pew Internet & American Life Project. Social Media Fact Sheet. http://www.pewinternet.org/fact-sheet/social-media/, 2016
47
48. Background
Health Recommender Systems
● Who?
○ “Young adults were among the earliest social media
adopters and continue to use these sites at high levels,
but usage by older adults has increased in recent
years”
● What?
○ Health information.
○ Disease information.
○ Wellbeing!
● How often?
○ “Roughly three-quarters of Facebook users – and
around half of Instagram users – visit these sites at
least once a day”
48
50. Background
Health Recommender Systems
● All previous approaches so far can be used together with
information from Personal Health Records to provide useful
recommendations.
● A way to exploit recommendation based on the Personal
Health Record is through semantic network.
● A semantic network is a network that represents relations
between concepts as a knowledge representation.
● Represented as a graph.
50
51. Background
Health Recommender Systems
Teodoro D, Pasche E, Gobeill J, Emonet S, Ruch P, Lovis C. Building a Transnational Biosurveillance
Network Using Semantic Web Technologies: Requirements, Design, and Preliminary Evaluation.
Journal of Medical Internet Research. 2012 May 29;14(3):e73.
51
52. Background
Health Recommender Systems
○ Health professionals and patients are taking
advantage of social media, to find and share health
information. They connect with each other, and to
provide other helpful insights, but it is essential to be
aware of possibly unwanted consequences.
○ One way to avoid these effects is to take advantage of
health educational resources. Recommender systems
and personalized health education work together to
filter the overwhelming information and focus on the
actual user needs.
Roitman H, Messika Y, Tsimerman Y, Maman Y. Increasing patient safety using explanation-driven
personalized content recommendation. Proceedings of the 1st ACM International Health Informatics
Symposium. New York, NY, USA: ACM; 2010. p. 430–4. Available from:
http://doi.acm.org/10.1145/1882992.1883057.
● Ontologies and semantic approach
52
53. Background
Health Recommender Systems
○ Natural Language Processing (NLP) interacts between
computers and human (natural) languages.
○ It conforms the following analysis:
■ Syntax → Morphological segmentation, stemming
and word segmentation.
■ Semantic → Meaning of individual words, automatic
translation, synonyms and antonyms, OCR.
■ Discourse → Summarization, coreference,
sentiment analysis.
■ Speech → Speech recognition and segmentation.
Jensen PB, Jensen LJ, Brunak S. Mining electronic health records: towards better research
applications and clinical care. Nat Rev Genet. 2012;13:395–405.
Uzuner O, Stubbs A. Practical applications for natural language processing in clinical research: The
2014 i2b2/UTHealth shared tasks. J Biomed Inform. 2015;58:S1–5.
Wang Y, Luo J, Hao S, Xu H, Shin AY, Jin B, et al. NLP based congestive heart failure case finding: a
prospective analysis on statewide electronic medical records. Int J Med Inf. 015;84:1039–47.
● Ontologies and semantic approach
53
54. Background
Health Recommender Systems
○ Natural Language Processing Natural (NLP) constitutes
an element that, together with ontologies, improves
the levels of analysis on a recommendation.
○ It opens great possibilities in the improvement of the
algorithm, especially when the user’s profile is
incorporated.
○ It include techniques, which can combine syntactic,
semantic, and contextual analyses.
● NLP and Sentiment Analysis
54
57. Methodology
HealthRecSys: a health recommender system
● Objectives:
1. To analyze and obtain access channels and resources for patients in
web 2.0 environments.
2. Search repositories with sources of health resources, to share,
validate and manage existing ones and create awareness among
patients.
3. Design and implement recommendations algorithms based on
collaborative techniques.
a. Algorithm to collect videos from a platform.
b. Algorithm advisor in health.
4. Recommend contents that are positively valued by users with similar
characterístics.
5. Usage of social network analysis and natural language processing
techniques characteristics such a reputation of popularity.
6. Evaluation of recommender algorithm.
57
58. Methodology
HealthRecSys: a health recommender system
Fernandez-Luque L, Karlsen R, Melton GB. HealthTrust: a social network approach for
retrieving online health videos. J Med Internet Res. 2012;14(1): e22. doi:10.2196/jmir.1985.
● Context application - Saluteca
HealthTrust
58
59. Methodology
HealthRecSys: a health recommender system
Sánchez-Bocanegra, C. L., Rivero, A.,Femandez-Luque, L., and Sevillano, J. L. (2012).
Saluteca, a spanish health video portal. Proceedings IV Workshop on Technology for
Health-care and Healthy Lifestyle.
● Context application - Saluteca
59
60. Methodology
HealthRecSys: a health recommender system
Sánchez-Bocanegra, C., Rivero-Rodriguez.,Fernández-Luque, L., and Sevillano, J. L2013.
Diavideos: a diabetes health video portal. In Lehmann, C. U., Ammen-werth, E., and Nøhr,
C., editors, MedInfo, volume 192 of Studies in Health Technology and Informatics, page
● Context application - Diavideos
○ Diavideos is a web platform that retrieves trustworthy
diabetes videos from Youtube and offers a high-quality
set of them to the user.
60
61. Methodology
HealthRecSys: a health recommender system
● Context application - Diavideos
○ A selection of trustworthy diabetes-related users from
YouTube is made using the algorithm HealthTrust.
○ A web crawler (see figure right) uses the YouTube API
to extract diabetes videos and their metadata from
trusted sources.
○ The web crawler periodically updates information about
the videos and integrates it into web portal.
○ The web portal is based on Drupal, an Open Source
Content Management System.
61
63. Methodology
HealthRecSys: a health recommender system
● Context application - Diavideos
Sánchez-Bocanegra, C., Rivero-Rodriguez.,Fernández-Luque, L., and Sevillano, J. L2013.
Diavideos: a diabetes health video portal. In Lehmann, C. U., Ammen-werth, E., and Nøhr,
C., editors, MedInfo, volume 192 of Studies in Health Technology and Informatics, page
63
64. Methodology
HealthRecSys: a health recommender system
● Objectives:
1. To analyze and obtain access channels and resources for patients in
web 2.0 environments.
2. Search repositories with sources of health resources, to share,
validate and manage existing ones and create awareness among
patients.
3. Design and implement recommendations algorithms based on
collaborative techniques.
a. Algorithm to collect videos from a platform.
b. Algorithm advisor in health.
4. Recommend contents that are positively valued by users with similar
characterístics.
5. Usage of social network analysis and natural language processing
techniques characteristics such a reputation of popularity.
6. Evaluation of recommender algorithm.
64
66. Methodology
HealthRecSys: a health recommender system
● Youtube tags!
http://newmediarockstars.com/2012/08/youtube-disables-public-video-tags-in-their-api-will-it-make-a-difference/66
68. Methodology
HealthRecSys: a health recommender system
● Title and Description
○ “Diabetes Mellitus”
○ “Insulin”
○ “Diabetes Type 2”
○ “Blood Sugar”
Diabetes: How insulin works
https://www.youtube.com/watch?v=VgGHSddJrUQ
● RoadMap
68
70. Methodology
HealthRecSys: a health recommender system
Rivero-Rodriguez A, Konstantinidis ST, Sanchez-Bocanegra CL, Fernandez-Luque L. A
health information recommender system: Enriching YouTube health videos with Medline
Plus information by using SnomedCT terms. IEEE 26th International Symposium on
Computer-Based Medical Systems (CBMS). P. 257-61. 2013.
Method A
Method B
Method D
Method C
● Searching Terms
70
71. Methodology
HealthRecSys: a health recommender system
○ A method that estimates the precise of recommended
links with trustworthiness health videos.
○ Two different experiments:
■ Diabetes and high blood pressure.
■ 4 clinicians evaluated recommended links from:
● 6 health videos (recollected from diavideos portal) with more
than 3 recommended results.
● 17 of High Blood Pressure (HTA) most popular videos with more
than 3 recommended results.
■ 23 vídeos provided 114 ratings.
Fernandez-Luque L., Sánchez-Bocanegra C., Ramos J.L, Karlsen R. A Semantic Web
Health Recommender System: Enriching YouTube Health Videos. Medicine 2.0 Congress
● Term from Health Ontologies: bioontology.org
71
73. Methodology
HealthRecSys: a health recommender system
○ Online tools and a Web portal enabling them to access,
review, and integrate disparate health ontological
resources in all aspects of biomedical investigation
and clinical practice.
July 2017 - https://www.bioontology.org/
● Term from Health Ontologies: bioontology.org
73
78. Methodology
HealthRecSys: a health recommender system
Fernandez-Luque L., Sánchez-Bocanegra C., Ramos J.L, Karlsen R. A Semantic Web
Health Recommender System: Enriching YouTube Health Videos. Medicine 2.0 Congress
avavailable at http://www.medicine20congress.com/ocs/index.php/med/med2014b/paper/view/2455. 2015.
● Term from Health Ontologies: bioontology.org
78
83. Methodology
HealthRecSys: a health recommender system
Sánchez-Bocanegra C., Sevillano J.L., Rizo C, Civit A., Fernandez-Luque L. HealthRecSys:
A semantic content-based recommender system to complement health videos. Medical
Informatics Decision Making - BioMed Central Journal. DOI:10.1186/s12911-017-0431-
7.2017.
● Term from Health Ontologies and NLP
83
84. Methodology
HealthRecSys: a health recommender system
● Term from Health Ontologies: bioontology.org
○ cTakes is a health-specific NLP implementation that
extracts SNOMED-CT health terms from text.
84
85. ● Objectives:
1. To analyze and obtain access channels and resources for patients in
web 2.0 environments.
2. Search repositories with sources of health resources, to share,
validate and manage existing ones and create awareness among
patients.
3. Design and implement recommendations algorithms based on
collaborative techniques.
a. Algorithm to collect videos from a platform.
b. Algorithm advisor in health.
4. Recommend contents that are positively valued by users with similar
characterístics.
5. Usage of social network analysis and natural language processing
techniques characteristics such a reputation of popularity.
6. Evaluation of recommender algorithm.
Methodology
HealthRecSys: a health recommender system
85
87. Results
HealthRecSys: a health recommender system
Sánchez-Bocanegra, C. L., Rivero, A.,Femandez-Luque, L., and Sevillano, J. L. (2012).
Saluteca, a spanish health video portal. Proceedings IV Workshop on Technology for
Health-care and Healthy Lifestyle.
● Saluteca
87
88. Results
HealthRecSys: a health recommender system
Sánchez-Bocanegra, C. L., Rivero, A.,Femandez-Luque, L., and Sevillano, J. L. (2012).
Saluteca, a spanish health video portal. Proceedings IV Workshop on Technology for
Health-care and Healthy Lifestyle.
● Saluteca
More than 1.000 trusted videos
22 spanish channels
88
89. Results
HealthRecSys: a health recommender system
Sánchez-Bocanegra, C. L., Rivero, A.,Femandez-Luque, L., and Sevillano, J. L. (2012).
Saluteca, a spanish health video portal. Proceedings IV Workshop on Technology for
Health-care and Healthy Lifestyle.
● Saluteca
Saluteca Youtube
89
90. Results
HealthRecSys: a health recommender system
Sánchez-Bocanegra, C. L., Rivero, A.,Femandez-Luque, L., and Sevillano, J. L. (2012).
Saluteca, a spanish health video portal. Proceedings IV Workshop on Technology for
Health-care and Healthy Lifestyle.
● Saluteca
Saluteca Youtube
90
91. Results
HealthRecSys: a health recommender system
● Diavideos
The crawler of Diavideos has integrated over 1000 videos
from 31 trusted channels (e.g. reputable member of
the diabetes community, health authorities).
The number of videos increases automatically due to the
synchronization with YouTube.
The crawler can update up to 100 videos every 30
minutes.
Videos are inserted and indexed in a web platform based
on Drupal (CMS), allowing advanced search and
recommendations.
Sánchez-Bocanegra, C., Rivero-Rodriguez.,Fernández-Luque, L., and Sevillano, J. L2013.
Diavideos: a diabetes health video portal. In Lehmann, C. U., Ammen-werth, E., and Nøhr,
C., editors, MedInfo, volume 192 of Studies in Health Technology and Informatics, page
91
92. Results
HealthRecSys: a health recommender system
● Searching Terms (CBMS)
○ Methods A and D were applied on a set of 1000
videos, extracted from the twenty most popular
American hospital channels in YouTube.
○ The methods calculate the MedlinePlus
recommendation links. Those links were evaluated of
their relativeness with the video content itself by 2
expert.
Rivero-Rodriguez A, Konstantinidis ST, Sanchez-Bocanegra CL, Fernandez-Luque L. A
health information recommender system: Enriching YouTube health videos with Medline
Plus information by using SnomedCT terms. IEEE 26th International Symposium on
Computer-Based Medical Systems (CBMS). P. 257-61. 2013.
92
93. Results
HealthRecSys: a health recommender system
● Searching Terms
○ It was calculated the hit rate of the method as the
average of the hit rage of the method for the five
selected videos, evaluated by the two users.
■ Hit rate for method A was 4.94%
■ Hit rate for method D was 46.44%
Rivero-Rodriguez A, Konstantinidis ST, Sanchez-Bocanegra CL, Fernandez-Luque L. A
health information recommender system: Enriching YouTube health videos with Medline
Plus information by using SnomedCT terms. IEEE 26th International Symposium on
Computer-Based Medical Systems (CBMS). P. 257-61. 2013.
93
94. Results
HealthRecSys: a health recommender system
● Term from Health Ontologies: bioontology.org
Fernandez-Luque L., Sánchez-Bocanegra C., Ramos J.L, Karlsen R. A Semantic Web
Health Recommender System: Enriching YouTube Health Videos. Medicine 2.0 Congress
avavailable at http://www.medicine20congress.com/ocs/index.php/med/med2014b/paper/view/2455. 2015.
○ Evaluated for 4 health professionals.
○ 23 vídeos.
■ 6 Diabetes.
■ 17 HTA.
○ 114 ratings.
94
95. Results
HealthRecSys: a health recommender system
● Term from Health Ontologies and NLP
○ Assigned 26 health professionals (raters)
○ 53 videos with 510 recommended links.
■ 10 from general medical videos (CBMS) with 48
recommended links.
■ 22 from diabetes with 102 recommended links.
■ 21 from HTA with 103 recommended links.
○ 2 evaluation methods: precision@k and DCG
(Discounted Cumulative Gain)
○ 2 scenarios:
■ Robust: consider as relevant only thos ilinks that
are supported by both raters.
■ Moderate: consider a link to be relevant if at least
one rater agreed.
Sánchez-Bocanegra C., Sevillano J.L., Rizo C, Civit A., Fernandez-Luque L. HealthRecSys:
A semantic content-based recommender system to complement health videos. Medical
Informatics Decision Making - BioMed Central Journal. DOI:10.1186/s12911-017-0431-
7.2017.
95
96. Results
HealthRecSys: a health recommender system
● Term from Health Ontologies and NLP
Sánchez-Bocanegra C., Sevillano J.L., Rizo C, Civit A., Fernandez-Luque L. HealthRecSys: A
semantic content-based recommender system to complement health videos. Medical Informatics
Decision Making - BioMed Central Journal. DOI:10.1186/s12911-017-0431-7.2017.
96
97. Results
HealthRecSys: a health recommender system
● Term from Health Ontologies and NLP
Sánchez-Bocanegra C., Sevillano J.L., Rizo C, Civit A., Fernandez-Luque L. HealthRecSys: A
semantic content-based recommender system to complement health videos. Medical Informatics
Decision Making - BioMed Central Journal. DOI:10.1186/s12911-017-0431-7.2017.
97
98. Results
HealthRecSys: a health recommender system
● Term from Health Ontologies and NLP
Sánchez-Bocanegra C., Sevillano J.L., Rizo C, Civit A., Fernandez-Luque L. HealthRecSys: A
semantic content-based recommender system to complement health videos. Medical Informatics
Decision Making - BioMed Central Journal. DOI:10.1186/s12911-017-0431-7.2017.
98
99. Methodology
HealthRecSys: a health recommender system
● Objectives:
1. To analyze and obtain access channels and resources for patients in
web 2.0 environments.
2. Search repositories with sources of health resources, to share,
validate and manage existing ones and create awareness among
patients.
3. Design and implement recommendations algorithms based on
collaborative techniques.
a. Algorithm to collect videos from a platform.
b. Algorithm advisor in health.
4. Recommend contents that are positively valued by users with similar
characterístics.
5. Usage of social network analysis and natural language processing
techniques characteristics such a reputation of popularity.
6. Evaluation of recommender algorithm.
99
101. Conclusions
HealthRecSys: a health recommender system
● Saluteca & Diavideos
○ Youtube provides an easy way to produce mashups
and use the potential and use the potential involved
in audiovisual communication.
○ We can find many Google groups who give support and
quick answers too.
○ They contain one of the largest repositories of videos,
which are provided by trustworthy content providers
based on the metric HealthTrust and hand-picked
channels.
101
102. Conclusions
HealthRecSys: a health recommender system
● Searching Terms (CBMS)
○ In the limitations of our study could be considered that
the recommendations are based exclusively in the
titles, which sometimes are not very representative of
the content.
○ It is relevant to underline that Method D offers two
advantages: on one hand the number of related links is
higher and there are more relevant information for the
users. On the other hand there are less unrelated links
and consequently, it is much more probable to find
information related to the video.
Rivero-Rodriguez A, Konstantinidis ST, Sanchez-Bocanegra CL, Fernandez-Luque L. A
health information recommender system: Enriching YouTube health videos with Medline
Plus information by using SnomedCT terms. IEEE 26th International Symposium on
Computer-Based Medical Systems (CBMS). P. 257-61. 2013.
102
103. Conclusions
HealthRecSys: a health recommender system
● Term from Health Ontologies: bioontology.org
Fernandez-Luque L., Sánchez-Bocanegra C., Ramos J.L, Karlsen R. A Semantic Web
Health Recommender System: Enriching YouTube Health Videos. Medicine 2.0 Congress
avavailable at http://www.medicine20congress.com/ocs/index.php/med/med2014b/paper/view/2455. 2015.
○ Evaluated for 4 health professionals (raters)
○ The method showed promising results (specially in
HTN)
○ The research indicates that the use of semantic web
for recomender links seems to be precise with trusted
video contents.
103
104. Conclusions
HealthRecSys: a health recommender system
● Term from Health Ontologies and NLP
○ This study demonstrated that a semantic-based
recommender algorithm can provide relevant education
health websites as further reading for a given health
video.
○ The relevance of websites recommended by our
system decreased as we provided more
recommendations, but HealthRecSys still performed
well with up to five recommended.
○ This approach can reduce the burden of health
consumers when searching for reliable additional health
educational content.
○ It was evaluated for 26 health professionals (raters)
Sánchez-Bocanegra C., Sevillano J.L., Rizo C, Civit A., Fernandez-Luque L. HealthRecSys: A
semantic content-based recommender system to complement health videos. Medical Informatics
Decision Making - BioMed Central Journal. DOI:10.1186/s12911-017-0431-7.2017.
104
114. Discusion
HealthRecSys: a health recommender system
● Target
○ Our aim is to enrich trustworthy health information.
○ Recommender System can be applied in Health
Networks.
● Future works
○ Study the quality of the health ontologies.
○ Include a EHR on the recommendation.
○ Open perspectives with other ontologies for health
professionals.
■ Pubmed/MESH.
■ ClinicalTrials.
○ Increase the cases for evaluation.
○ Use of SparQL.
○ Include machine/deep learning techniques.
114