Tag-based Semantic Website Recommendation
- 1. «Tag-based Semantic Website Recommendation for Turkish Language» Onur Yılmaz mail@onuryilmaz.me
- 2. Outline Introduction Related Work Problem Definition and Algorithm Experimental Evaluation Conclusion Future Work Demo
- 3. Introduction - Definitions Tags non-hierarchical keyword or term Reasons categorizing, memorizing, archiving and sharing…
- 4. Introduction - Motivation Dramatic increase in the number of the websites on the internet 7.14 billion pages Difficulty in finding and exploring new websites Social bookmarking Recommendation systems
- 5. Introduction – Turkish Effect Recommendation systems search within user inputs Users tend to use their own language on the internet Turkey is listed as 32nd country in English proficiency Turkish and English is very different languages!
- 6. Introduction – What is proposed? Tag-based recommendation system For Turkish-language Which is based on similarity, tag weight, tag popularity; Where semantic properties of tags are taken into account
- 7. Related Work Collaborative filtering Widely accepted No context! Topic and pattern extraction Usage of WordNet A lexical database for the English language 2 papers are found for Turkish WordNet but no source
- 8. Related Work Similarity calculation methods Durao & Dolog (2009) Reference paper Tag popularity, tag representativeness and taguser affinity Without any semantics analysis, 60 % acceptance level achieved
- 9. Problem Definition Take inputs Websites and tags Recommendation System
- 10. Problem Definition Provide personal recommendations Websites and tags Recommendation System
- 11. Problem Definition Aim -> User satisfaction Recommend websites User wants to use in the future, Already using and finds interesting
- 12. Problem Definition Challenge -> Different tagging purposes and expectations Website Tag Potential Purpose zaytung.com zaytung Archiving eksisozluk.com alışkanlık (ENG: habit) Internet usage habit evekitap.com ücretsiz kargo (ENG: free shipping) Categorizing 9gag.com eğlenceli (ENG: funny) Definition Data are taken from experiment
- 13. Algorithm Steps of the algorithm Spell-check Stemming Semantics Analysis Similarity Calculation
- 14. Algorithm – Spell-Checking Spell check on the tags Add a single letter, Delete a single letter, Replace one letter and Transpose two letters Estimated tags occur or not in Turkish National Corpus.
- 15. Algorithm – Spell-Checking Correction on URLs Original URL Corrected URL https://www.deviantart.com/ deviantart.com http://www.sahadan.com/Default.aspx sahadan.com http://www.yemeksepeti.com/AnonymouseDefault.aspx yemeksepeti.com Data are taken from experiment
- 16. Algorithm Steps of the algorithm Spell-check Stemming Semantics Analysis Similarity Calculation
- 17. Algorithm – Stemming Stems of the tags are extracted by removing suffices. Website facebook.com metu.edu.tr deviantart.com Original Tag Corrected Tag arkadaşlık arkadaş (ENG: friendship) (ENG: friend) mühendislik mühendis (ENG: engineering) (ENG: engineer) eğlenceli eğlence (ENG: funny) (ENG: fun) Data are taken from experiment
- 18. Algorithm Steps of the algorithm Spell-check Stemming Semantics Analysis Similarity Calculation
- 19. Algorithm – Semantics Analysis An open source «Turkish Thesaurus» project 125.022 <Word, Synonym> pairs
- 20. Algorithm – Semantics Analysis Algorithm applied: for each “tag” in ALL-DATA do: for each “synonym” of “tag” in SYNONYM-LIST do: if “synonym” occurs in ALL-DATA then: add <user, site, “synonym”> to ALL-DATA
- 21. Algorithm – Semantics Analysis User Website Tag User1 milliyet.com.tr haber (ENG: news) User2 sabah.com.tr gazete (ENG: newspaper) Original data (ALL-DATA) Word Synonym haber (ENG: news) gazete (ENG: newspaper) Synonym List (SYNONYM-LIST) User Website Tag User1 milliyet.com.tr gazete (ENG: newspaper) User2 sabah.com.tr haber (ENG: news) Added data to ALL-DATA Data are taken from experiment
- 22. Algorithm – Semantics Analysis An environment where all users provide tags and their potential meanings which other people may have already used.
- 23. Algorithm Steps of the algorithm Spell-check Stemming Semantics Analysis Similarity Calculation
- 24. Algorithm – Website Rating = Similarity Calculation Tag Popularity 𝑜𝑣𝑒𝑟 𝑎𝑙𝑙 𝑡𝑎𝑔𝑠 How often this tag is used? x Tag Representativeness How much a tag can represent a document? The more used for document, the more representative
- 25. Algorithm – Similarity = (a,b) Similarity Calculation Document + Score (a) Document x Score (b) Cosine Similarity (a,b) Tags as vectors, Cosine similarity between vectors
- 26. Experimental Evaluation Call for participation Gather websites and tags Find recommendations Ask for evaluation
- 27. Experimental Evaluation Call for participation Gather websites and tags www.eksiduyuru.com Find recommendations Ask for evaluation
- 28. Experimental Evaluation Call for participation Gather websites and tags 25 users 122 websites 366 tags Find recommendations Ask for evaluation bit.ly/oneri-sistemi
- 29. Experimental Evaluation Call for participation Gather websites and tags 20 of 25 Users Find recommendations Ask for evaluation bit.ly/oneri-sistemi-degerlendirme
- 30. Experimental Evaluation Expected Results Recommendation Acceptance 50 % Not acceptable 80 % Excellent (not expected)
- 31. Experimental Evaluation Results For top 5 28% Accepted Recomm. recommendations Accepted Rejected 72% 5 4 3 2 Accepted 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 User Accepted recommendations by each user (5 Recommendations)
- 32. Experimental Evaluation Results For top 3 22% Accepted Accepted Recomm. recommendations Rejected 78% 3 2 1 Accepted 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 User Accepted recommendations by each user (3 Recommendations)
- 33. Conclusion What is presented? Turkish-language tag-based recommendation system Based on similarity, tag weight, tag popularity Semantic properties of tags are taken into account
- 34. Conclusion Main contribution Combining Well-known similarity measures and calculations Turkish semantics analysis
- 35. Conclusion Evaluation An experiment with 25 people Participants provide websites and tags Then evaluate recommendations
- 36. Future Work Pre-processing Stage English inputs Site yandex.com Tags harita, e-mail, arama Turkish inputs with English letters Site eksiduyuru.com Tags duyuru, alinik, satilik Translation or control over them
- 37. Future Work Semantic Analysis Small set of synonyms list 125.022 <word, synonym> pairs Larger and more comprehensive theasurus
- 38. Demo 2 users from experiment
- 39. Demo XXX@candasdemir.com Website Tags http://candasdemir.com pazarlama, kişisel, blog http://radikal.com.tr haber, gündem, güncel http://mynet.com portal, genel, haber http://sahibinden.com alışveriş, market, sahibinden http://markafoni.com moda, e-ticaret, alışveriş
- 40. Demo XXX@candasdemir.com Website Tags http://candasdemir.com pazarlama, kişisel, blog http://radikal.com.tr haber, gündem, güncel http://mynet.com portal, genel, haber http://sahibinden.com alışveriş, market, sahibinden http://markafoni.com moda, e-ticaret, alışveriş mynet.com bilgi mynet.com gazete radikal.com.tr bilgi radikal.com.tr gazete Added after semantics analysis
- 42. Demo demirkolXXX@hotmail.com Website Tags http://www.sahadan.com/Default.aspx eğlence, merak, futbol http://www.erepublik.com/ iletişim, strateji, oyun http://www.1907unifeb.org/forums fenerbahçe, sohbet, eğlence http://ligtv.com.tr/ maç özetleri, haber, futbol https://www.tuttur.com/ para, futbol, eğlence
- 43. Demo demirkolXXX@hotmail.com Website Tags http://www.sahadan.com/Default.aspx eğlence, merak, futbol http://www.erepublik.com/ iletişim, strateji, oyun http://www.1907unifeb.org/forums fenerbahçe, sohbet, eğlence http://ligtv.com.tr/ maç özetleri, haber, futbol https://www.tuttur.com/ para, futbol, eğlence ligtv.com.tr bilgi ligtv.com.tr gazete Added after semantics analysis
- 45. References Adrian, B., Sauermann, L., & Roth-berghofer, T. (2007). ConTag: A Semantic Tag Recommendation System. Proceedings of ISemantics’ 07 Aksan, Y. et al. (2012). Construction of the Turkish National Corpus (TNC). In Proceedings of the Eight International Conference on Language Resources and Evaluation (LREC 2012). İstanbul. Turkiye. http://www.lrec-conf.org/proceedings/lrec2012/papers.html Brill, E., & Moore, R. C. (2000). An Improved Error Model for Noisy Channel Spelling Correction. (Microsoft Research) Cattuto, C., Benz, D., Hotho, A., & Stumme, G. (2008). Semantic Grounding of Tag Relatedness in Social Bookmarking Systems. In The Semantic Web - ISWC 2008. 2008: Springer Durao, F., & Dolog, P. (2009). A Personalized Tag-based Recommendation in Social Web Systems. International Workshop on Adaptation and Personalization for Web 2.0
- 46. References Education First, (2012). EF EPI Country Rankings Frankfurt International School, (2001). The Differences Between English and Turkish ISPA (Investment Support and Promotion Agency) of Turkey, (2010).Turkish Information and Communication Technologies Industry. Deloitte Nakamoto, R., Nakajima, S., Miyazaki, J., & Uemura, S. (2007). Tagbased Contextual Collaborative Filtering. IAENG International Journal of Computer Science Özbek, A. (2012). Türkçe Eşanlamlı Kelimeler Sözlüğü Projesi (Turkish Thesaurus Project). Retrieved from http://github.com/maidis/mythes-tr
- 47. Thank you!