Content Recommendation through Semantic Annotation of User Reviews and Linked Data
2 likes620 views
This document discusses a method for enhancing recommender systems by utilizing semantic annotation of user reviews and linked data, aiming to improve recommendation diversity and novelty. It presents an algorithm, semrevrec, that processes user reviews to generate ranked item recommendations across various media categories, including books, movies, and music. The results, including precision and recall metrics for several algorithms, demonstrate the effectiveness of the proposed approach in providing better recommendations.
Content Recommendation through Semantic Annotation of User Reviews and Linked Data
- 1. Content Recommendation through Semantic Annotation of User Reviews and Linked Data Iacopo Vagliano, ZBW Kiel Diego Monti, Politecnico di Torino Maurizio Morisio, Politecnico di Torino Ansgar Scherp, ZBW Kiel K-CAP 2017, Austin, TX, USA 1
- 2. Recommender Systems • Recommender systems suggest items – 80% of movies watched on Netflix from recommendations1 – 60% of video clicks from recommendations in YouTube2 – Typically based on rating 2 “Which movie to watch?” 1. Gomez-Uribe & Hunt. ŒThe netflƒix recommender system: Algorithms, business value, and innovation. ACM TMIS 6, 4 (2016), 13. 2. Davidson et al. ThŒe YouTube Video Recommendation System. In: RecSys ’10. ACM, NY, USA, pp. 293–296
- 3. User Reviews • More expressive than ratings (multi-faceted users’ opinions) • May reveal connections among items 3 The visual pattern also has some references to the Kubrick´s „Odissey“ and they are pleasing to the eye A review of „Interstellar“
- 4. Goals • Extract information from user reviews and discover additional knowledge from Linked Data • Increase the diversity of recommendations • Increase the novelty of recommendations 4
- 5. Annotating and Discovering Entities 5 The visual pattern also has some references to the Kubrick´s „Odissey“ and they are pleasing to the eye dbr:Stanley_Kubrick dbr:2001:_A_Space_Odyssey dbr:Full_Metal_Jacket dbr:A_Clockwork_Orange dbo:Film dbp:is_director_of rdf:typedbo:Person A review of „Interstellar“
- 6. SemRevRec • Annotated and discovery entities feeding • Generation of candidate recommendations and ranking • Independence from the knowledge base 6
- 8. Ranking • Frequency of entities in the reviews • Interlinking of entities – Between a discovered entity and its origin – Between each candidate recommendation and the initial item 8
- 9. Results - Books • Top-10 recommendations • Ratings from Library Things • Reviews from Library Things 9 Algorithm Precision Recall nDCG EBN Diversity SemRevRec 0.0530 0.0530 0.0536 0.1946 0.9118 SPrank 0.0379 0.0346 0.0337 0.1562 0.8037 ItemKNN 0.0620 0.0564 0.0662 1.4956 0.2259 BPR 0.0862 0.0817 0.0895 0.6043 0.7177 Popular 0.0423 0.0343 0.0447 1.6034 0.6483
- 10. Results - Movies • Top-10 recommendations • Ratings from Movielens 1M • Reviews from IMDb 10 Algorithm Precision Recall nDCG EBN Diversity SemRevRec 0.0857 0.0561 0.0686 0.7820 0.2431 SPrank 0.0445 0.0254 0.0280 0.8813 0.1612 ItemKNN 0.1626 0.1105 0.1302 2.6846 0.0696 BPR 0.2347 0.1737 0.1930 1.8358 0.1769 Popular 0.1325 0.0840 0.0969 2.7439 0.1412
- 11. Results - Music • Top-10 recommendations • Ratings from HotRec 2011 LastFM • Reviews from Amazon 11 Algorithm Precision Recall nDCG EBN Diversity SemRevRec 0.0536 0.0549 0.0502 0.2411 0.9329 SPrank 0.0156 0.0158 0.0176 0.1834 0.9077 ItemKNN 0.1392 0.1428 0.1720 1.6023 0.4730 BPR 0.1545 0.1583 0.1808 0.9404 0.6547 Popular 0.0686 0.0703 0.0791 2.0360 0.6519
- 12. Conclusions • Extract information from user reviews and discover additional knowledge from Linked Data • Increase the diversity of recommendations • Increase the novelty of recommendations 12MOVING is funded by the EU Horizon 2020 Programme under the project number INSO-4-2015: 693092 i.vagliano@zbw.eu @maponaso More detailed results at https://arxiv.org/abs/1709.09973