SlideShare a Scribd company logo
ONLINE RECOMMENDER
SYSTEM FOR RADIO STATION
HOSTING: EXPERIMENTAL
RESULTS REVISITED
Dmitry I. Ignatov1, Sergey Nikolenko1,2, Taimuraz Abaev1, and
Natalia Konstantinova3
1National Research University Higher School of Economics, Russia
2Steklov Institute of Mathematics at St. Petersburg of the RAS,
Russia
3University of Wolverhampton, UK
IEEE/WIC/ACM International
Conference on Web Intelligence
August 11-14, 2014
Warsaw, Poland
OUTLINE
• FMhost Online Radio Hosting
• Recommender Model
• Data
• Model and Algorithms
• Quality of Service Evaluation (QoS)
• User and Radio Station Activity Analaysis
• Results for the proposed methods
• Comparison with SVD-based recommender
• Conclusion
ONLINE RADIO HOSTING
FMHOST
• FMhost.me or Host.fm
• Real radio, not a streamer
• Social network
• Lives
• New features
• Listener oriented
• Likes
• Favorites
USERS
• Unauthorized
• Listeners
• DJs
• Station owners
ONLINE RADIO HOSTING
FMHOST IN 2012
MUSIC RECOMMENDATION
Conferences and workshops:
• International Society for Music Information Retrieval Conference
(ISMIR)
• Recommender Systems Conference (RecSys)
• Workshop on Music Recommendation and Discovery (WOMRAD)
Web services:
• Last.fm
• Pandora
• iTunes
6
PREVIOUS WORK
Usually methods for music recommendation use
quite limited data sources:
• Collaborative filtering exploits only users’ ratings
• Acoustic methods relies on acoustic information
• Hybrid approaches combine different methods
7
PREVIOUS WORK
• B. Logan. Music recommendation from song sets. In Proc. the 5th
International Conference on Music Information Retrieval, Barcelona, Spain,
2004.
• O. Celma. Foafing the music: Bridging the semantic gap in music
recommendation. In Proc. the 5th International Semantic Web Conference,
Athens, Georgia, 2006.
• K. Yoshii, M. Goto, K. Komatani, T. Ogata, and H. G. Okuno. Hybrid
collaborative and content-based music recommendation using probabilistic
model with latent user preferences. In Proc. the 7th International
Conference on Music Information Retrieval, Victoria, Canada, 2006.
• S. Pauws, W. Verhaegh, and M. Vossen. Fast generation of optimal music
playlists using local search. In Proc. the 7th International Conference on
Music Information Retrieval, Victoria, Canada, 2006.
• Dmitry I. Ignatov, Andrey V. Konstantinov, Sergey I. Nikolenko, Jonas
Poelmans, Vasily Zaharchuk: Online Recommender System for Radio
Station Hosting. BIR 2012: 1-12.
8
THE PREVIOUS ALGORITHM
• Ignatov et al. 2011
MOTIVATION
• It is rare case when different approaches to
recommendations are used together (e.g. history of
listening and tags)
• Too few research activity in radiostation
recommendation
10
PROBLEM SETTING
• To propose models and algorithms for radiostation
(and music) recommendation
• To implement the proposed algorithms, test and
compare them on real data of radio hosting FMHost
11
FMHOST DATA
Entity Count
User 4266
Tag 3618
Radiostation
s
2209
Tracks 4165
12
Relation Count
User-tag 38504
Radiostation-tag 18539
User-Radiostation 24803
Track-tag 18781
Radiostation-track 22525
Users
Radiostatio
ns
Tags
Track
s
THE MODEL: DATA
• U is a set of users, R is a set of radio stations, T is a set
of tags
• A=(aut), B=(brt), C=(cur), X=(xst)
• frequency vectors
• Normalized matrices, e.g.
THE MODEL: ARCHITECTURE
METHODS: INDIVIDUAL-
BASED RECOMMENDER
SYSTEM (IBRS)
Distance between user and radiostation:
𝑑 𝑢0, 𝑟 =
𝑡∈𝑇
|𝑎 𝑢0 𝑡 − 𝑏 𝑟𝑡|
Relevance of radiostation 𝑟𝑖 for user 𝑢0:
𝑠𝑐𝑜𝑟𝑒 𝑟𝑖 = 1 − 𝑑(𝑢0, 𝑟𝑖)/ max
𝑟 𝑗∈𝑅
𝑑(𝑢0, 𝑟𝑗)
15
METHODS: COLLABORATIVE-
BASED RECOMMENDER
SYSTEM (CBRS)
Users’ similarity: sim 𝑢0, 𝑢 = 𝑡∈𝑇 𝑢0𝑡 𝑢 𝑡
𝑡∈𝑇 𝑢0𝑡
2 ∙ 𝑡∈𝑇 𝑢 𝑡
2
𝑠𝑖𝑚 𝑢0, 𝑢 = 1 − 𝑑 𝑢0 𝑢/ max
𝑢′∈𝑈
𝑑 𝑢0 𝑢′
Relevance of radiostation for the target user 𝑢0:
𝑠𝑐𝑜𝑟𝑒 𝑟 = 𝑠𝑖𝑚 𝑢∗ ∙ 𝑐 𝑓𝑢∗ 𝑟
𝑢∗ = 𝑎𝑟𝑔𝑚𝑎𝑥 𝑢∈𝑈 𝑢0,𝑟∈𝑅/𝐿(𝑢0) 𝑠𝑖𝑚(𝑢) ∙ 𝑐𝑓𝑢𝑟
𝑐𝑓𝑢𝑟 – frequency of visits of radiostation 𝑟 by user 𝑢
𝐿(𝑢0) – the set of radiostations listened by user 𝑢0
𝑈 𝑢0
– the set of k most similar users with the target user 𝑢0
16
METHODS: FUSION
RECOMMENDER
SYSTEM (FRS)
For each recommendation list of size n:
𝛽∗ ∙ 𝑠𝑐𝑜𝑟𝑒 𝐶 𝑟 + 1 − 𝛽∗ ∙ 𝑠𝑐𝑜𝑟𝑒 𝐼 𝑟
we maximize 𝛽 by a chosen quality measure:
𝛽∗ = 𝑎𝑟𝑔𝑚𝑎𝑥 𝛽∈[0,1] 𝐹−𝑚𝑒𝑎𝑠𝑢𝑟𝑒
𝛽∗ = 𝑎𝑟𝑔𝑚𝑎𝑥 𝛽∈[0,1] 𝑁𝐷𝐶𝐺
17
METHODS: SVD-BASED
RECOMMENDER
• log 𝑐 𝑢𝑟 + 1 ~𝜇 + 𝑏 𝑢 + 𝑏 𝑟 + 𝑣 𝑢∙
𝑇
𝑣𝑟∙
𝑐 𝑢𝑟 is the number of times user 𝑢 listened to radio
station 𝑟
𝜇 is the general mean, 𝑏 𝑢 and 𝑏 𝑟 are the baseline
predictors for the user 𝑢 and the station 𝑟
𝑣 𝑢∙ and 𝑣𝑟∙ are the vectors of the user and station
features
18
METHODS:
MUSIC RECOMMENDATION TO USERS &
REPERTOIRE RECOMMENDATION FOR
RADIOSTATIONS
Similarity of users and songs:
𝑠𝑖𝑚 𝑢0, 𝑠 =
𝑡∈𝑇 𝑢0𝑡 𝑠𝑡
𝑡∈𝑇 𝑢0𝑡
2
∗ 𝑡∈𝑇 𝑠𝑡
2
Relevance of song 𝑠𝑖 for user 𝑢0 via
distance:
𝑠𝑐𝑜𝑟𝑒 𝑠𝑖 = 1 − 𝑑(𝑢0, 𝑠𝑖)/ max
𝑠 𝑗∈𝑆
𝑑(𝑢0, 𝑠𝑗)
Similarity of radiostations and songs:
𝑠𝑖𝑚 𝑟0, 𝑠 =
𝑡∈𝑇 𝑟0𝑡 𝑠𝑡
𝑡∈𝑇 𝑟0𝑡
2
∗ 𝑡∈𝑇 𝑠𝑡
2
Relevance of song 𝑠𝑖
for radiostation 𝑟0 via distance:
𝑠𝑐𝑜𝑟𝑒 𝑠𝑖 = 1 − 𝑑(𝑟0, 𝑠𝑖)/ max
𝑠 𝑗∈𝑆
𝑑(𝑟0, 𝑠𝑗)
19
QoS: DISTRIBUTION ANALYSIS
• Looking for Power Law P(x)=Cx-a
QoS: DISTRIBUTION ANALYSIS
QoS: DISTRIBUTION ANALYSIS
• Pareto Principle (20%:80%)
• 50%:80% for radio stations
• 50%:83% for user visits
RESULTS: IBRS
23
RESULTS: CBRS
24
RESULTS: FRS
MAXIMIZATION OF F-MEASURE
25
𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐶𝐵𝑅𝑆 𝑟 + 1 − 𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐼𝐵𝑅𝑆 𝑟
RESULTS: FRS
MAXIMIZATION OF NDCG
26
𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐶𝐵𝑅𝑆
𝑟 + 1 − 𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐼𝐵𝑅𝑆
𝑟
RESULTS: SVD
27
Solid line denotes error on the validation set;
dashed line, error on the training set.
RESULTS: COMPARISON
28
RESULTS: COMPARISON
29
RESULTS: COMPARISON
30
RESULTS: MUSIC
RECOMMENDATION TO USERS
31
RESULTS: RECOMMENDATION
OF REPERTOIRE FOR
RADIOSTATIONS
32
CONCLUSION
• We have described the underlying models, algorithms, and system
architecture of the new improved FMHost service and tested it on the
available real dataset.
• By using bimodal cross-validation, we have built a hybrid algorithm FRS
tuned to maximize either F-measure or NDCG for various values of N
and 𝛽. The FRS algorithm performs better than the three other
approaches, namely IBRS, CBRS, and SVD, both in terms of F-measure
and in terms of NDCG.
• According to the NDCG@n measure, IBRS is strictly better than CBRS,
so the former one is a better ranker.
• Surprisingly, in our experiments the state-of-the-art SVD-based
technique performed poorly in comparison to our proposed algorithms.
This can be explained by the small size and sparseness of our dataset.
33
Thank you!
Questions?
34

More Related Content

PPTX
Wireless comm ppt1
PDF
Rf planning wikipedia, the free encyclopedia
PDF
Spectrum Sensing with Energy Detection in Cognitive Radio Networks
PDF
WPMC2013_SBhattarai
PPTX
Cognitive Radio Spectrum Sensing 1586 ppt
PPTX
LOS Throughput Measurements in Real-Time with a 128-Antenna Massive MIMO Testbed
PDF
Mobile radio network_RAN Optimization
PPTX
Intelligent Selection of Transmission Waveform for Active Sonar Using Predict...
Wireless comm ppt1
Rf planning wikipedia, the free encyclopedia
Spectrum Sensing with Energy Detection in Cognitive Radio Networks
WPMC2013_SBhattarai
Cognitive Radio Spectrum Sensing 1586 ppt
LOS Throughput Measurements in Real-Time with a 128-Antenna Massive MIMO Testbed
Mobile radio network_RAN Optimization
Intelligent Selection of Transmission Waveform for Active Sonar Using Predict...

Viewers also liked (17)

PDF
Promotion of radiostation
PPTX
AIST 2016 Opening Slides
PDF
A lattice-based consensus clustering
PDF
Введение в рекомендательные системы. 3 case-study без NetFlix.
PDF
A One-Pass Triclustering Approach: Is There any Room for Big Data?
PDF
Experimental Economics and Machine Learning workshop
PDF
On the Family of Concept Forming Operators in Polyadic FCA
PDF
A lattice-based consensus clustering
PDF
Putting OAC-triclustering on MapReduce
PPTX
Boolean matrix factorisation for collaborative filtering
PPTX
NIPS 2016, Tensor-Learn@NIPS, and IEEE ICDM 2016
PDF
Pattern-based classification of demographic sequences
PDF
Intro to Data Mining and Machine Learning
PDF
Sequence mining
PDF
Поиск частых множеств признаков (товаров) и ассоциативные правила
PDF
Context-Aware Recommender System Based on Boolean Matrix Factorisation
PPTX
Recommender system introduction
Promotion of radiostation
AIST 2016 Opening Slides
A lattice-based consensus clustering
Введение в рекомендательные системы. 3 case-study без NetFlix.
A One-Pass Triclustering Approach: Is There any Room for Big Data?
Experimental Economics and Machine Learning workshop
On the Family of Concept Forming Operators in Polyadic FCA
A lattice-based consensus clustering
Putting OAC-triclustering on MapReduce
Boolean matrix factorisation for collaborative filtering
NIPS 2016, Tensor-Learn@NIPS, and IEEE ICDM 2016
Pattern-based classification of demographic sequences
Intro to Data Mining and Machine Learning
Sequence mining
Поиск частых множеств признаков (товаров) и ассоциативные правила
Context-Aware Recommender System Based on Boolean Matrix Factorisation
Recommender system introduction
Ad

Similar to Online Recommender System for Radio Station Hosting: Experimental Results Revisited (20)

PPTX
Radio recommender system for FMHost
PPT
P1121106496
PPT
Cognitive Radio & DSA ;Introduction et concept.ppt
PDF
Performance Analysis of MIMO–OFDM for PCHBF , RELAY Technique with MMSE For T...
PDF
Hablemos de ALMA — Wideband Sensitivity Upgrade
PDF
When a FILTER makes the di fference in continuously answering SPARQL queries ...
PPTX
Major Project Presentation final 1 (3).pptx
PDF
IRJET- Dynamic Spectrum Sensing using Matched Filter Method and MATLAB Simula...
PDF
Reference Network Models
PDF
IMCOM_2017_Presentation
PPTX
WiOpt2015_v1.0
PDF
Introduction to RADAR by NI
PPTX
Paper Review: ENERGY-EFFICIENT WIRELESS COMMUNICATIONS TUTORIAL, SURVEY, AND ...
PPTX
Experimental Evaluation of a Novel Fast Beamsteering Algorithm for Link Re-Es...
PPT
hans-werner_braun
PDF
OPTIMIZATION TECHNIQUES FOR WATER SUPPLY NETWORK: A CRITICAL REVIEW
PPTX
Color flow medical cardiac ultrasound
PDF
Radar 2009 a 15 parameter estimation and tracking part 1
PPTX
RFID 2013 talk-eve
Radio recommender system for FMHost
P1121106496
Cognitive Radio & DSA ;Introduction et concept.ppt
Performance Analysis of MIMO–OFDM for PCHBF , RELAY Technique with MMSE For T...
Hablemos de ALMA — Wideband Sensitivity Upgrade
When a FILTER makes the di fference in continuously answering SPARQL queries ...
Major Project Presentation final 1 (3).pptx
IRJET- Dynamic Spectrum Sensing using Matched Filter Method and MATLAB Simula...
Reference Network Models
IMCOM_2017_Presentation
WiOpt2015_v1.0
Introduction to RADAR by NI
Paper Review: ENERGY-EFFICIENT WIRELESS COMMUNICATIONS TUTORIAL, SURVEY, AND ...
Experimental Evaluation of a Novel Fast Beamsteering Algorithm for Link Re-Es...
hans-werner_braun
OPTIMIZATION TECHNIQUES FOR WATER SUPPLY NETWORK: A CRITICAL REVIEW
Color flow medical cardiac ultrasound
Radar 2009 a 15 parameter estimation and tracking part 1
RFID 2013 talk-eve
Ad

More from Dmitrii Ignatov (13)

PDF
Interpretable Concept-Based Classification with Shapley Values
PPTX
AIST2019 – opening slides
PDF
Turning Krimp into a Triclustering Technique on Sets of Attribute-Condition P...
PDF
Personal Experiences of Publishing with Springer from both Editor and Author ...
PDF
Searching for optimal patterns in Boolean tensors
PDF
Pattern Mining and Machine Learning for Demographic Sequences
PPTX
RAPS: A Recommender Algorithm Based on Pattern Structures
PPTX
Aist2014
PDF
Social Learning in Networks: Extraction Deterministic Rules
PPTX
Orpailleur -- triclustering talk
PDF
CoClus ICDM Workshop talk
PPTX
Pseudo-triclustering
PDF
CrowDM system
Interpretable Concept-Based Classification with Shapley Values
AIST2019 – opening slides
Turning Krimp into a Triclustering Technique on Sets of Attribute-Condition P...
Personal Experiences of Publishing with Springer from both Editor and Author ...
Searching for optimal patterns in Boolean tensors
Pattern Mining and Machine Learning for Demographic Sequences
RAPS: A Recommender Algorithm Based on Pattern Structures
Aist2014
Social Learning in Networks: Extraction Deterministic Rules
Orpailleur -- triclustering talk
CoClus ICDM Workshop talk
Pseudo-triclustering
CrowDM system

Recently uploaded (20)

DOCX
Viruses (History, structure and composition, classification, Bacteriophage Re...
PPTX
Classification Systems_TAXONOMY_SCIENCE8.pptx
PDF
An interstellar mission to test astrophysical black holes
PPTX
G5Q1W8 PPT SCIENCE.pptx 2025-2026 GRADE 5
PPTX
TOTAL hIP ARTHROPLASTY Presentation.pptx
PPTX
EPIDURAL ANESTHESIA ANATOMY AND PHYSIOLOGY.pptx
PPTX
GEN. BIO 1 - CELL TYPES & CELL MODIFICATIONS
PDF
IFIT3 RNA-binding activity primores influenza A viruz infection and translati...
PPTX
cpcsea ppt.pptxssssssssssssssjjdjdndndddd
PPTX
Microbiology with diagram medical studies .pptx
PDF
Formation of Supersonic Turbulence in the Primordial Star-forming Cloud
PPTX
ECG_Course_Presentation د.محمد صقران ppt
PPT
protein biochemistry.ppt for university classes
PPTX
The KM-GBF monitoring framework – status & key messages.pptx
PPTX
Cell Membrane: Structure, Composition & Functions
PDF
The scientific heritage No 166 (166) (2025)
PPTX
neck nodes and dissection types and lymph nodes levels
PPTX
2Systematics of Living Organisms t-.pptx
PPTX
Introduction to Cardiovascular system_structure and functions-1
PPTX
Introduction to Fisheries Biotechnology_Lesson 1.pptx
Viruses (History, structure and composition, classification, Bacteriophage Re...
Classification Systems_TAXONOMY_SCIENCE8.pptx
An interstellar mission to test astrophysical black holes
G5Q1W8 PPT SCIENCE.pptx 2025-2026 GRADE 5
TOTAL hIP ARTHROPLASTY Presentation.pptx
EPIDURAL ANESTHESIA ANATOMY AND PHYSIOLOGY.pptx
GEN. BIO 1 - CELL TYPES & CELL MODIFICATIONS
IFIT3 RNA-binding activity primores influenza A viruz infection and translati...
cpcsea ppt.pptxssssssssssssssjjdjdndndddd
Microbiology with diagram medical studies .pptx
Formation of Supersonic Turbulence in the Primordial Star-forming Cloud
ECG_Course_Presentation د.محمد صقران ppt
protein biochemistry.ppt for university classes
The KM-GBF monitoring framework – status & key messages.pptx
Cell Membrane: Structure, Composition & Functions
The scientific heritage No 166 (166) (2025)
neck nodes and dissection types and lymph nodes levels
2Systematics of Living Organisms t-.pptx
Introduction to Cardiovascular system_structure and functions-1
Introduction to Fisheries Biotechnology_Lesson 1.pptx

Online Recommender System for Radio Station Hosting: Experimental Results Revisited

  • 1. ONLINE RECOMMENDER SYSTEM FOR RADIO STATION HOSTING: EXPERIMENTAL RESULTS REVISITED Dmitry I. Ignatov1, Sergey Nikolenko1,2, Taimuraz Abaev1, and Natalia Konstantinova3 1National Research University Higher School of Economics, Russia 2Steklov Institute of Mathematics at St. Petersburg of the RAS, Russia 3University of Wolverhampton, UK IEEE/WIC/ACM International Conference on Web Intelligence August 11-14, 2014 Warsaw, Poland
  • 2. OUTLINE • FMhost Online Radio Hosting • Recommender Model • Data • Model and Algorithms • Quality of Service Evaluation (QoS) • User and Radio Station Activity Analaysis • Results for the proposed methods • Comparison with SVD-based recommender • Conclusion
  • 3. ONLINE RADIO HOSTING FMHOST • FMhost.me or Host.fm • Real radio, not a streamer • Social network • Lives • New features • Listener oriented • Likes • Favorites
  • 6. MUSIC RECOMMENDATION Conferences and workshops: • International Society for Music Information Retrieval Conference (ISMIR) • Recommender Systems Conference (RecSys) • Workshop on Music Recommendation and Discovery (WOMRAD) Web services: • Last.fm • Pandora • iTunes 6
  • 7. PREVIOUS WORK Usually methods for music recommendation use quite limited data sources: • Collaborative filtering exploits only users’ ratings • Acoustic methods relies on acoustic information • Hybrid approaches combine different methods 7
  • 8. PREVIOUS WORK • B. Logan. Music recommendation from song sets. In Proc. the 5th International Conference on Music Information Retrieval, Barcelona, Spain, 2004. • O. Celma. Foafing the music: Bridging the semantic gap in music recommendation. In Proc. the 5th International Semantic Web Conference, Athens, Georgia, 2006. • K. Yoshii, M. Goto, K. Komatani, T. Ogata, and H. G. Okuno. Hybrid collaborative and content-based music recommendation using probabilistic model with latent user preferences. In Proc. the 7th International Conference on Music Information Retrieval, Victoria, Canada, 2006. • S. Pauws, W. Verhaegh, and M. Vossen. Fast generation of optimal music playlists using local search. In Proc. the 7th International Conference on Music Information Retrieval, Victoria, Canada, 2006. • Dmitry I. Ignatov, Andrey V. Konstantinov, Sergey I. Nikolenko, Jonas Poelmans, Vasily Zaharchuk: Online Recommender System for Radio Station Hosting. BIR 2012: 1-12. 8
  • 9. THE PREVIOUS ALGORITHM • Ignatov et al. 2011
  • 10. MOTIVATION • It is rare case when different approaches to recommendations are used together (e.g. history of listening and tags) • Too few research activity in radiostation recommendation 10
  • 11. PROBLEM SETTING • To propose models and algorithms for radiostation (and music) recommendation • To implement the proposed algorithms, test and compare them on real data of radio hosting FMHost 11
  • 12. FMHOST DATA Entity Count User 4266 Tag 3618 Radiostation s 2209 Tracks 4165 12 Relation Count User-tag 38504 Radiostation-tag 18539 User-Radiostation 24803 Track-tag 18781 Radiostation-track 22525 Users Radiostatio ns Tags Track s
  • 13. THE MODEL: DATA • U is a set of users, R is a set of radio stations, T is a set of tags • A=(aut), B=(brt), C=(cur), X=(xst) • frequency vectors • Normalized matrices, e.g.
  • 15. METHODS: INDIVIDUAL- BASED RECOMMENDER SYSTEM (IBRS) Distance between user and radiostation: 𝑑 𝑢0, 𝑟 = 𝑡∈𝑇 |𝑎 𝑢0 𝑡 − 𝑏 𝑟𝑡| Relevance of radiostation 𝑟𝑖 for user 𝑢0: 𝑠𝑐𝑜𝑟𝑒 𝑟𝑖 = 1 − 𝑑(𝑢0, 𝑟𝑖)/ max 𝑟 𝑗∈𝑅 𝑑(𝑢0, 𝑟𝑗) 15
  • 16. METHODS: COLLABORATIVE- BASED RECOMMENDER SYSTEM (CBRS) Users’ similarity: sim 𝑢0, 𝑢 = 𝑡∈𝑇 𝑢0𝑡 𝑢 𝑡 𝑡∈𝑇 𝑢0𝑡 2 ∙ 𝑡∈𝑇 𝑢 𝑡 2 𝑠𝑖𝑚 𝑢0, 𝑢 = 1 − 𝑑 𝑢0 𝑢/ max 𝑢′∈𝑈 𝑑 𝑢0 𝑢′ Relevance of radiostation for the target user 𝑢0: 𝑠𝑐𝑜𝑟𝑒 𝑟 = 𝑠𝑖𝑚 𝑢∗ ∙ 𝑐 𝑓𝑢∗ 𝑟 𝑢∗ = 𝑎𝑟𝑔𝑚𝑎𝑥 𝑢∈𝑈 𝑢0,𝑟∈𝑅/𝐿(𝑢0) 𝑠𝑖𝑚(𝑢) ∙ 𝑐𝑓𝑢𝑟 𝑐𝑓𝑢𝑟 – frequency of visits of radiostation 𝑟 by user 𝑢 𝐿(𝑢0) – the set of radiostations listened by user 𝑢0 𝑈 𝑢0 – the set of k most similar users with the target user 𝑢0 16
  • 17. METHODS: FUSION RECOMMENDER SYSTEM (FRS) For each recommendation list of size n: 𝛽∗ ∙ 𝑠𝑐𝑜𝑟𝑒 𝐶 𝑟 + 1 − 𝛽∗ ∙ 𝑠𝑐𝑜𝑟𝑒 𝐼 𝑟 we maximize 𝛽 by a chosen quality measure: 𝛽∗ = 𝑎𝑟𝑔𝑚𝑎𝑥 𝛽∈[0,1] 𝐹−𝑚𝑒𝑎𝑠𝑢𝑟𝑒 𝛽∗ = 𝑎𝑟𝑔𝑚𝑎𝑥 𝛽∈[0,1] 𝑁𝐷𝐶𝐺 17
  • 18. METHODS: SVD-BASED RECOMMENDER • log 𝑐 𝑢𝑟 + 1 ~𝜇 + 𝑏 𝑢 + 𝑏 𝑟 + 𝑣 𝑢∙ 𝑇 𝑣𝑟∙ 𝑐 𝑢𝑟 is the number of times user 𝑢 listened to radio station 𝑟 𝜇 is the general mean, 𝑏 𝑢 and 𝑏 𝑟 are the baseline predictors for the user 𝑢 and the station 𝑟 𝑣 𝑢∙ and 𝑣𝑟∙ are the vectors of the user and station features 18
  • 19. METHODS: MUSIC RECOMMENDATION TO USERS & REPERTOIRE RECOMMENDATION FOR RADIOSTATIONS Similarity of users and songs: 𝑠𝑖𝑚 𝑢0, 𝑠 = 𝑡∈𝑇 𝑢0𝑡 𝑠𝑡 𝑡∈𝑇 𝑢0𝑡 2 ∗ 𝑡∈𝑇 𝑠𝑡 2 Relevance of song 𝑠𝑖 for user 𝑢0 via distance: 𝑠𝑐𝑜𝑟𝑒 𝑠𝑖 = 1 − 𝑑(𝑢0, 𝑠𝑖)/ max 𝑠 𝑗∈𝑆 𝑑(𝑢0, 𝑠𝑗) Similarity of radiostations and songs: 𝑠𝑖𝑚 𝑟0, 𝑠 = 𝑡∈𝑇 𝑟0𝑡 𝑠𝑡 𝑡∈𝑇 𝑟0𝑡 2 ∗ 𝑡∈𝑇 𝑠𝑡 2 Relevance of song 𝑠𝑖 for radiostation 𝑟0 via distance: 𝑠𝑐𝑜𝑟𝑒 𝑠𝑖 = 1 − 𝑑(𝑟0, 𝑠𝑖)/ max 𝑠 𝑗∈𝑆 𝑑(𝑟0, 𝑠𝑗) 19
  • 20. QoS: DISTRIBUTION ANALYSIS • Looking for Power Law P(x)=Cx-a
  • 22. QoS: DISTRIBUTION ANALYSIS • Pareto Principle (20%:80%) • 50%:80% for radio stations • 50%:83% for user visits
  • 25. RESULTS: FRS MAXIMIZATION OF F-MEASURE 25 𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐶𝐵𝑅𝑆 𝑟 + 1 − 𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐼𝐵𝑅𝑆 𝑟
  • 26. RESULTS: FRS MAXIMIZATION OF NDCG 26 𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐶𝐵𝑅𝑆 𝑟 + 1 − 𝛽 ∙ 𝑠𝑐𝑜𝑟𝑒 𝐼𝐵𝑅𝑆 𝑟
  • 27. RESULTS: SVD 27 Solid line denotes error on the validation set; dashed line, error on the training set.
  • 33. CONCLUSION • We have described the underlying models, algorithms, and system architecture of the new improved FMHost service and tested it on the available real dataset. • By using bimodal cross-validation, we have built a hybrid algorithm FRS tuned to maximize either F-measure or NDCG for various values of N and 𝛽. The FRS algorithm performs better than the three other approaches, namely IBRS, CBRS, and SVD, both in terms of F-measure and in terms of NDCG. • According to the NDCG@n measure, IBRS is strictly better than CBRS, so the former one is a better ranker. • Surprisingly, in our experiments the state-of-the-art SVD-based technique performed poorly in comparison to our proposed algorithms. This can be explained by the small size and sparseness of our dataset. 33

Editor's Notes

  • #4: Возможно этот слайд можно разбить. Здесь стоит упомянуть разницу между Fmhost-ом и Last-ом. Сказать так же, что наш проект позволяет устраивать интерактив и у каждого лайва есть свой блог, так же есть стена у каждой станции и у каждого зарегистрированного пользователя. Лайвы – это отдельная плюшка, живой интерактив с диджеем в блоге и исключительно живое проведение, человек сам должен отыграть лайв. Можно даже живую группу подсоединить, если есть желание. Новые опции – это новая панелька, которая позволяет вести лайвы с многими плюшками и удобствами, ну и так, кое какие архитектурные решения, которые сильно упрощают жизнь диджеям. Ориентация на слушателей, и привлечение их на проект (чему и служит рекомендательная система), что позволяет раскручивать станции и дижеев. Задача проекта вывести интернет радио на один уровень с обычным.
  • #5: 4 уровня пользователей: Неавторизованные - имеют право слушать станции и лайвы, просматривать информацию по диджеям, станциям и исполнителям. Слушатели – имеют право слушать все, но при этом они могут стать диджеем, проведя хоть один лайв. Так же они могут пользоваться рекомендательной системой. Могут оставлять комментарии и принимать участие в интерактиве лайвов. Так же они могут подписываться на лайвы и вступать в фан-клубы исполнителей (это новая фишка. После релиза я буду модернизировать систему и добавлять все эти плюшки в РС). DJs – получают статус диджея, после проведенного лайва. Имеют право вступать в команду станции, и получают рейтинг, который строится на основании лайвов, которые они проводят. Владельцы станций – имеют право набирать команду и строить вещание, а так же назначать лайвы.
  • #6: Возможно этот слайд можно разбить. Здесь стоит упомянуть разницу между Fmhost-ом и Last-ом. Сказать так же, что наш проект позволяет устраивать интерактив и у каждого лайва есть свой блог, так же есть стена у каждой станции и у каждого зарегистрированного пользователя. Лайвы – это отдельная плюшка, живой интерактив с диджеем в блоге и исключительно живое проведение, человек сам должен отыграть лайв. Можно даже живую группу подсоединить, если есть желание. Новые опции – это новая панелька, которая позволяет вести лайвы с многими плюшками и удобствами, ну и так, кое какие архитектурные решения, которые сильно упрощают жизнь диджеям. Ориентация на слушателей, и привлечение их на проект (чему и служит рекомендательная система), что позволяет раскручивать станции и дижеев. Задача проекта вывести интернет радио на один уровень с обычным.