Using Social Media Data for Online Television Recommendation Services at RTÉ Ireland
Report
•3 likes•678 views
![Centre for Data Analytics
16
Community-based Recommendation
# Users: 34
# Tweets: 18
# Retweets: 30
[Laurenmx15]
"I wonder who's gonna move in the old
no 23 Slater household.. #eastenders"
[Tobiiiaaas]
"Ah psycho Abi is back #EastEnders"
"(On the plus side yay an episode
focusing on Dylan)! #Casualty"](https://image.slidesharecdn.com/recsysandreabarrazarteproject-151013225521-lva1-app6891/75/using-social-media-data-for-online-television-recommendation-services-at-rt-ireland-16-2048.jpg?cb=1668570192)
Using Social Media Data for Online Television Recommendation Services at RTÉ Ireland
•3 likes•678 views
Download to read offline
Report
Presentation at RecSysTV 2015
Recommended
More Related Content
What's hot
What's hot(20)
Viewers also liked
Viewers also liked(17)
Similar to Using Social Media Data for Online Television Recommendation Services at RTÉ Ireland
Similar to Using Social Media Data for Online Television Recommendation Services at RTÉ Ireland (20)
Recently uploaded
Recently uploaded(20)
Using Social Media Data for Online Television Recommendation Services at RTÉ Ireland
- 1. USING SOCIAL MEDIA DATA FOR ONLINE TELEVISION RECOMMENDATION SERVICES AT RTÉ IRELAND Unit for Information Mining and Retrieval (UIMR) The Insight Centre for Data Analytics National University of Ireland, Galway, Ireland Andrea Barraza-Urbina, Hugo Hromic, Ioana Hulpus, Benjamin Heitmann, Conor Hayes, Neal Cantle 2nd Workshop on Recommendation Systems for Television and Online Video – RecSysTV September 19th, 2015 Vienna, Austria
- 2. Centre for Data Analytics Agenda 2 Introduction • RTÉ Use Case, Challenges, Research Goal Our Approach • Data Integration • Proposed Solution Approach Conclusion and Future Work
- 3. Centre for Data Analytics Agenda 3 Introduction • RTÉ Use Case, Challenges, Research Goal Our Approach • Data Integration • Proposed Solution Approach Conclusion and Future Work
- 4. Centre for Data Analytics 4 Introduction 500 + watching hours Live Broadcast
- 5. Centre for Data Analytics 5 Introduction Recommendation Service Lack of personal preference data such as ratings. Lack of historical user session information. Dynamic inventory and limited life span of recommendable items. Challenges:
- 6. Centre for Data Analytics 6 Introduction
- 7. Centre for Data Analytics 7 Introduction Amazing acting this evening Rachel #thevixenisback @RCPilkington @RTEFairCity great script Don't think I've ever laughed as much, @damoandivor is hilarious tonight, still trying to catch my breath from all the laughing! Really interesting RTÉ documentary about the late Brian Lenihan. Didn't see it advertised much. Worth a watch
- 8. Centre for Data Analytics 8 RTÉ Project RTÉ Programme Preferences Implicit RTÉ Programme Preferences Transfer Learning
- 9. Centre for Data Analytics 9 Introduction: Research Goal Design a Recommender System that can offer programme suggestions to an anonymous user, based solely on information about the user’s current session and inferred preferences of other RTÉ users extracted from social media.
- 10. Centre for Data Analytics Agenda 10 Introduction • RTÉ Use Case, Challenges, Research Goal Our Approach • Data Integration • Proposed Solution Approach Conclusion and Future Work
- 11. Centre for Data Analytics 11 Data Integration USER TWEET Fair City EPISODE MEDIA LINKS Amazing acting this evening Rachel #thevixenisback @RCPilkington @RTEFairCity great script IMDb DBpedia URI PROGRAMME SOCIAL MEDIA RTÉ CONTENT LINKED DATA
- 12. Centre for Data Analytics 12 Generating Recommendations Collaborative Filtering Recommendation User-Item Matrix
- 13. Centre for Data Analytics 13 Generating Recommendations Collaborative Filtering Recommendation Number of common users between all pairs of programmes Item-Item Matrix Limitation: Data Sparsity
- 14. Centre for Data Analytics 14 mentions reply retweets Community-based Recommendation User – User Graph w1 w2 w3
- 15. Centre for Data Analytics 15 Window time range: From 2015-07-29 21:00:00 To 2015-07-30 20:59:59 # Users: 36648 # Tweets: 23232
- 16. Centre for Data Analytics 16 Community-based Recommendation # Users: 34 # Tweets: 18 # Retweets: 30 [Laurenmx15] "I wonder who's gonna move in the old no 23 Slater household.. #eastenders" [Tobiiiaaas] "Ah psycho Abi is back #EastEnders" "(On the plus side yay an episode focusing on Dylan)! #Casualty"
- 17. Centre for Data Analytics 17 Generating Recommendations Hybrid Recommendation
- 18. Centre for Data Analytics 18 Generating Recommendations USER TWEET Fair City EPISODE MEDIA LINKS Amazing acting this evening Rachel #thevixenisback @RCPilkington @RTEFairCity great script IMDb DBpedia URI PROGRAMME SOCIAL MEDIA RTÉ CONTENT LINKED DATA
- 19. Centre for Data Analytics Agenda 19 Introduction • RTÉ Use Case, Challenges, Research Goal Our Approach • Data Integration • Proposed Solution Approach Conclusion and Future Work
- 20. Centre for Data Analytics 20 Conclusion and Future Work • RTÉ project presents a representative use case to explore the potential use of microblogging data to enhance Recommendation Systems. • We proposed novel recommendation approaches that learn the user-item models from Twitter, and apply them to the context of an online TV Player. • Evaluate our proposed solutions with user testing.
- 21. 21
Editor's Notes
- Good Morning, My name is Andrea Barraza Today I am presenting the industrial case study at RTÉ Ireland which focuses on the use of social media for online television reco
- RTÉ is the national provider of television and radio in Ireland. The company makes its programmes available online through the RTÉ Player service which is a video on demand service RTE allows its users to both catch up on programmes they might have missed or watch streaming of live broadcasts The player holds a wide variety of content From both national and international providers From its 4 different tv channel targeted to a wide audience including children -national international -4 channels -different audience -
- Given RTE’s large product catalogue A recommendation service is a desirable feature However the development of such a service faces 3 main challenges Users don’t tend to login and it is private info Private info Products only available limited time
- However, RTE communicates with their users through other platforms And encourage users to participate and engage with each other in social media highly encourages users to discuss and share in social media their opinions and preferences concerning programmes. This provides a key opportunity in using data immersed within social media to better understand user show preferences. ---------------- To further relate with their audience, RTÉ also publishes content using social media channels such as Twitter, Facebook and YouTube. However, RTÉ uses social media to further relate with their users For example, to For example, from the RTE player service a user can Tweet about a programme To further relate with their audience, RTÉ also publishes content using social media channels such as Twitter, Facebook and YouTube. However, RTÉ engages with users through social media And we believe we can find user programme prefences implicit… ------------------- If linked with the Facebook account, users can easily express their views on RTÉ content using buttons such as “Watching it”, “Like” and “Share”. Also, the Player has a “Tweet” button that allows users to publish in Twitter about a given episode. As can be seen, RTÉ highly encourages users to discuss and share in social media their opinions and preferences concerning programmes. Finally, we highlight that RTÉ also offers content outside the RTÉ Player page. Besides possibly having their own website, programmes could also have: a Facebook page (e.g., to announce related news), Twitter user accounts (e.g., to promote events and gossip) and YouTube video channels (e.g., posting trailers).
- The analysis of social media can give us great insights into how user communities are formed around TV shows, the impact of the TV shows, the sentiment(s) expressed with respect to the programmes, as well as insights into their content, and ultimately provide users with valuable recommendations.
- How can we leverage knowledge from Social Media for Recommendation? For a solution, we draw inspiration from transfer learning [10] approaches that deal with the application of a model learnt in a source domain, to draw conclusions about a separate target domain.
- We would like to offer to an anonymous user … based on the user’s current session Recommendations based on the preferences of other users extracted from social media Current approaches: Require a rich user profile Do not consider input from social media
- To do this, we integrate data from heterogeneous sources Our approach relies on the integration between three types of data: social media content, RTÉ content and Linked Data. For seamless integration, flexibility and scalability we propose a graph-based data model. These connections are automatically created. Currently, for the annotation of the Tweets with Programmes, we use a manually curated set of hashtags as described in section 5.1. Episodes can further be inferred by analysing the date and time of the tweet as well as its content. Named entity recognition, linking and disambiguation can be used for linking Tweet content to mentioned DBpedia concepts.
- we can obtain a User‑Item matrix This graph could be used by several RS approaches described in [4], such as random walk similarity using ItemRank.
- All communities To form community clusters, the structure of twitter interactions among users is considered: user mentions user, user retweets post of user and user reply user "user retweets posts from other users" para q quede mas claro. Recuerda que este subgrafo es user-user weighted así que se considera homogéneo, directed y weighted.
- (48 tweets total) A specific community with its keywords and associated programmes East Enders is a british tv series<- soap Holby city is a spin-off of Casualty… british tv series<- drama series Neighbours …Also like a soap opera first screened in united kingdom
- The main intuition is to use graph algorithms over the Linked Data in order to assess semantic relatedness between tweets and ultimately between programmes. The key to this approach is the URI-URI matrix that captures semantic relatedness between URI’s. For instance, given a tweet about Ryan Tubridy, the computation of semantic relatedness identifies films transmitted by RTÉ in which Ryan Tubridy plays. A novel method for computing the URI-URI associations is Heitmann et al.’s approach [8], a form of spreading activation to provide content‑based recommendations, where recommended items are represented by URI’s.
- Introduction – RTÉ Use Case, Challenges and Research Goal
- The RTÉ project presents a representative use case to explore the potential use of microblogging data to enhance Recommendation Systems. Due to the requirements of this use case we have to treat all users of the RTÉ Player as anonymous. In this paper, we proposed novel recommendation approaches that learn the user-item models from Twitter, and apply them to the context of an online TV Player. To this end, we designed collaborative filtering as well as hybrid approaches, and showed how Linked Data can potentially be used to further enrich the knowledge extracted from social media We plan to evaluate our proposed RS solutions with user testing, both in a controlled environment and on-line. The purpose of the RS service is to increase user retention and increase user engagement with content on the RTÉ Player. Promising metrics to measure user engagement are the length of user sessions, the number of recurrent users, the number of videos watched by a user and the number of tweets sent via the RTÉ Player. Evaluations must be first carried out in a controlled environment with a sample of user volunteers. Later, A/B testing could be used to temporarily test a RS approach on the live service.
- We found that only 1,373 users tweeted about more than one programme (3.5% from retrieved users). We found 46 programmes had at least one shared user with another programme. However, most of the programme pairs had less than ten overlapping users. The maximum user overlap found was of 145 users between the programmes: RTÉ News: Six One and The Sunday Game. These users generated 888 tweets commenting on these two programmes.
- We found that only 1,373 users tweeted about more than one programme (3.5% from retrieved users). We found 46 programmes had at least one shared user with another programme. However, most of the programme pairs had less than ten overlapping users. The maximum user overlap found was of 145 users between the programmes: RTÉ News: Six One and The Sunday Game. These users generated 888 tweets commenting on these two programmes.