Applied analytics takes many forms at Rakuten-Viki Global TV Recommender Challenge

Source: Rakuten.

Rakuten has concluded the Rakuten-Viki Global TV Recommender Challenge, which was held in partnership with Dextra Singapore.

Participants in the Challenge were tasked to develop a personalised recommender system for global TV site Viki while following a set of business and consumer considerations. Competitors were also asked to predict the movies watched in February and March of 2015. Teams were scored on accuracy, engineering considerations as well as business and user insights gleaned from the data.

Teams were given access to over 7 million lines of anonymised and partially masked data data, including viewer demographics, preferences and popular content features, after which they submitted recommendation systems for Viki. Six finalists were then chosen to present their solutions live during the finals:

Gabor Benedek, Associate Professor at Corvinus University of Budapest, represented Team Gbenedek. He explained that graph theory can be used to analyse data.

Gabor Benedek, Associate Professor at Corvinus University of Budapest, represented Team Gbenedek in presenting a solution based on graph theory. He ran the data through an existing analytic solution, LynxKite from Lynx Analytics, to identify the top videos for Viki customers.

To predict what was watched in February 2015, Associate Professor Benedek identified a connection between individual countries and the newly-released movies watched by many people the month before. For March however, he identified new films which are most similar to blockbusters.

Dr Liu Guimei, Research Scientist, Data Analytics Department, Institute for Infocomm Research, represented Team GM. She lists some of the considerations taken in building her model.

Dr Liu Guimei, Research Scientist, Data Analytics Department, Institute for Infocomm Research, explained how Team GM used historical viewing data to create its recommendation engine. The model recommends what is popular for infrequent users, as there is insufficient data about their behaviour; whereas actual behavioural data is used to generate recommendations for frequent users.

Advertising ROI Analyst Pritish Kakodkar, representing Team Pritish, analysed behaviour and user preferences with gradient boosted models (GBM) to generate new predictions.

Source: Le Nguyen The Dat, Senior Data Technologist at Commercialize TV, spoke for Team Le Nguyen The Dat. He used Tableau Public for visualising viewer data.

Le Nguyen The Dat, Senior Data Technologist at Commercialize TV, created a framework that will allow Rakuten and Viki to create recommendation engines based on different purposes. He also relied on Tableau Software for visualisation purposes. Team Le Nguyen The Dat used Tableau Public visualisation for illustration. He noted that a model that works for February data may not work for March data, and commented that audience behaviour is likely to be affected by home page banners and other promotions.

Team haipt, represented by Pham Thanh Hai, Co-founder and CTO of Teevers, used parameters like recency of the movies that an individual watches, where the movie is produced and video genres to come up with recommendations. Pham noted that liking a movie (preferences) and actually getting round to watching are separate things. "Some might want to watch immediately. Since might wait," he said.

The team also discovered that viewers love Korean shows, followed by Japanese and Taiwanese entertainment.

Team Merlion described a three-step approach to recommending videos, which began by predicting the videos that a user is likely to choose, followed by shortlisting the popular ones, and then filtering the shortlist to retain those which actually manage to hold viewers' attention.

The judges comprised Masaya Mori, Global Head, Rakuten Institute of Technology (RIT), Rohit Dewan, CTO, Viki, Dr Ewa Szymanska, Head of Research, RIT Singapore, Takashi Umeda, Data Scientist, RIT Tokyo, and Adam Lyle, Chairman, Newton Circus.

Interested?

Read the TechTrade Asia blog post on the welcome address highlighting Singapore's opportunity in providing applied analytics expertise.

posted from Bloggeroid