=Paper=
{{Paper
|id=Vol-1247/recsys14_poster19
|storemode=property
|title=Improving IMDb Movie Recommendations with Interactive Settings and Filters
|pdfUrl=https://ceur-ws.org/Vol-1247/recsys14_poster19.pdf
|volume=Vol-1247
|dblpUrl=https://dblp.org/rec/conf/recsys/DoomsPM14
}}
==Improving IMDb Movie Recommendations with Interactive Settings and Filters==
https://ceur-ws.org/Vol-1247/recsys14_poster19.pdf
Improving IMDb Movie Recommendations with Interactive
Settings and Filters
Simon Dooms Toon De Pessemier Luc Martens
iMinds-Ghent University iMinds-Ghent University iMinds-Ghent University
G. Crommenlaan 8, box 201 G. Crommenlaan 8, box 201 G. Crommenlaan 8, box 201
Ghent, Belgium Ghent, Belgium Ghent, Belgium
Simon.Dooms@UGent.be Toon.DePessemier@UGent.be Luc1.Martens@UGent.be
ABSTRACT 2. A 3-TIER ARCHITECTURE
IMDb is a widely known online movie platform that offers The MovieBrain system consists of a 3-tier architecture
movie information, allows to rate movies and recommends which includes a calculation back-end, a middleware web-
interesting movies to users. The IMDb movie recommenda- server and a front-end Chrome extension. All recommen-
tions do not however offer any means for interactivity or user dation calculations are performed on a high-performance
control, which inherently limits their contextual adaptabil- computing (HPC) infrastructure which iteratively retrains
ity. In this work we describe our Google Chrome extension recommendation models in the background and in real time
– called MovieBrain – which offers interactive movie recom- responds to user requests. A Google Chrome extension pro-
mendations and integrates the IMDb website for user rating vides a visual web-based user interface, and a webserver links
data. Dynamic settings and genre filters are available, allow- the back and front-end of the system while adding control,
ing users to manually fine-tune the recommendation process caching and data management.
and its results.
3. RECOMMENDATION ALGORITHMS
Categories and Subject Descriptors The MovieBrain recommender system is driven by a dy-
H.3.3 [Information Search and Retrieval]: Information namic hybrid recommendation strategy which we described
filtering, Relevance feedback, User-centered design in previous work [2]. Multiple individual recommendation
algorithms can be included and their recommendation re-
sults are dynamically optimized in a weighted hybridiza-
Keywords tion scenario. While the hybrid system allows to automati-
Recommender systems, IMDb, MovieBrain, Chrome exten- cally optimize a given evaluation metric (e.g., RMSE ), users
sion, MovieTweetings, Movies themselves can influence the recommendation process by
manually overriding the weight vectors associated with each
individual recommendation algorithm. For the MovieBrain
1. INTRODUCTION recommender system we integrated 4 recommendation algo-
While the Internet Movie Database (IMDb) website1 is rithms including MatrixFactorization, user-based collabora-
a widely known and popular online movie platform, its in- tive filtering, a popular and a most recent recommendation
tegrated recommendations are static and non-interactive in approach.
nature. Users have no way of controlling, influencing or User ratings from IMDb are complemented with the Movi-
fine-tuning the suggested movies other than by rating more eTweetings dataset [3], which is a live rating dataset com-
movies and waiting for the recommendation results to change. posed of IMDb ratings posted to Twitter. This dataset
It has been shown however that interactivity and user in- nicely complements the public ratings on the IMDb plat-
volvement in the recommendation process increases user sat- form, helps to alleviate any cold start symptoms and guaran-
isfaction and recommendation relevance [1]. In this work we tees the continued inclusion of recent, relevant and popular
illustrate our approach to improve the default IMDb recom- movies.
mendations by offering interactive recommendation settings
and filters in a Google Chrome extension called MovieBrain.
4. GOOGLE CHROME EXTENSION
1 Browser plugins integrate seamlessly in people’s everyday
http://www.imdb.com
Internet activity (i.e., browsing the web), they allow to inject
custom code into existing websites and track user browsing
behavior. Our MovieBrain Chrome extension extends the
IMDb website functionality by offering customizable movie
recommendations based on IMDb ratings that users have
provided. The beauty in this workflow lies in the fact that
for users interested in the MovieBrain service, ratings al-
ready available on IMDb can simply be re-used. Further-
Copyright is held by the author/owner(s).
RecSys 2014 Poster Proceedings, October 6-10, 2014, Foster City, Silicon Valley,
more, a Chrome extension has an additional advantage of
USA. scalability. Since an extension is a self-contained file, hosted
D
User interaction: Settings
Interpretable algorithm descriptions
User interaction: Genre filtering
Draggable genre buttons
Implicit feedback tracking
Exclude and include area
Figure 1: Screenshot of the MovieBrain front-end illustrating the general layout, user interaction options
via the Settings and Genre filtering and implicit feedback tracking through the monitoring of the Hide, Open
IMDb, and More links.
at the client side, the impact on the webserver will be limited Chrome extension that integrates public IMDb ratings from
to HTTP calls to its API. users and provides an enhanced recommendation experience
The user interface is based on a two-level visual design by allowing users to influence the recommendation process
where movies are initially presented by their movie poster using settings and filters.
and title. Clicking a movie, triggers a popover information
pane with more detailed movie information. For every rec- 6. ACKNOWLEDGMENTS
ommended movie two action links are available which allow
The described research activities were funded by a PhD
to hide a movie, or open its corresponding IMDb page. At
grant to Simon Dooms of the Agency for Innovation by Sci-
the bottom of the page a more button allows to load more
ence and Technology (IWT Vlaanderen). The experiments
recommendations.
in this work were carried out using the Stevin Supercom-
Users can change the individual importance weights of the
puter Infrastructure at Ghent University, funded by Ghent
integrated algorithms and filter movies by intuitively drag-
University, the Hercules Foundation and the Flemish Gov-
ging genre buttons to an exclude or include area in the user
ernment - department EWI.
interface as illustrated in Fig. 1. All requests to the mid-
dleware API are logged so that user interaction with the
front-end and typical user behavioral patterns (e.g., implicit 7. REFERENCES
feedback) can potentially be analyzed. The MovieBrain ex- [1] Svetlin Bostandjiev, John O’Donovan, and Tobias
tension source code can be found on Github2 . Over 70 users Höllerer. Tasteweights: A visual interactive hybrid
currently have installed the extension. As more users in- recommender system. In Proceedings of the Sixth ACM
stall and use the extension, more data is collected which Conference on Recommender Systems, RecSys ’12,
can ultimately be used to evaluate the user experience of pages 35–42, New York, NY, USA, 2012. ACM.
our integrated recommendation algorithms and interaction [2] Simon Dooms, Toon De Pessemier, and Luc Martens.
process in a very realistic usage scenario. Offline optimization for user-specific hybrid
recommender systems. Multimedia Tools and
Applications, pages 1–24, 2013.
5. CONCLUSIONS
[3] Simon Dooms, Toon De Pessemier, and Luc Martens.
While IMDb provides movie recommendations to users Movietweetings: a movie rating dataset collected from
who have rated movies, the recommendations are static and twitter. In Workshop on Crowdsourcing and Human
can not be interacted with. We have created a Google Computation for Recommender Systems, CrowdRec at
2 RecSys, volume 13, 2013.
http://github.com/sidooms/MovieBrain