CyberTube

In previous posts (literature survey, existing applications) devoted to a field of interests’ investigation, we published an overview of existing applications and researches related to our initial idea. The survey showed that the demand in the web market and research community for recommendation systems, has led to its mass production in the past few years. Some of these suggestion tools are quite successful (Jinni, Filmaster Blog, Faves for Facebook, etc.) as they provide rich functionality and a good quality of recommendations.  However, in this sketch we will differentiate our idea from existing solutions and also outline strengths of the proposed system.

See the difference

The majority of existing web market suggestion tools exploits traditional recommendation algorithms as Collaborative Filtering or Item-to-Item Recommendations. If we are talking about recommendation systems on top of social networks, these solutions, as a rule, use Collaborative Filtering as this approach allows utilizing existing relations between users. In our implementation we rejected traditional algorithms and decided to make a shift to trust-based recommendations. This means that the user will not just find out what movie to watch, but rather will be also provided with an information as who suggested the movie (in our case will be displayed only suggestions made by users from a friends list which will be extracted from the social network user profile).  The idea is that this additional information will boost the trust to the recommendations, since the user knows the person whom made the suggestion. However, we are not the first who considers trust-based approach (literature survey). Research community (Woerndl,  Groh and Zhou) already implemented some trust-based systems and ran a testing which proved decisive superiority of proposed approach over traditional recommendation algorithms. However, these projects still remain within the research field.

Rich functionality

Moving on, our system is not a simple recommendation tool that only provides a search box and a list of suggestions as many existing products. We want to tell our users not just what to watch and who suggested the movie, but also what is the best place for them to see it (the closest and most qualitative cinema) and who from their friends might be interested in watching it. In the other words, we want to make our system multifunctional providing opportunity to comment, rate and suggest movies and cinemas to friends, create invitations (movie events) and “follow” your friends. By “follow” we mean that some of your friends that have similar interests with you and some don’t, thus you will be interested in recommendations made by friends with similar preferences at first. Therefore our system provides subscription mechanism on recommendations and comments of your “like-minded” friends.  In addition to these core features, we plan to extend functionality by providing user reputation information, stimulation tools (credits to active users) and other useful features as “best comment/critics” or user activity reports.

CyberTube API

As for design’s point of view we decided to build an open system providing an API for potential developers. Moreover, it is planned that system data will be published in RDF format which is a standard way to provide information to the public and go with the times keeping in mind forthcoming era of semantic web. Provision of data in RDF will enable users that already have RDF data to point to our generated RDF in order  to state what movies they changed, their comments on it, etc. This information can be potentially used by agents (e.g. find a movie which was watched by a group of people more than 3 individuals).

Future work

The first variant of our system will support authorization using Facebook ID. However, it is planned to extend the list of social networks with MySpace or Twitter. In this case the user can define all social networks ID’s which are supported by our system what will allow to create some kind of a mash-up compiling final friends list from different “sources”. To clarify, we propose to build a new social movie-oriented network on the fly absorbing existing relationships from different networks. The main benefit of this mix is that it will bring more qualitative recommendations to the user as it will gather suggestion data from extended friends-lists. Another direction for future improvement is enhancement of an underlying recommendation algorithm. We realize that the trust-based recommendations will work quite well in case of a large friends-list. However, recently registered users or users with a small number of relations will not benefit from trust-based suggestions in full. Due to this reason we may consider combination of trust-based suggestions with other algorithms.

Our goal is to build a universal and highly flexible suggestion system which undoubtedly will be in demand, as the importance of qualitative information filtering tools increases from day to day. We strongly believe that selected approach based on aggregation of data for recommendations from distributed resources (themoviedb API, social networks infrastructure, geo-location data) allows the provision of precise, trustful and comprehensive suggestions. In addition, we are keens to make our system open and fully accessible to third parties which can harvest our data and use it for own purposes. Finally, we would like to point out once again main differences of our system from existing solutions:

1. Use of trust-based recommendations;
2. Use of geo-location information to narrow suggestions;
3. Rich functionality;
4. Openness of a system (CyberTube API, data distribution in RDF format);
5. Future support of a multiple social networks.