CyberTube

Our area of interests is focused on geo-based recommendation systems which exploit relations between users in social networks. In order to evaluate, adjust and extend our initial idea we performed a survey looking for both commercial products and academic researchers related to these topics. However, this post is devoted entirely to academic works. In our investigation we focused, first of all, on location-based recommendation systems on top of social networks, suggestion algorithms exploiting social network relations and the level of trust in recommendation systems and trust factors. Furthermore, we considered materials about the level of anonymity and privacy issues in social networks, legal issues with user-generated content,  social and economic impact of recommendation systems.

Recommendation systems and algorithms

The growing need of a good suggestion tool in the web of data boosted the development of different sorts of recommendation algorithms. Most popular and widely used are collaborative filtering (CF) and item to item recommendations. The first one suggests items exploiting the similarity between users when the second one provides recommendations measuring the similarity between items. To some extent both algorithms work quite well and produce good quality suggestions, however, they suffer from the number of limitations as impossibility to include new items which were not rated in the final suggestion list or take into account recently registered users. In order to overcome these drawbacks authors of [1] research proposed to use standard recommendation algorithms in conjunction with social networks infrastructure. They considered what suggestion algorithm can benefit from being used within the social network and their answer is definitely – collaborative filtering as it can consume existing information of user profile. However, researches stress that the pure CF recommendation based on the measurement of resemblance among all system users is less trust-worthy than the suggestion made by friend or family member. In the other words, paper proposes to utilize classical CF approach, physical (user location, time and environment) and social (interests, friend’s lists, etc.) contexts of user profiles in order to provide more precise and trust-worthy suggestions to users of mobile devices. Research team developed live demo and ran a set of tests which confirmed high effectiveness of proposed solution.

Some ideas expressed in a previous research were reflected by Chinese researcher Lixin Zhou [2] who also considered trust as a powerful tool for enhancing the effectiveness of any recommendation system. In order to provide precise recommendations he proposed to exploit Social Network Analysis technique which allows to extract valuable information from connections between users and recommend items with a reference to a friends list. In support to his proposal a prototype of trust-based recommendation system was developed. However, research paper doesn’t provide any empirical data which could prove the effectiveness of a new method in comparison with a standard CF.

Next research [3]  utilizes the Facebook Social Network infrastructure and API’s provided by third party web sites like www.maxmind.com for geo location information, www.last.fm for events gathering in order to make event recommendation based on the content that events (Content Based Method) have and the rating that users assign to them (Collaborative Method). Based on this information the system is able to make recommendation to a Facebook user and also suggest friend that might be interested in attending the selected event.

In addition to described above works [4] and [5] also consider benefits from using CF in conjunction with social networks infrastructure and trust weight and provide evidences of its successful use.

Recommendation quality factors: contribution and participation

Another interesting aspect which affects the quality of recommendations is the willing of social community to contribute and participate. “Social comparison and contributions to online communities (the MovieLens study case)” paper [6] raises questions about “under-contribution” and “non-participation” of social network members. Current study revealed that in content-sharing systems, for example, 25% of users share 98% of the content while the rest (75%) only consumes. This state-of-art creates a serious problem for recommendation systems as they rely heavily on user’s ratings and reviews in order to provide high quality recommendations. In follows researches questioned what can motivate community members to contribute? Results of a study showed that after notifying users about their social network behavior and the level of participation “passive” users show higher level of contribution (530% increase) while “active” users slow down their activity (62% decrease).

Presentation logic

The last paper in the current survey set considers the effect of a suggested items presentation on user’s persuasion and satisfaction. In the other words, researches focused on presentation logic of recommended items: they tested different presentation methods as N-items lists,  structured overview and recommendation modality (i.e. simple text, combination of text and image, and combination of text and video). Empirical evaluation of different methods revealed that the most efficient presentation modes are structured overview and recommendation modality (text + video or text + image).

Structured overview

Text plus image and text plus video presentation modes

Summary

Considering collected information we can conclude:

1. The use of a social networks infrastructure (relations between users, profile information and location data) in coherence with standard recommender algorithms allows to provide more qualitative and trust-worthy suggestions;
2. In order to support recommendation systems we need a high level of participation (reports about social activity and comparison with other users are needed);
3. Presentation logic plays an important role in user’s persuasion and satisfaction;

Features which might be useful in our project

1. The use of social networks infrastructure for recommendations;
2. Reports about social activity and the level of contribution;
3. The use of structured overviews or modality methods for recommended items presentation.

What’s next?

In this post we considered works related to recommender algorithms on top of social networks and factors which have an influence on the quality of suggestions as presentation methods and the level of participation. In the next post devoted to literature survey we will cover topics as geo location systems in social networks, the level of anonymity and privacy issues, legal issues with user-generated content and social and economic impact of recommendation systems.

References

1. W. Woerndl, G. Groh, “Utilizing Physical and Social Context to Improve Recommender Systems”, IEEE/WIC/ACM International Conferences on Web Intelligence and Intelligent Agent Technology, 2007.
2. L. Zhou, “Trust Based Recommendation System with Social Network Analysis”,  Information Engineering and Computer Science International Conference, 2009.
3. Kayaalp, M. Ozyer, T. Ozyer, S.T., “A Collaborative and Content Based Event Recommendation System Integrated with Data Collection Scrapers and Services at a Social Networking Site”, International Conference on Advances in Social Network Analysis and Mining, 2009.
4. Zhang, Ruiquan Sun, Baohui Kang, Wei Zhu, Tingshao, “GLOSS: A Social Networks-Based Recommender System”, 5th International Conference on Pervasive Computing and Applications (ICPCA), 2010.
5. Faqing Wu; Liang He; Weiwei Xia; Lei Ren, “A collaborative filtering algorithm based on Users’ Partial Similarity”, 10th International Conference on Control, Automation, Robotics and Vision, 2008.
6. Y.F. Chen, M. Harper, J. Konstan, S.X. Li, “Social Comparisons and Contributions to Online Communities:
   A Field Experiment on MovieLens”, Ph.D. Thesis, University of Minnesota, 2009.
7. T. Nanou, G. Lekakos, K. Fouskas, “The effects of recommendations’ presentation on persuasion and satisfaction in a movie recommender system”, vol.16 Multimedia Intelligent Services and Technologies,  2010.