Recommender Watch: Stumble Upon

This week I played around with my Preferences in Stumble Upon to see how it reacts to these changes.

I set my preferences to only cricket and it recommended some good websites.
In one case it also recommended a video being played online in one of the news telecast website.

Then I added University/College to the list and it recommended some useful websites for students.

Next I added Dating, Relationships and Love to the list and it brought up websites for the same.

In an interesting case, it recommended me a link for a course in the Cornell University, in which an article on the “Science of love” was posted maybe as some course material.

Recommended Website: http://instruct1.cit.cornell.edu/courses/econ669/love.html

In another case, it showed up a website having the topic “A Systems Engineering Approach to Dating and Relationships”

Ref: http://www.geocities.com/SouthBeach/1285/syspaper.html

It also showed up many website having a similar topic “How to say I Love You In different Languages”

How LikeMinds generates Recommendations

By now we all are familiar with the different types of recommender systems, but what we dont know much is how they are used in real world applications.

Lets look into one such recommender application named LikeMinds and see how it generates Recommendations for users.

Read the entire article at:
http://publib.boulder.ibm.com/infocenter/wpdoc/v510/index.jsp?topic=/com.ibm.wp.ent.doc/pzn/pzn_likeminds_recommendation_engine.html

HPRS: A Profitability based Recommender

Traditional Recommender Systems learn about user preferences over time and recommends products that fit the learned model of user preferences.

In tradition, recommendations are provided to customers based on purchase probability and customers’ references, without considering the profitability factor or sellers. This work presents a new profitability-based recommender system, HPRS (Hybrid Perspective Recommender System), which attempts to integrate the profitability factor into the traditional recommender systems

For the entire article please view HPRS: A profitability based recommender system
Mu-Chen Chen,; Long-Sheng Chen,; Fei-Hao Hsu,; Yuanjia Hsu,; Hsiao-Ying Chou,;
in the IEEE Xplore.

Login to IEEE Xplore and read the entire article
http://www.ieeexplore.ieee.org/xpl/freeabs_all.jsp?isnumber=4419131&arnumber=4419183&count=438&index=51

Preventing shilling attacks in online recommender systems

Collaborative filtering techniques have been successfully employed in recommender systems in order to help users deal with information overload by making high quality personalized recommendations. However, such systems have been shown to be vulnerable to attacks in which malicious users with carefully chosen profiles are inserted into the system in order to push the predictions of some targeted items.
In this paper we propose several metrics for analyzing rating patterns of malicious users and evaluate their potential for detecting such shilling attacks. Building upon these results, we propose and evaluate an algorithm for protecting recommender systems against shilling attacks. The algorithm can be employed for monitoring user ratings and removing shilling attacker profiles from the process of computing recommendations, thus maintaining the high quality of the recommendations.

Read the entire article at:
http://www.l3s.de/web/upload/documents/chirita05preventing.pdf

Shilling Recommender Systems For Fun And Profit

Recommender systems have become particularly common in e-commerce, where recommendation of items can often help a customer find what he/she is interested in and, therefore can help drive sales.
Some users in order to increase their market penetration may find it profitable to "Shill" recommender systems by lying to the systems in order to have their products recommended more often than those of their competitors.
This paper explores four open questions that may affect the effectiveness of such shilling attacks: which recommender algorithm is being used, whether the application is producing recommendations or predictions,how detectable the attacks are by the operator of the system,
and what the properties are of the items being attacked.
The questions are explored experimentally on a large data set of movie ratings. The results of the paper suggest that new ways must be used to evaluate and detect shilling attacks on recommender systems.

Read the entire article at:
http://www.grouplens.org/papers/pdf/p333-lam.pdf

A New Hybrid Recommender System Using Dynamic Fuzzy Clustering

In this paper, a new hybrid system is proposed for combining collaborative and content-based approaches that resolves some limitations of them. By the proposed system, the novelty and diversity of recommendations improve remarkably.
Furthermore, the precision and recall of the proposedsystem is slightly less than those of the best existinghybrid system (collaborative via content) so thatemploying this system is justifiable. By this approach, the items that have not been yet rated by any user can be recommended. Collaborative and content-based systems utilized by this work, use a hybrid method based on fuzzy clustering model (fuzzy subtractive clustering) that combines model and memory-based approaches so that its precision is comparable with the precision of the memory-based approach and its scalability is comparable with the scalability of the model-based approach.
Furthermore, in this work, a dynamic fuzzy clustering algorithm was proposed in which a measure is presented to determine the stage at which a complete reclustering is required. By applying this algorithm, the system is able to adapt to the dynamic and changing environment in a much less expensive manner in terms of computation times and resources.

Read the entire paper at: www.profsite.um.ac.ir/~rmonsefi/conferences/baghebani1.pdf

Making Recommendations Better: An Analytic Model for Human-Recommender Interaction

Recommender systems do not always generate good recommendations for users. In order to improve recommender quality, we argue that recommenders need a deeper understanding of users and their information seeking tasks. Human-Recommender Interaction (HRI) provides a framework and a methodology for understanding users, their tasks, and recommender algorithms using a common language.

Further, by using an analytic process model, HRI
becomes not only descriptive, but also constructive. It
can help with the design and structure of a
recommender system, and it can act as a bridge
between user information seeking tasks and
recommender algorithms.

Read more at:
http://www.grouplens.org/papers/pdf/mcnee-chi06-hri.pdf