From Social Networks to Behavioral Networks in Recommender Systems

Recommender systems are widely used for personalization of information on the Web and information retrieval systems. collaborative filtering (CF) is the most popular recommendation technique. However, classical CF (CCF) systems use only direct links and common features to model relationships between users. This paper presents a new densified behavioral network based collaborative filtering model (D-BNCF), based on the BNCF approach that uses navigational patterns to model relationships between users. D-BNCF exploits additionally social networks techniques, such as prediction link methods, to discover new links throughout the behavioral network. The final aim is the involvement of these new links in prediction generation to improve the quality of recommendations. The approach proposed is evaluated in terms of accuracy on a real usage dataset. The experimentation shows the benefit of exploiting new links to compute predictions as a high precision is reached. Besides, the evaluation of a combined model (that exploits the more accurate D-BNCF models) shows also the interest of combining similarities based on two different link prediction methods and its impact on the accuracy of high predictions.

show abstract

“…1) are denoted as follows: Esslimani et al (2009)] Additionally, we compare the performance of D-BNCF models to…”

Section: Resultsmentioning

confidence: 99%

Densifying a behavioral recommender system by social networks link prediction methods

Esslimani

Brun

Boyer

2010

Soc. Netw. Anal. Min.

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, many of these networks depend on extensive user activities, such as building a graph of pages that the user viewed [5], to build the model. Others depend on the page attributes, such as link structure [7], to provide recommendation. In contrast, our approach works with sessions only and relies on page titles and URLs.…”

Section: Related Workmentioning

confidence: 99%

“…Chen [5] converts access patterns to two measures, Web access graph (WAG) and page interest estimator (PIE), to predict a user's interest in a certain page. Networks built using navigational information are used in many areas to model and predict user behavior [6,7,8]. However, many of these networks depend on extensive user activities, such as building a graph of pages that the user viewed [5], to build the model.…”

Section: Related Workmentioning

confidence: 99%

Recommendation based on Deduced Social Networks in an educational digital library

Akbar¹,

Shaffer²,

Fan³

et al. 2014

IEEE/ACM Joint Conference on Digital Libraries

View full text Add to dashboard Cite

Discovering useful resources can be difficult in digital libraries with large content collections. Many educational digital libraries (edu-DLs) host thousands of resources. One approach to avoiding information overload involves modeling user behavior. But this often depends on user feedback, along with the demographic information found in user account profiles, in order to model and predict user interests. However, edu-DLs often host collections with open public access, allowing users to navigate through the system without needing to provide identification. With few identifiable users, building models linked to user accounts provides insufficient data to recommend useful resources. Analyzing user activity on a per-session basis, to deduce a latent user network, can take place even without user profiles or prior use history. The resulting Deduced Social Network (DSN) can be used to improve DL services. An example of a DSN is a graph whose nodes are sessions and whose edges connect two sessions that view the same resource. In this paper we present a recommendation framework for edu-DLs that depends on deduced connections between users. Results show that a recommendation system built from DSN-dependent parameters can improve performance compared to when only text similarity between resources is used. Our approach can potentially improve recommendation for DL resources when implicit user activities (e.g., view, click, search) are abundant but explicit user activities (e.g., account creation, rating, comment) are unavailable.

show abstract

“…Esslimani et al [37] proposed a feedback effect between similarity and social influence in online communities. By utilizing the social relations, we can obtain the strength of social relationship between users, and we can use this social relationship to generate more accurate recommendation results.…”

Section: Related Workmentioning

confidence: 99%

Information Filtering via Biased Random Walk on Coupled Social Network

Nie

Zhang

Dong

et al. 2014

The Scientific World Journal

View full text Add to dashboard Cite

The recommender systems have advanced a great deal in the past two decades. However, most researchers focus their attentions on mining the similarities among users or objects in recommender systems and overlook the social influence which plays an important role in users' purchase process. In this paper, we design a biased random walk algorithm on coupled social networks which gives recommendation results based on both social interests and users' preference. Numerical analyses on two real data sets, Epinions and Friendfeed, demonstrate the improvement of recommendation performance by taking social interests into account, and experimental results show that our algorithm can alleviate the user cold-start problem more effectively compared with the mass diffusion and user-based collaborative filtering methods.

show abstract

From Social Networks to Behavioral Networks in Recommender Systems

Cited by 17 publications

References 15 publications

Densifying a behavioral recommender system by social networks link prediction methods

Densifying a behavioral recommender system by social networks link prediction methods

Recommendation based on Deduced Social Networks in an educational digital library

Information Filtering via Biased Random Walk on Coupled Social Network

Contact Info

Product

Resources

About