No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Social Network Collaborative Filtering
Abstract
This paper demonstrates that "social network collaborative filtering" (SNCF), wherein user-selected like-minded alters are used to make predictions, can rival traditional user-to-user collaborative filtering (CF) in predictive accuracy. Us-ing a unique data set from an online community where users rated items and also created social networking links specifically intended to represent like-minded â¬Sallies,â¬? we use SNCF and traditional CF to predict ratings by net-worked users. We find that SNCF using generic "friend" alters is moderately worse than the better CF techniques, but outperforms benchmarks such as by-item or by-user average rating; generic friends often are not like-minded. However, SNCF using "ally" alters is competitive with CF. These results are significant because SNCF is tremendously more computationally efficient than traditional user-user CF and may be implemented in large-scale web commerce and social networking communities. It is notoriously difficult to distinguish the contributions of social influence (where allies influence users) and "socialâ¬? selection (where users are simply effective at selecting like-minded people as their allies). Nonetheless, comparing similarity over time, we do show no evi-dence of strong social influence among allies or friends.
... It has long been observed in sociology [40] that users' "friends" on such networks have similar taste (homophily). It is thus natural that new techniques [65] extended previous RSs by making use of social network structures. However, it was realized that the type of interaction taken into account could have a dramatic impact on the quality of the obtained social recommender [65]. ...
... It is thus natural that new techniques [65] extended previous RSs by making use of social network structures. However, it was realized that the type of interaction taken into account could have a dramatic impact on the quality of the obtained social recommender [65]. In this section, we review three families of social recommender: one based on explicit social links, one based on trust and an emerging family based on implicit links. ...
... -In [64], they use a graph theoretic approach to compute users' similarity as the minimal distance between two nodes (using Dijkstra's algorithm for instance), instead of using the ratings' patterns as in traditional CF; it is assumed that the influence will exponentially decay as distance increases. They show that this method produces results worse than traditional CF; -In [65], the user's neighborhood is just simply its set of friends in the network (first circle). This approach provides results slightly worse than the best CF. ...
This paper presents a general formalism for Recommender Systems based on Social Network Analysis. After introducing the classical categories of recommender systems, we present our Social Filtering formalism and show that it extends association rules, classical Collaborative Filtering and Social Recommendation, while providing additional possibilities. This allows us to survey the literature and illustrate the versatility of our approach on various publicly available datasets, comparing our results with the literature.
... Among the various methods used by different recommender systems, collaborative filtering (CF) has become the most popular one (Herlocker, Konstan, Terveen, & Riedl, 2004). CF algorithm makes recommendation based on users' similar rates to the existing items (Zheng, Wilkinson, & Provost, 2008). In this manner an implicit network is formed among people who are the nodes and each link between two users represents the similarity between them. ...
... Social networks are good resources of similarity data for recommender systems. This data usually is provided in the form of explicit usergenerated connections linking pairs of users together (Zheng, Wilkinson, & Provost, 2008). In doing this, since links between the users illustrate their similarity, the computational step of identifying users with the similar rates to the rates of the target user is removed, and the complexity of CF algorithm is reduced from O(N 2 + NM) to O(1) -for a system of N users and M items (Golder, Wilkinson, & Huberman, 2007). ...
Ecommerce systems employ recommender systems to enhance the customer loyalty and hence increasing the cross-selling of products. In collaborative filtering - as the most popular method in recommender systems - an implicit network is formed among all the people. In any network, there are some individuals who have some inspirational power over the others leading them to influence their decisions and behaviours. But it seems that these methods do not support context awareness in mobile commerce environments. Furthermore, they lack high accuracy and also require high volume of computations due to not distinguish between neighbours as a friend or a stranger. This paper proposes a new model for recommender systems which are based on mobile data. This model uses these data to extract current users' context and also to identify individuals with the highest influence. Then, the system uses the information of these identified impressive users in the current context existed in the social networks for making recommendations. Beside of achieving higher accuracy, the proposed model has resolved cold start problem in collaborative filtering systems.
... If two users tag a bookmark with the same tags that signifies that both the users are looking at the bookmark from the same perspective, thus both the users are similar and if the users annotate same bookmark with different tags that signifies that both the users are looking at the bookmark from different perspective hence the degree of similarity between the users is low. Zheng, et al. [9,10] used social network information to only compute similarity between the users connected via social network. This reduces the number of comparisons required for finding out similar users and the resulting algorithm has similar accuracy with much lower runtime. ...
... Plain CF (labeled CF) algorithm used bookmarks for computing similarity between. Social Network Collaborative Filtering (labeled SocialNet) [9,10] also used bookmarks for computing similar users but instead of computing similarity between the user and all other users, similarity was computed only between users who were part of user's social network. Tags were used instead of bookmarks for computing similarity between users as suggested in [ (labeled as Tag) and called Tag based CF from here on. ...
Collaborative Filtering (CF) is a process used by many recommender systems for recommending content based on similarities between users. CF traditionally used item purchase history or rating information only, for finding out similar users. Nowadays users interact with the content in many different ways. Users can rate, tag, like, dislike or subscribe to the content and some information about relationship between users is also available usually in social network form. This additional interaction information can be used to filter out similar users more effectively. Various CF algorithms have been proposed to take into account this additional information and suggest better content. But it is very hard to figure out which algorithm will work best on a particular dataset. In this paper, we propose a generic data model, for storing user interaction data, which can easily support many collaborative filtering algorithms. We have also implemented a tool which assists in creating, running and comparing different collaborative filtering algorithms. The tool is based on the proposed data model and can easily be extended by using plugins. It is helpful in figuring out which algorithm works best on a particular domain.
... Existing recommendation algorithms based on the content of the recommendation algorithm 4 , based on the project of collaborative filtering 5 , based on the user's collaborative recommendation 6 , model for collaborative filtering 7 , recommendation based on social network analysis method 8 , and the recent rise of the recommendation algorithm based on network structure 9 . After studying the commonly used knowledge push approaches, Mingjian Zhou and Qiang Liao 1 presents a knowledge push approach based on attribute similarity. ...
Enterprise knowledge recommendation approach combines two elements of knowledge and context. And knowledge and context composed context data pool. Knowledge consist of associated knowledge elements. Context refers to the time context. Based on the analysis of time context, considering the effect of time context on knowledge recommendation. Gantt chart is applied to describe the time-sequence relationship in the knowledge recommendation process. And through visual analysis of R programming language, it confirms the importance of time context. Enterprise knowledge recommendation approach is proposed based on context-aware. It improves the knowledge recommendation accuracy, and helps employees’ knowledge reuse effectively.
... Primarily, OSNs provide benefit for the users by creating a medium to communicate and connect with friends, colleagues and family. Recently, online social networks have found application in e-Commerce, especially for advertisements and other business transactions [3]. ...
The explosion of Internet usage has drawn the attention of researchers towards online Social Networks Marketing (SNM). Research has shown that a number of the Internet users are distrustful and indecisive, when it comes to the use of social networks marketing system. Therefore, there is a need for researchers to identify some of the factors that determine users’ acceptance of social networks marketing using Technology Acceptance Model (TAM). This study extended the Technology Acceptance Model theoretical framework to predict consumer acceptance of social networks marketing within Western Cape Province of South Africa. The research model was tested using data collected from 470 questionnaires and analysed using linear regression. The results showed that user intentions to use SNM are strongly and positively correlated with user acceptance of using SNM systems. Empirical results confirmed that perceived credibility and perceived usefulness are the strongest determinant in predicting user intentions to use SNM system.
... Primarily, OSNs provide benefit for the users by creating a medium to communicate and connect with friends, colleagues and family. Recently, online social networks have found application in e-Commerce, especially for advertisements and other business transactions [3]. ...
The explosion of Internet usage has drawn the attention of researchers towards online Social Networks Marketing (SNM). Research has shown that a number of the Internet users are distrustful and indecisive, when it comes to the use of social networks marketing system. Therefore, there is a need for researchers to identify some of the factors that determine users’ acceptance of social networks marketing using Technology Acceptance Model (TAM). This study extended the Technology Acceptance Model theoretical framework to predict consumer acceptance of social networks marketing within Western Cape Province of South Africa. The research model was tested using data collected from 470 questionnaires and analysed using linear regression. The results showed that user intentions to use SNM are strongly and positively correlated with user acceptance of using SNM systems. Empirical results confirmed that perceived credibility and perceived usefulness are the strongest determinant in predicting user intentions to use SNM system.
... Recommendation system based on social network analysis (Social Network Analysis, SNA) applies SNA methods to study the relationships between users, and then recommends based on the relationships [17]. Some research use the node trust between nodes which stand for users or knowledge, and recommend the use of trust between them which can have higher recommended effect than collaborative recommended [18]. ...
The purpose of this paper is to review the literatures which have made an explicit study on personalized recommendation in E-Learning systems. By identifying the important research areas, which are in different perspectives, firstly, filtering recommendation is introduced before the illustration of how it has been developed in E-Learning systems. Then personalized recommendation is proposed for E-Learning system. Although social network is the basic way to improve the communication efficiency with others in E-Learning system, previous studies pay less attention on this. Therefore social network analysis should be taken into consideration for the recommendation in E-Learning system for further research.
... Primarily, online social networks provide benefit for the users by creating a medium to communicate and connect with friends, colleagues and family. Recently, online social networks have found application in e-Commerce, especially for advertisements and other business transactions (Zheng et al., 2008). ...
The explosion of Internet usage has drawn the attention of researchers towards online Social Networks Marketing (SNM). Research has shown that a number of the Internet users are distrustful and indecisive, when it comes to the use of social networks marketing system. Therefore, there is a need for researchers to identify some of the factors that determine users’ acceptance of social networks marketing using Technology Acceptance Model (TAM). This study extended the Technology Acceptance Model theoretical framework to predict consumer acceptance of social networks marketing. The research model was tested using data collected from 470 questionnaires and analysed using linear regression. The results showed that user’s intentions to use SNM are strongly and positively correlated with user acceptance of using SNM system. Empirical results confirmed that perceived credibility and perceived usefulness are the strongest determinant in predicting user’s intention to use SNM system.
... The inclusion of social media links to improve collaborative fi ltering has been explored by a number of works. Zheng et al. 36 for example introduce social network links to complement traditional collaborative fi ltering approaches and predict rated items obtained from online communities, concluding that this provides a net advantage over simple by-item or by-user approaches. The method simply relies on weighted averages of ratings of the same items from self-selected friends. ...
The last few years have seen significant investment in social media as an advertising, marketing and customer outreach opportunity. In the US alone, in 2010, almost $1.7 bn was spent by advertisers on social media marketing, with 53 per cent specifically allocated to Facebook1. Due to the explicit links that users maintain with each other, social media platforms are perceived as a highly suited environment for network-based marketing: word-of-mouth marketing, diffusion of innovation, or buzz and viral marketing2 all aim to take advantage of the relationships between users to facilitate the spread of awareness or adoption. In order to predetermine the effectiveness of such campaigns, it is important to be able to estimate potential return on investment. In particular, the ability to model existing networks, track the propagation of marketing messages and estimate customer exposures and impressions are essential for this purpose. A wide range of techniques to measure notions such as user engagement on such platforms have been developed and there also exists a significant amount of research on modelling contagion and diffusion in network-based environments that can be exploited to generally refine an overall marketing strategy. However, the structure and properties of different social media platforms introduce various constraints on both the means via which data propagate and the visibility of content and nodes, constraints that must be taken into account when modelling or measuring the impact of social media campaigns. Perfect information about exposures within a given graph to a given message will not be available and as such it is important to investigate and define methodologies for diffusion monitoring that are suited to specific platforms.
... In [121], Zheng et. al. used social network collaborative filtering (SNCF) for recommendation. ...
The importance of machine learning for social network analysis is realized as an inevitable tool in forthcoming years. This is due to the unprecedented growth of social-related data, boosted by the proliferation of social media websites and the embedded heterogeneity and complexity. Alongside the machine learning derives much effort from psychologists to build computational model for solving tasks like recognition, prediction, planning and analysis even in uncertain situations. In this chapter, we have presented different network analysis concepts. Then we have discussed implication of machine learning for network data preparation and different learning techniques for descriptive and predictive analysis. Finally we have presented some machine learning based findings in the area of community detection, prediction, spatial-temporal and fuzzy analysis.
... Other recent studies have investigated the manifestation on SNSs of race and ethnicity [13], religion [14], gender [15,16], and national identity [17]. Other research has illustrated that online friendship networks can be exploited to improve shopper recommendation systems on websites such as Amazon [18]. (Presumably, this is becoming increasingly prominent in practice.) ...
We study the social structure of Facebook “friendship” networks at one hundred American colleges and universities at a single point in time, and we examine the roles of user attributes–gender, class year, major, high school, and residence–at these institutions. We investigate the influence of common attributes at the dyad level in terms of assortativity coefficients and regression models. We then examine larger-scale groupings by detecting communities algorithmically and comparing them to network partitions based on user characteristics. We thereby examine the relative importance of different characteristics at different institutions, finding for example that common high school is more important to the social organization of large institutions and that the importance of common major varies significantly between institutions. Our calculations illustrate how microscopic and macroscopic perspectives give complementary insights on the social organization at universities and suggest future studies to investigate such phenomena further.
... Blogs are likely to link to content that other blogs have linked to [27]. The insight that individuals tend to like (or like to have) the same things that their friends like can be used to improve collaborative filtering algorithms [32]. While social network information has been commonly utilized , there are relatively few studies that have included direct transfers between users. ...
Social influence determines to a large extent what we adopt and when we adopt it.This is just as true in the digi- tal domain as it is in real life, and has become of increas- ing importance due to the deluge of user-created content on the Internet. In this paper, we present an empirical study of user-to-user content transfer occurring in the context of a time-evolving social network in Second Life, a massively multiplayer virtual world. We identify and model social influence based on the change in adoption rate following the actions of one's friends and find that the social network plays a significant role in the adoption of content. Adoption rates quicken as the number of friends adopting increases and this e!ect varies with the connectivity of a particular user. We further find that sharing among friends occurs more rapidly than sharing among strangers, but that content that di!uses primarily through social influence tends to have a more lim- ited audience. Finally, we examine the role of individuals, finding that some play a more active role in distributing content than others, but that these influencers are distinct from the early adopters.
... An important attention has been also devoted to the combination of social networks and recommender systems. Some studies like [16] incorporate social network information into Collaborative Filtering. Nodes are consumers and links are the social relationships among them. ...
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 systems use only direct links and common features to model relationships between users. This paper presents a new Collaborative Filtering approach (BNCF) based on a behavioral network that uses navigational patterns to model relationships between users and exploits social networks techniques, such as transitivity, to explore additional links throughout the behavioral network. The final aim consists in involving these new links in prediction generation, to improve recommendations quality. BNCF is evaluated in terms of accuracy on a real usage dataset. The experimentation shows the benefit of exploiting new links to compute predictions. Indeed, BNCF highly improves the accuracy of predictions, especially in terms of HMAE.
... community, demonstrating how a local trust metric achieves higher accuracy than a global one in predicting how much trust users should place in each other. Similarly, [Zheng et al. 2007] performed a study on a dataset from Amazon.com to assess the impact of incorporating social-network information into CF, with the conclusion that neighbours' selection based on social-network information yielded remarkably more accurate results than those obtained with non-socially-connected neighbours. All these approaches have been developed with the aim of improving the accuracy of collaborative filtering. ...
Digital content production and distribution has radically changed our business models. An unprecedented volume of supply is now on offer, whetted by the demand of millions of users from all over the world. Since users cannot be expected to browse through millions of different items to find what they might like, filtering has become a popular technique to connect supply and demand: trusted users are first identified, and their opinions are then used to create recommendations. In this domain, users' trustworthiness has been measured according to one of the following two criteria: taste similarity (i.e., “I trust those who agree with me”), or social ties (i.e., “I trust my friends, and the people that my friends trust”). The former criterion aims at identifying concordant users, but is subject to abuse by malicious behaviors. The latter aims at detecting well-intentioned users, but fails to capture the natural subjectivity of tastes. In this article, we propose a new definition of trusted recommenders, addressing those users that are both well-intentioned and concordant. Based on this characterisation, we propose a novel approach to information filtering that we call dependable filtering. We describe alternative algorithms realizing this approach, and demonstrate, by means of extensive performance evaluation on a variety of real large-scale datasets, the high degree of both accuracy and robustness they entail.
Ecommerce systems employ recommender systems to enhance the customer loyalty and hence increasing the cross-selling of products. In collaborative filtering—as the most popular method in recommender systems—an implicit network is formed among all the people. In any network, there are some individuals who have some inspirational power over the others leading them to influence their decisions and behaviours. But it seems that these methods do not support context awareness in mobile commerce environments. Furthermore, they lack high accuracy and also require high volume of computations due to not distinguish between neighbours as a friend or a stranger. This paper proposes a new model for recommender systems which are based on mobile data. This model uses these data to extract current users' context and also to identify individuals with the highest influence. Then, the system uses the information of these identified impressive users in the current context existed in the social networks for making recommendations. Beside of achieving higher accuracy, the proposed model has resolved cold start problem in collaborative filtering systems.
There are an increasing number of health-related communities and forums developed on the Internet, where people discuss certain health issues and exchange social support with each other. However, due to the huge amount and loose structure of user-generated content in the health communities, it is difficult for users to find relevant topics or peers to discuss with. In this paper, we focus on an online smoking cessation forum, QuitStop. We extract user discussion content from the forum, apply machine learning technology to classify posts in the forum, and develop recommendation techniques to help users find valuable topics. Using text and health feature sets, the classifiers are developed and optimized to categorize posts in terms of user intentions and social support types. The recommender systems are then developed to make a recommendation of posts to users, in which the classification results are incorporated in the neighbor-based collaborative filtering approach. It is found that the combination of text and health feature sets can achieve satisfactory classification result. Integrating classification result could help relieving cold start problem in the recommendation. It can greatly improve the recall of recommendation when limited knowledge is known for a thread.
The port-starboard ambiguity in the conventional single towed linear array sonar is one of the most deceiving obstacles for the spatial spectrum estimation. In order to improve the performance of target detection and Direction of Arrival (DOA) estimation, a spatial spectrum sparse representation model is established using multiple beam-space measurements. The forward-backward operator splitting method is adopted for solving the sparse optimization problem. Simulation results demonstrate that, compared with Conventional Beam Forming (CBF) method, the proposed method has evident advantages in ambiguity suppression and estimation performance.
Digital content production and distribution has radically changed our business models. An unprecedented volume of supply is now on offer, whetted by the demand of millions of users from all over the world. Since users cannot be expected to browse through millions of different items to find what they might like, filtering has become a popular technique to connect supply and demand: trusted users are first identified, and their opinions are then used to create recommendations. In this domain, users' trustworthiness has been measured according to one of the following two criteria: taste similarity (i.e., “I trust those who agree with me”), or social ties (i.e., “I trust my friends, and the people that my friends trust”). The former criterion aims at identifying concordant users, but is subject to abuse by malicious behaviors. The latter aims at detecting well-intentioned users, but fails to capture the natural subjectivity of tastes. In this article, we propose a new definition of trusted recommenders , addressing those users that are both well-intentioned and concordant. Based on this characterisation, we propose a novel approach to information filtering that we call dependable filtering . We describe alternative algorithms realizing this approach, and demonstrate, by means of extensive performance evaluation on a variety of real large-scale datasets, the high degree of both accuracy and robustness they entail.
In a Web dominated by social media for information, relationships and communication, the established dynamics between content, people and technology change radically. Given the relevance of the user-generated content in such a scenario and its essentially relational nature, successfully locating the best sources of information requires the development of recommender systems that incorporate the social characteristic of scenarios built on a network that goes beyond the original Internet. The article provides a review of current approaches to the process of recommendation, placing them in the context of consolidated trends associated with the phenomenon of Social Computing, and highlights some lines of development in the area of redefining the problem of recommendation in a context dominated by social networks and user-generated content.
The rapid development of internet has brought us into the era of information explosion, which brings the widespread application of the personalized recommendation system. Collaborative filtering recommendation algorithm is the most widely used algorithms in the personalized recommended system, but it faces problems like the data sparsity, cold start," idler" attack. with the development of social network, many e-commerce sites, social network sites introduce the trust mechanism, which becomes the new approach to overcome the problems of the traditional collaborative filtering algorithms. This paper first introduces the collaborative filtering recommendation algorithm and the existing problems, and then summarizes the current trust based collaborative filtering algorithms. for the trust based collaborative filtering algorithms, this paper summarizes the trust expressions and metrics, and focus on analyzing the representative trust models.
Peer-to-peer (p2p) networks are used by millions for searching and downloading content. Recently, clustering algorithms were shown to be useful for helping users find content in large networks. Yet, many of these algorithms overlook the fact that p2p networks follow graph models with a power-law node degree distribution. This paper studies the obtained clusters when applying clustering algorithms on power-law graphs and their applicability for finding content. Driven by the observed deficiencies, a simple yet efficient clustering algorithm is proposed, which targets a relaxed optimization of a minimal distance distribution of each cluster with a size balancing scheme. A comparative analysis using a song-similarity graph collected from 1.2 million Gnutella users reveals that commonly used efficiency measures often overlook search and recommendation applicability issues and provide the wrong impression that the resulting clusters are well suited for these tasks. We show that the proposed algorithm performs well on various measures that are well suited for the domain.
Transitive trust propagation on webs of trust (social networks representing trust relationships between individuals) is a well-known pattern used to compute reputation, to defend against Sybil attacks and to introduce incentives to cooperation in peer-to-peer applications. Based on the fact that it is dicult for an attacker to get trusted by many honest users, it is possible to prove that some methods based on transitive trust propagation are indeed resilient to attacks mounted by malicious users. The other side of the coin is that honest users can be mistaken for malicious ones: such an event is likely to happen when the web of trust has a high mixing time. Performing a random walk over the web of trust, mixing time is the number of steps needed to ensure that the landing point of the random walk does not depend on the starting node. To help assessing the eectiveness of methods based on transitive trust propagation, we measure the mixing time of four large and publicly available webs of trust. Our experimental results show that mixing time is not directly related with well-known characteristics such as degree distribution or clustering coecient. Rather, they suggest that mixing time is an interesting metrics on its own, explained by the presence or absence of long-range links in the social network.
New generation wireless sensor networks are demanding in in-time updating. Traditional trust evaluation is based entities and needs lengthy time to establish. The data security would be neglected in entity-based sensor networks. False data is another problem because it is hard to be filtered in these entity-based trust mechanisms. The propagation of false data and redundant information wastes a lot of system energy. This paper proposes a new notion to address these challenges: data-centric trust evaluation mechanism (DTSN) and a new method: Proof-of-Reputation-Relevance (PoRR) to realize DTSN. This trust evaluation mechanism protects both entities and data via authentic consensus on event reports and aggregate related reports’ trust via DST. Then DTSN decision logic makes a decision according to the output of DST. Performance evaluation and security analysis confirm the efficiency and security of the proposed mechanism.
Consumers who are close to one another in a social network often make similar purchase decisions. This similarity can result from latent homophily or social influence, as well as common exogenous factors. Latent homophily means consumers who are connected to one another are likely to have similar characteristics and product preferences. Social influence refers to the ability of one consumer to directly influence another consumer's decision based upon their communication. We present an empirical study of purchases of caller ring-back tones using data from an Asian mobile network that predicts consumers' purchase timing and choice decisions. We simultaneously measure latent homophily and social influence, while also accounting for exogenous factors. Identification is achieved due to our dynamic, panel data structure and the availability of detailed communication data. We find strong influence effects and latent homophily effects in both the purchase timing and product choice decisions of consumers.
This paper was accepted by Sandra Slaughter, information systems.
Collaborative filtering or recommender systems use a database about user
preferences to predict additional topics or products a new user might like. In
this paper we describe several algorithms designed for this task, including
techniques based on correlation coefficients, vector-based similarity
calculations, and statistical Bayesian methods. We compare the predictive
accuracy of the various methods in a set of representative problem domains. We
use two basic classes of evaluation metrics. The first characterizes accuracy
over a set of individual predictions in terms of average absolute deviation.
The second estimates the utility of a ranked list of suggested items. This
metric uses an estimate of the probability that a user will see a
recommendation in an ordered list. Experiments were run for datasets associated
with 3 application areas, 4 experimental protocols, and the 2 evaluation
metrics for the various algorithms. Results indicate that for a wide range of
conditions, Bayesian networks with decision trees at each node and correlation
methods outperform Bayesian-clustering and vector-similarity methods. Between
correlation and Bayesian networks, the preferred method depends on the nature
of the dataset, nature of the application (ranked versus one-by-one
presentation), and the availability of votes with which to make predictions.
Other considerations include the size of database, speed of predictions, and
learning time.
this article, we propose to deal with this sparsity problem by applying an associative retrieval framework and related spreading activation algorithms to explore transitive associations among consumers through their past transactions and feedback. Such transitive associations are a valuable source of information to help infer consumer interests and can be explored to deal with the sparsity problem. To evaluate the effectiveness of our approach, we have conducted an experimental study using a data set from an online bookstore. We experimented with three spreading activation algorithms including a constrained Leaky Capacitor algorithm, a branch-and-bound serial symbolic search algorithm, and a Hopfield net parallel relaxation search algorithm. These algorithms were compared with several collaborative filtering approaches that do not consider the transitive associations: a simple graph search approach, two variations of the user-based approach, and an item-based approach. Our experimental results indicate that spreading activation-based approaches significantly outperformed the other collaborative filtering methods as measured by recommendation precision, recall, the F-measure, and the rank score. We also observed the over-activation effect of the spreading activation approach, that is, incorporating transitive associations with past transactional data that is not sparse may "dilute" the data used to infer user preferences and lead to degradation in recommendation performance
A fundamental open question in the analysis of social networks is to understand the interplay between similarity and social ties. People are similar to their neighbors in a social network for two distinct reasons: first, they grow to resemble their current friends due to social influence; and second, they tend to form new links to others who are already like them, a process often termed selection by sociologists. While both factors are present in everyday social processes, they are in tension: social influence can push systems toward uniformity of behavior, while selection can lead to fragmentation. As such, it is important to understand the relative effects of these forces, and this has been a challenge due to the difficulty of isolating and quantifying them in real settings.
We develop techniques for identifying and modeling the interactions between social influence and selection, using data from online communities where both social interaction and changes in behavior over time can be measured. We find clear feedback effects between the two factors, with rising similarity between two individuals serving, in aggregate, as an indicator of future interaction -- but with similarity then continuing to increase steadily, although at a slower rate, for long periods after initial interactions. We also consider the relative value of similarity and social influence in modeling future behavior. For instance, to predict the activities that an individual is likely to do next, is it more useful to know the current activities of their friends, or of the people most similar to them?
Recommender systems apply knowledge discovery techniques to the problem of making personalized recommendations for information, products or services during a live interaction. These systems, especially the k-nearest neighbor collaborative filtering based ones, are achieving widespread success on the Web. The tremendous growth in the amount of available information and the number of visitors to Web sites in recent years poses some key challenges for recommender systems. These are: producing high quality recommendations, performing many recommendations per second for millions of users and items and achieving high coverage in the face of data sparsity. In traditional collaborative filtering systems the amount of work increases with the number of participants in the system. New recommender system technologies are needed that can quickly produce high quality recommendations, even for very large-scale problems. To address these issues we have explored item-based collaborative filtering techniques. Itembased techniques first analyze the user-item matrix to identify relationships between different items, and then use these relationships to indirectly compute recommendations for users. In this paper we analyze different item-based recommendation generation algorithms. We look into different techniques for computing item-item similarities (e.g., item-item correlation vs. cosine similarities between item vectors) and different techniques for obtaining recommendations from them (e.g., weighted sum vs. regression model). Finally, we experimentally evaluate our results and compare them to the basic k-nearest neighbor approach. Our experiments suggest that item-based algorithms provide dramatically better performance than user-based algorithms, while at the same time p...
Recommender systems attempt to reduce information overload and retain customers by selecting a subset of items from a universal set based on user preferences. While research in recommender systems grew out of information retrieval and filtering, the topic has steadily advanced into a legitimate and challenging research area of its own. Recommender systems have traditionally been studied from a content-based filtering vs. collaborative design perspective. Recommendations, however, are not delivered within a vacuum, but rather cast within an informal community of users and social context. Therefore, ultimately all recommender systems make connections among people and thus should be surveyed from such a perspective. This viewpoint is under-emphasized in the recommender systems literature. We therefore take a connection-oriented viewpoint toward recommender systems research. We posit that recommendation has an inherently social element and is ultimately intended to connect people either directly as a result of explicit user modeling or indirectly through the discovery of relationships implicit in extant data. Thus, recommender systems are characterized by how they model users to bring people together: explicitly or implicitly. Finally, user modeling and the connection-centric viewpoint raise broadening and social issues--such as evaluation, targeting, and privacy and trust--which we also briefly address.
Identifying Social Effects Using Networked ECommerce Data. SCECR-08 Murdoch's MySpace Makes E-Commerce Debut
- Oestreicher
- G Singer
- A Sundararajan
Oestreicher-Singer, G. and A. Sundararajan (2008). Identifying Social Effects Using Networked ECommerce Data. SCECR-08. r23 Olsen, P./ (2006) Murdoch's MySpace Makes E-Commerce Debut, Forbes.com , May 15, 2006 accessed 9/22/2008