Jonathan Gemmell

Personalized recommendation in social tagging systems using hierarchical clustering

Collaborative tagging applications allow Internet users to annotate resources with personalized tags. The complex network created by many annotations, often called a folksonomy, permits users the freedom to explore tags, resources or even other users profiles unbound from a rigid predefined conceptual hierarchy. However, the freedom afforded users comes at a cost: an uncontrolled vocabulary can result in tag redundancy and ambiguity hindering navigation. Data mining techniques, such as clustering, provide a means to remedy these problems by identifying trends and reducing noise. Tag clusters can also be used as the basis for effective personalized recommendation assisting users in navigation. We present a personalization algorithm for recommendation in folksonomies which relies on hierarchical tag clusters. Our basic recommendation framework is independent of the clustering method, but we use a context-dependent variant of hierarchical agglomerative clustering which takes into account the users current navigation context in cluster selection. We present extensive experimental results on two real world dataset. While the personalization algorithm is successful in both cases, our results suggest that folksonomies encompassing only one topic domain, rather than many topics, present an easier target for recommendation, perhaps because they are more focused and often less sparse. Furthermore, context dependent cluster selection, an integral step in our personalization algorithm, demonstrates more utility for recommendation in multi-topic folksonomies than in single-topic folksonomies. This observation suggests that topic selection is an important strategy for recommendation in multi-topic folksonomies.

Resource recommendation in social annotation systems: A linear-weighted hybrid approach

Social annotation systems enable the organization of online resources with user-defined keywords. Collectively these annotations provide a rich information space in which users can discover resources, organize and share their finds, and connect to other users with similar interests. However, the size and complexity of these systems can lead to information overload and reduced utility for users. For these reasons, researchers have sought to apply the techniques of recommender systems to deliver personalized views of social annotation systems. To date, most efforts have concentrated on the problem of tag recommendation - personalized suggestions for possible annotations. Resource recommendation has not received the same systematic evaluation, in part because the task is inherently more complex. In this article, we provide a general formulation for the problem of resource recommendation in social annotation systems that captures these variants, and we evaluate two cases: basic resource recommendation and tag-specific resource recommendation. We also propose a linear-weighted hybrid framework for resource recommendation. Using six real-world datasets, we show that its integrative approach is essential for this recommendation task and provides the most adaptability given the varying data characteristics in different social annotation systems. We find that our algorithm is more effective than other more mathematically-complex techniques and has the additional advantages of flexibility and extensibility.

The impact of ambiguity and redundancy on tag recommendation in folksonomies

Collaborative tagging applications have become a popular tool allowing Internet users to manage online resources with tags. Most collaborative tagging applications permit unsupervised tagging resulting in tag ambiguity in which a single tag has many different meanings and tag redundancy in which several tags have the same meaning. Common metrics for evaluating tag recommenders may overestimate the utility of ambiguous tags or ignore the appropriateness of redundant tags. Ambiguity and redundancy may even burden the user with additional effort by requiring them to clarify an annotation or forcing them to distinguish between highly related items. In this paper we demonstrate that ambiguity and redundancy impede the evaluation and performance of tag recommenders. Five tag recommendation strategies based on popularity, collaborative filtering and link analysis are explored. We use a cluster-based approach to define ambiguity and redundancy and provide extensive evaluation on three real world datasets.

Tag-based resource recommendation in social annotation applications

Social annotation systems enable the organization of online resources with user-defined keywords. The size and complexity of these systems make them excellent platforms for the application of recommender systems, which can provide personalized views of complex information spaces. Many researchers have concentrated on the important problem of tag recommendation. Less attention has been paid to the recommendation of resources in the context of social annotation systems. In this paper, we examine the specific case of tag-based resource recommendation and propose a linear-weighted hybrid for the task. Using six real world datasets, we show that our algorithm is more effective than other more mathematically complex techniques.

Evaluating the Impact of Attacks in Collaborative Tagging Environments

Abstract—The proliferation of social web technologies such as collaborative tagging has led to an increasing awareness of their vulnerability to misuse. Attackers may attempt to distort the system’s adaptive behavior by inserting erroneous or misleading annotations, thus altering the way in which information is presented to legitimate users. Prior work on recommender systems has shown that studying different attack types, their properties and their impact, can help identify robust algorithms that make these systems more secure and less vulnerable to manipulation.Unlike traditional recommender systems, a tagging systemincludes multiple retrieval algorithms to facilitate browsing of resources, users and tags. The challenge is, therefore, evaluating the impact of various types of attacks across different navigation options. In this paper we develop a framework for characterizingattacks against tagging systems. We then propose a methodology for evaluating their global impact based on PageRank. Using real data from a popular tagging systems, we empirically evaluate the effectiveness of several attack types. Our results help us understand how much effort is needed from an attacker to change the behavior of a tagging system and which attack types are more successful against such systems.

Resource Recommendation in Collaborative Tagging Applications

Collaborative tagging applications enable users to annotate online resources with user-generated keywords. The collection of these annotations and the way they connect users and resources produce a rich information space for users to explore. However the size, complexity and chaotic structure of these systems hamper users as they search for information. Recommenders can assist the user by suggesting resources, tags or even other users. Previous work has demonstrated that an integrative approach which exploits all three dimensions of the data (users, resources, tags) produce superior results in tag recommendation. We extend this integrative philosophy to resource recommendation. Specifically, we propose an approach for designing weighted linear hybrid resource recommenders. Through extensive experimentation on two large real world datasets, we show that the hybrid recommenders surpass the effectiveness of their constituent components while inheriting their simplicity, computational efficiency and explanatory capacity. We further introduce the notion of information channels which describe the interaction of the three dimensions. Information channels can be used to explain the effectiveness of individual recommenders or explain the relative contribution of components in the hybrid recommender.

Infusing Collaborative Recommenders with Distributed Representations

Recommender systems assist users in navigating complex information spaces and focus their attention on the content most relevant to their needs. Often these systems rely on user activity or descriptions of the content. Social annotation systems, in which users collaboratively assign tags to items, provide another means to capture information about users and items. Each of these data sources provides unique benefits, capturing different relationships. In this paper, we propose leveraging multiple sources of data: ratings data as users report their affinity toward an item, tagging data as users assign annotations to items, and item data collected from an online database. Taken together, these datasets provide the opportunity to learn rich distributed representations by exploiting recent advances in neural network architectures. We first produce representations that subjectively capture interesting relationships among the data. We then empirically evaluate the utility of the representations to predict a users rating on an item and show that it outperforms more traditional representations. Finally, we demonstrate that traditional representations can be combined with representations trained through a neural network to achieve even better results.

Recommendation in the Social Web

Recommender systems are a means of personalizing the presentation of information to ensure that users see the items most relevant to them. The social web has added new dimensions to the way people interact on the Internet, placing the emphasis on user-generated content. Users in social networks create photos, videos and other artifacts, collaborate with other users, socialize with their friends and share their opinions online. This outpouring of material has brought increased attention to recommender systems, as a means of managing this vast universe of content. At the same time, the diversity and complexity of the data has meant new challenges for researchers in recommendation. This article describes the nature of recommendation research in social web applications and provides some illustrative examples of current research directions and techniques. It is difficult to overstate the impact of the social web. This new breed of social applications is reshaping nearly every human activity from the way people watch movies to how they overthrow governments. Facebook allows its members to maintain friendships whether they live next door or on another continent. With Twitter, users from celebrities to ordinary folks can launch their 140 character messages out to a diverse horde of ‘‘followers.” Flickr and YouTube users upload their personal media to share with the world, while Wikipedia editors collaborate on the world’s largest encyclopedia.

Recommendation by Example in Social Annotation Systems

Recommendation by example is common in contemporary Internet applications providing resources similar to a user-selected example. In this paper this task is considered as a function available within a social annotation system offering new ways to model both users and resources. Using three real-world datasets we motivate several conclusions. First, a personalized approach outperforms non-personalized approaches suggesting that users perceive the similarity between resources differently. Second, the manner in which users interact with social annotation systems vary producing datasets with variable characteristics and requiring different recommendation strategies to best satisfy their needs. Third, a hybrid recommender constructed from several component recommenders can produce superior results by exploiting multiple dimensions of the data. The hybrid remains powerful, flexible and extensible despite the underlying characteristics of the data.

User Partitioning Hybrid for Tag Recommendation

Tag recommendation is a fundamental service in today’s social annotation systems, assisting users as they collect and annotate resources. Our previous work has demonstrated the strengths of a linear weighted hybrid, which weights and combines the results of simple components into a final recommendation. However, these previous efforts treated each user the same. In this work, we extend our approach by automatically discovering partitions of users. The user partitioning hybrid learns a different set of weights for these user partitions. Our rigorous experimental results show a marked improvement. Moreover, analysis of the partitions within a dataset offers interesting insights into how users interact with social annotations systems.