In this paper, we present the factorization model PITF (Pairwise Interaction Tensor Factorization) which is a special case of the TD model with linear runtime both for learning and prediction. PITF explicitly models the pairwise interactions between users, items and tags. The model is learned with an adaption of the Bayesian personalized ranking (BPR) criterion which originally has been introduced for item recommendation. Empirically, we show on real world datasets that this model outperforms TD largely in runtime and even can achieve better prediction quality. Besides our lab experiments, PITF has also won the ECML/PKDD Discovery Challenge 2009 for graph-based tag recommendation.}, acmid = {1718498}, address = {New York, NY, USA}, author = {Rendle, Steffen and Schmidt-Thieme, Lars}, booktitle = {Proceedings of the third ACM international conference on Web search and data mining}, doi = {10.1145/1718487.1718498}, interhash = {ce8fbdf2afb954579cdb58104fb683a7}, intrahash = {10fe730b391b08031f3103f9cdbb6e1a}, isbn = {978-1-60558-889-6}, location = {New York, New York, USA}, numpages = {10}, pages = {81--90}, publisher = {ACM}, title = {Pairwise interaction tensor factorization for personalized tag recommendation}, url = {http://doi.acm.org/10.1145/1718487.1718498}, year = 2010 } @inproceedings{musto2010combining, abstract = {The explosion of collaborative platforms we are recently witnessing, such as social networks, or video and photo sharing sites, radically changed the Web dynamics and the way people use and organize information. The use of tags, keywords freely chosen by users for annotating resources, offers a new way for organizing and retrieving web resources that closely reflects the users' mental model and also allows the use of evolving vocabularies. However, since tags are handled in a purely syntactical way, the annotations provided by users generate a very sparse and noisy tag space that limits the effectiveness of tag-based approaches for complex tasks. Consequently, systems called tag recommenders recently emerged, with the purpose of speeding up the so-called tag convergence, providing users with the most suitable tags for the resource to be annotated. This paper presents a tag recommender system called STaR (Social Tag Recommender), which extends the social approach presented in a previous work [14] with a content-based approach able to extract tags directly from the textual content of HTML pages. Results of experiments carried out on a large dataset gathered from Bibsonomy, show that the use of content-based techniques improves the predictive accuracy of the tag recommender. }, address = {Berlin/Heidelberg}, author = {Musto, Cataldo and Narducci, Fedelucio and Lops, Pasquale and de Gemmis, Marco}, booktitle = {E-Commerce and Web Technologies}, doi = {10.1007/978-3-642-15208-5_2}, editor = {Buccafurri, Francesco and Semeraro, Giovanni}, interhash = {60254c70491f83c365ee71b019d65344}, intrahash = {bdd023e357c901c749580d038b4f2059}, isbn = {978-3-642-15207-8}, pages = {13--23}, publisher = {Springer}, series = {Lecture Notes in Business Information Processing}, title = {Combining Collaborative and Content-Based Techniques for Tag Recommendation.}, url = {http://dx.doi.org/10.1007/978-3-642-15208-5_2}, volume = 61, year = 2010 } @inproceedings{rae2010improving, abstract = {In this paper we address the task of recommending additional tags to partially annotated media objects, in our case images. We propose an extendable framework that can recommend tags using a combination of different personalised and collective contexts. We combine information from four contexts: (1) all the photos in the system, (2) a user's own photos, (3) the photos of a user's social contacts, and (4) the photos posted in the groups of which a user is a member. Variants of methods (1) and (2) have been proposed in previous work, but the use of (3) and (4) is novel.
For each of the contexts we use the same probabilistic model and Borda Count based aggregation approach to generate recommendations from different contexts into a unified ranking of recommended tags. We evaluate our system using a large set of real-world data from Flickr. We show that by using personalised contexts we can significantly improve tag recommendation compared to using collective knowledge alone. We also analyse our experimental results to explore the capabilities of our system with respect to a user's social behaviour.}, address = {Paris, France}, author = {Rae, Adam and Sigurbjörnsson, Börkur and van Zwol, Roelof}, booktitle = {Adaptivity, Personalization and Fusion of Heterogeneous Information}, interhash = {2595ff47e852a64c7f1c88b915c7e9ad}, intrahash = {98034c615577fd3558fd326fbe03f894}, location = {Paris, France}, pages = {92--99}, publisher = {Le Centre De Hautes Etudes Internationales d'Informatique Documentaire}, series = {RIAO '10}, title = {Improving tag recommendation using social networks}, url = {http://portal.acm.org/citation.cfm?id=1937055.1937077}, year = 2010 } @book{jaeschke2011formal, abstract = {One of the most noticeable innovation that emerged with the advent of the Web 2.0 and the focal point of this thesis are collaborative tagging systems. They allow users to annotate arbitrary resources with freely chosen keywords, so called tags. The tags are used for navigation, finding resources, and serendipitous browsing and thus provide an immediate benefit for the user. By now, several systems for tagging photos, web links, publication references, videos, etc. have attracted millions of users which in turn annotated countless resources. Tagging gained so much popularity that it spread into other applications like web browsers, software packet managers, and even file systems. Therefore, the relevance of the methods presented in this thesis goes beyond the Web 2.0. The conceptual structure underlying collaborative tagging systems is called folksonomy. It can be represented as a tripartite hypergraph with user, tag, and resource nodes. Each edge of the graph expresses the fact that a user annotated a resource with a tag. This social network constitutes a lightweight conceptual structure that is not formalized, but rather implicit and thus needs to be extracted with knowledge discovery methods. In this thesis a new data mining task – the mining of all frequent tri-concepts – is presented, together with an efficient algorithm for discovering such implicit shared conceptualizations. Our approach extends the data mining task of discovering all closed itemsets to three-dimensional data structures to allow for mining folksonomies. Extending the theory of triadic Formal Concept Analysis, we provide a formal definition of the problem, and present an efficient algorithm for its solution. We show the applicability of our approach on three large real-world examples and thereby perform a conceptual clustering of two collaborative tagging systems. Finally, we introduce neighborhoods of triadic concepts as basis for a lightweight visualization of tri-lattices. The social bookmark and publication sharing system BibSonomy, which is currently among the three most popular systems of its kind, has been developed by our research group. Besides being a useful tool for many scientists, it provides interested researchers a basis for the evaluation and integration of their knowledge discovery methods. This thesis introduces BibSonomy as an exemplary collaborative tagging system and gives an overview of its architecture and some of its features. Furthermore, BibSonomy is used as foundation for evaluating and integrating some of the discussed approaches. Collaborative tagging systems usually include tag recommendation mechanisms easing the process of finding good tags for a resource, but also consolidating the tag vocabulary across users. In this thesis we evaluate and compare several recommendation algorithms on large-scale real-world datasets: an adaptation of user-based Collaborative Filtering, a graph-based recommender built on top of the FolkRank algorithm, and simple methods based on counting tag co-occurences. We show that both FolkRank and Collaborative Filtering provide better results than non-personalized baseline methods. Moreover, since methods based on counting tag co-occurrences are computationally cheap, and thus usually preferable for real time scenarios, we discuss simple approaches for improving the performance of such methods. We demonstrate how a simple recommender based on counting tags from users and resources can perform almost as good as the best recommender. Furthermore, we show how to integrate recommendation methods into a real tagging system, record and evaluate their performance by describing the tag recommendation framework we developed for BibSonomy. With the intention to develop, test, and evaluate recommendation algorithms and supporting cooperation with researchers, we designed the framework to be easily extensible, open for a variety of methods, and usable independent from BibSonomy. We also present an evaluation of the framework which demonstrates its power. The folksonomy graph shows specific structural properties that explain its growth and the possibility of serendipitous exploration. Clicklogs of web search engines can be represented as a folksonomy in which queries are descriptions of clicked URLs. The resulting network structure, which we will term logsonomy is very similar to the one of folksonomies. In order to find out about its properties, we analyze the topological characteristics of the tripartite hypergraph of queries, users and bookmarks on a large folksonomy snapshot and on query logs of two large search engines. We find that all of the three datasets exhibit similar structural properties and thus conclude that the clicking behaviour of search engine users based on the displayed search results and the tagging behaviour of collaborative tagging users is driven by similar dynamics. In this thesis we further transfer the folksonomy paradigm to the Social Semantic Desktop – a new model of computer desktop that uses Semantic Web technologies to better link information items. There we apply community support methods to the folksonomy found in the network of social semantic desktops. Thus, we connect knowledge discovery for folksonomies with semantic technologies. Alltogether, the research in this thesis is centered around collaborative tagging systems and their underlying datastructure – folksonomies – and thereby paves the way for the further dissemination of this successful knowledge management paradigm. }, address = {Heidelberg, Germany}, author = {Jäschke, Robert}, interhash = {dcb2cd1cd72ae45d77c4d8755d199405}, intrahash = {9db90c2ff04f514ada9f6b50fde46065}, isbn = {978-3-89838-332-5}, month = jan, publisher = {Akademische Verlagsgesellschaft AKA}, series = {Dissertationen zur Künstlichen Intelligenz}, title = {Formal Concept Analysis and Tag Recommendations in Collaborative Tagging Systems}, url = {http://www.aka-verlag.com/de/detail?ean=978-3-89838-332-5}, vgwort = {413}, volume = 332, year = 2011 } @inproceedings{sarwar2001item, abstract = {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. Item-based 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 providing better quality than the best available user-based algorithms.}, address = {New York, NY, USA}, author = {Sarwar, Badrul and Karypis, George and Konstan, Joseph and Riedl, John}, booktitle = {WWW '01: Proceedings of the 10th International Conference on World Wide Web}, doi = {10.1145/371920.372071}, interhash = {043d1aaba0f0b8c01d84edd517abedaf}, intrahash = {a6461157c8102d34b8001c7d33a42684}, isbn = {1-58113-348-0}, location = {Hong Kong}, pages = {285--295}, publisher = {ACM}, title = {Item-based collaborative filtering recommendation algorithms}, url = {http://portal.acm.org/citation.cfm?id=372071}, year = 2001 } @article{naaman2008zonetag, abstract = {We describe ZoneTag, a camera phone application that allows users to capture, annotate, and share photos directly from their phone.}, address = {Los Alamitos, CA, USA}, author = {Naaman, Mor and Nair, Rahul}, doi = {http://doi.ieeecomputersociety.org/10.1109/MMUL.2008.69}, editor = {Panchanathan, Sethuraman}, interhash = {0eaa336bc425a2710aefe9b2a4f3711d}, intrahash = {3a6067d3abc1cfdb9b59aa8e222ea664}, issn = {1070-986X}, journal = {IEEE MultiMedia}, number = 3, pages = {34-40}, publisher = {IEEE Computer Society}, title = {ZoneTag's Collaborative Tag Suggestions: What is This Person Doing in My Phone?}, volume = 15, year = 2008 } @inproceedings{tso2008tag, abstract = {Recommender Systems (RS) aim at predicting items or ratings of items that the user are interested in. Collaborative Filtering (CF) algorithms such as user- and item-based methods are the dominant techniques applied in RS algorithms. To improve recommendation quality, metadata such as content information of items has typically been used as additional knowledge. With the increasing popularity of the collaborative tagging systems, tags could be interesting and useful information to enhance RS algorithms. Unlike attributes which are "global" descriptions of items, tags are "local" descriptions of items given by the users. To the best of our knowledge, there hasn't been any prior study on tag-aware RS. In this paper, we propose a generic method that allows tags to be incorporated to standard CF algorithms, by reducing the three-dimensional correlations to three two-dimensional correlations and then applying a fusion method to re-associate these correlations. Additionally, we investigate the effect of incorporating tags information to different CF algorithms. Empirical evaluations on three CF algorithms with real-life data set demonstrate that incorporating tags to our proposed approach provides promising and significant results.}, address = {New York, NY, USA}, author = {Tso-Sutter, Karen H. L. and Marinho, Leandro Balby and Schmidt-Thieme, Lars}, booktitle = {SAC '08: Proceedings of the 2008 ACM symposium on Applied computing}, doi = {http://doi.acm.org/10.1145/1363686.1364171}, interhash = {61f74fe4bb3a72220c69438010ae9962}, intrahash = {792034671682f8720177801e2729d4c7}, isbn = {978-1-59593-753-7}, location = {Fortaleza, Ceara, Brazil}, pages = {1995--1999}, publisher = {ACM}, title = {Tag-aware recommender systems by fusion of collaborative filtering algorithms}, url = {http://portal.acm.org/citation.cfm?id=1364171}, year = 2008 } @inproceedings{mishne06-autotag, abstract = {This paper describes AutoTag, a tool which suggests tags for weblog posts using collaborative filtering methods. An evaluation of AutoTag on a large collection of posts shows good accuracy; coupled with the blogger's final quality control, AutoTag assists both in simplifying the tagging process and in improving its quality.}, address = {New York, NY, USA}, author = {Mishne, Gilad}, booktitle = {WWW '06: Proceedings of the 15th International Conference on World Wide Web}, interhash = {62d370a3abfaed3469bd2ee6a8ef6473}, intrahash = {4094419a2f26ac85175ad48207311d61}, note = {paper presented at the poster track}, pages = {953--954}, publisher = {ACM Press}, title = {AutoTag: a collaborative approach to automated tag assignment for weblog posts}, url = {http://portal.acm.org/citation.cfm?id=1135777.1135961}, year = 2006 } @techreport{Vojnovic2007ranking, abstract = {We consider collaborative tagging systems where users can attach tags to information objects. Such systems are widely used to add keywords meta-data to photos, videos, or web pages (social bookmarking applications). The meta-data is then used by information retrieval mechanism to provide accurate query answers. To that end, the goal of collaborative tagging systems is to quickly discover the true ranking of a tag for an information object with respect to a given ranking criteria. In this paper, we consider the popularity rank as a ranking criteria. Many collaborative tagging systems help users tagging of an object by making suggestions based on the tagging history of an information object. The problem with making tag suggestions is that they may reinforce some tags and the system fails to discover the true popularity rank of a tag for a given information object. To investigate this issue, we propose and study several algorithms for ranking and suggesting tags in collaborative tagging systems where we focus on the following design objectives: (a) learn true popularity rank, (b) make relevant suggestions, and (c) learn fast. We find that simple incremental updates of the suggestion set, which besides suggestion set size require no configuration constants, offer good performance. Performance is evaluated by analysis and numerical results for which we used a dataset of complete tagging histories of urls that we crawled from a popular social bookmarking web service over a month period.}, author = {Vojnovic, M. and Cruise, J. and Gunawardena, D. and Marbach, P.}, date = {Feb 2007}, howpublished = {MSR-TR-2007-06}, institution = {Microsoft Research}, interhash = {687129a4106fab9a2cd1c032ae52fc73}, intrahash = {b2ca76b60ffbd28b55f3707085986e7a}, number = {MSR-TR-2007-06}, tech = {TR-2007-06}, title = {Ranking and Suggesting Tags in Collaborative Tagging Applications}, url = {http://research.microsoft.com/~milanv/MSR-TR-2007-06.pdf}, year = 2007 }