Tag based recommender system
- 1. Образец заголовка Tag-Based Recommender System by Xiao Xin Li (xli147) Prepared as an assignment for CS410: Text Information Systems in Spring 2016
- 2. Образец заголовкаOverview 1. The Recommender System 2. Traditional Recommendation Methods: definition, pros, and cons 1) Collaborative Filtering 2) Content-based Recommendations 3) Knowledge-based systems 4) Hybrid Approaches 3. Enhance Recommender Systems with User Profiles – Research papers 4. Leveraging Tagging Systems with User Information – Research papers 5. Tutorial Conclusions 6. Acknowledgements
- 3. Образец заголовка The Recommender System
- 4. Образец заголовкаThe Recommender System
- 5. Образец заголовкаThe Recommender System
- 6. Образец заголовкаThe Recommender System • Traditional definition: Estimate a utility function that automatically predicts how a user will like an item. • Based on: – Past behavior – Relations to other users – Item similarity – Context – …
- 7. Образец заголовка Traditional Recommendation Methods • Collaborative Filtering • Content-based Recommendations • Knowledge-based systems • Hybrid Approaches
- 9. Образец заголовкаCollaborative Filtering • Widely used in e-commerce • Find users in a community that share the same interests in the past to predict what the current user will be interested in.
- 11. Образец заголовкаAlgorithms Collaborative Filtering Non-probabilistic Algorithms Probabilistic Algorithms User-based nearest neighbor Item-based nearest neighbor Reducing dimensionality Bayesian-network models EM algorithm
- 12. Образец заголовкаUser-Based CF • A collection of user ui , i=1, …, n and a collection of products pj , j=1, …, m • An n × m matrix of ratings vij , with vij = ? if user i did not rate product j • Prediction for user i and product j is computed • Similarity can be computed by Pearson correlation
- 13. Образец заголовкаUser-Based CF The similarity of Alice to User1 is:
- 15. Образец заголовкаItem-Based CF 1. Look into the items the target user has rated 2. Compute how similar they are to the target item – Similarity only using past ratings from other users 3. Select k most similar items 4. Compute Prediction by taking weighted average on the target user’s ratings on the most similar items
- 16. Образец заголовкаItem Similarity Computation • Cosine-based Similarity (difference in rating scale between users is not taken into account) • Adjusted Cosine Similarity (takes care of difference in rating scale) U = set of users that rated both items a and b
- 17. Образец заголовкаUser-Based CF The cosine similarity of Item5 and Item1 is:
- 18. Образец заголовкаUser-Based CF The adjusted cosine similarity value for Item5 and Item1 is:
- 19. Образец заголовкаMemory-Based CF • Use the entire user-item database to generate a prediction • Usage of statistical techniques to find the neighbors – e.g. nearest-neighbor.
- 20. Образец заголовкаModel-Based CF • First develop a model of user • Type of model: – Probabilistic (e.g. Bayesian Network) – Clustering – Rule-based approaches (e.g. Association Rules) – Classification – Regression – LDA – …
- 21. Образец заголовкаPros & Cons Pros: • Requires minimal knowledge engineering efforts • Users and products are symbols without any internal structure or characteristics • Produces good-enough results in most cases Cons: • Sparsity – evaluation of large itemsets where user/item interactions are under 1% • Scalability - Nearest neighbor require computation that grows with both the number of users and the number of items
- 24. Образец заголовкаContent-Based Recommenders • Recommendations based on content of items rather than on other users’ opinions/interactions • Common for recommending text-based products
- 25. Образец заголовкаSimilarity-Based Retrieval • Nearest Neighbors • Relevance Feedback and Rocchio’s Algorithm • Probabilistic approaches based on Naïve Bayes • Linear classifiers and machine learning • Decision Tree
- 26. Образец заголовкаHow they work? • Items to recommend are “described” by their associated features (e.g. keywords) • User Model structured in a “similar” way as the content: features/keywords more likely to occur in the preferred documents (lazy approach) • The user model can be a classifier based on whatever technique (Neural Networks, Naïve Bayes...)
- 27. Образец заголовкаPros & Cons • Pros – User independence • No cold-start or sparsity – Able to recommend to users with unique tastes – Able to recommend new and unpopular items – Can provide explanations by listing content-features • Cons – Requires content that can be encoded as meaningful features (difficult in some domains/catalogs) – Users represented as learnable function of content features – Difficult to implement serendipity – Easy to overfit (e.g. for a user with few data points)
- 28. Образец заголовкаCF vs. CB CF CB Compare Users interest Item info Similarity Set of users User profile Item info Text document Shortcoming Other users’ feedback matters Coverage Unusual interest Feature matters Over-specialize Eliciting user feedback
- 30. Образец заголовкаKnowledge-Based Systems Explanation subsystem Inference engine Knowledge acquisition subsystem Case specific database Knowledge base User interface Developer's interface User Knowledge engineer
- 31. Образец заголовкаKnowledge-Based Systems • Select items from the catalog that fulfill a set of applicable constraints specified by the user • Two basic types: – Constraint-based – Case-based
- 32. Образец заголовкаPseudocode 1. Users specify the requirements 2. Systems try to identify solutions 3. If no solution can be found, users change requirements
- 33. Образец заголовкаConstraint-Based vs. Case-Based • Case-based: – Based on different types of similarity measures – Retrieve items that are similar to specified requirements • Constraint-based: – Rely on explicitly defined set of rules – Retrieve items that fulfill the rules – Critiquing is an effective way to support navigation in item space to find useful alternatives
- 34. Образец заголовкаPros & Cons • Pros – Cold-start problem doesn’t exist • recommendations are calculated independently of user ratings – Does not have to gather information about a particular user • Judgments are independent of individual tastes • Cons – High cost and effort – The nature of knowledge • Knowledge is specific to the domain • Can not be shared without the presence of expert even the knowledge is available – The level of risk • Development cost is very high • Cost goes higher and higher in maintaining these systems
- 36. Образец заголовка Hybrid Recommender Systems: Survey and Experiments CF-Based Recommender Content-Based Recommender Combiner Reco Input Input
- 37. Образец заголовка Hybrid Recommender Systems: Survey and Experiments • Well-known survey of the design space of different hybrid recommendation algorithms by Robin Burke • Proposes a taxonomy of different classes of recommendation algorithms • Seven different hybridization strategies can be abstracted into three base designs: – Monolithic hybrids – Parallelized hybrids – Pipelined hybrids
- 38. Образец заголовкаMonolithic • Incorporates aspects of several recommendation strategies in one algorithm implementation • Data-specific preprocessing steps are used to transform the input data into a representation that can be exploited by a specific algorithm paradigm • Advantageous if little additional knowledge is available for inclusion on the feature level
- 39. Образец заголовкаMonolithic • Feature combination hybrid – uses a diverse range of input data • Feature augmentation hybrid – integrate several recommendation algorithms
- 40. Образец заголовкаParallelized • Employ several recommenders side by side and employ a specific hybridization mechanism to aggregate their outputs • Least invasive to existing implementations • Act as an additional post-processing step
- 41. Образец заголовкаParallelized • Mixed – combines the results of different recommender systems at the level of the user interface – results from different techniques are presented together. • Weighted – combines the recommendations of two or more recommendation systems by computing weighted sums of their scores. • Switching – require an oracle that decides which recommender should be used in a specific situation, depending on the user profile and/or the quality of recommendation results.
- 42. Образец заголовкаPipelined • Implement a staged process in which several techniques sequentially build one another before the final one produces recommendations for the user • Most ambitious hybridization designs • Require deeper insight into algorithm’s functioning to ensure efficient runtime computations
- 43. Образец заголовкаPipelined • Cascade hybrids – based on a sequenced order of techniques – each succeeding recommender only refines the recommendations of its predecessor • Meta-level hybridization design – one recommender builds a model that is exploited by the principal recommender to make recommendations
- 44. Образец заголовкаSummary Collaborative Filtering Content-based Knowledge-based Hybrid User-Based CF Item-Based CF Memory-Based CF Similarity- Based Retrieval Case-Based Constraint-base Monolithic Parallelized Pipelined Model-Based CF
- 45. Образец заголовка Enhance Recommender Systems with User Profiles
- 46. Образец заголовкаRecommendations Just For You
- 48. Образец заголовкаWhy Using User Profile? • A profile of the user's interests is used by most recommendation systems • Used to provide personalized recommendations • Describes the types of items the user likes • Compares items to the user profile to determine what to recommend • Created and updated automatically in response to feedback on the desirability of items that have been presented to the user
- 49. Образец заголовка Accounting for Taste: Using Profile Similarity to Improve Recommender Systems Philip Bonhard , Clare Harries , John McCarthy , M. Angela S
- 50. Образец заголовкаBackground • User-user collaborative filtering comes closest to emulating real world recommendations – based on user rather than item matching • Recommender system research focus: – Precision effectiveness: tested against the real ratings – Prediction efficiency: computational cost in terms of time and resources for calculating predictions • Recommender systems can be made more effective and usable by appropriating some functionality from social systems Philip Bonhard , Clare Harries , John McCarthy , M. Angela Sasse, Accounting for taste: using profile similarity to improve recommender systems
- 51. Образец заголовкаExperiment • Independent variables: recommender profile characteristics – familiarity, profile similarity, and rating overlap • Dependent variable: choices people make in a recommender system context • Hypotheses and results: 1. Familiar recommenders will be preferred – not supported 2. Similar recommenders will be preferred – overwhelmingly supported 3. Recommenders with high rating overlap will be preferred – supported Philip Bonhard , Clare Harries , John McCarthy , M. Angela Sasse, Accounting for taste: using profile similarity to improve recommender systems
- 52. Образец заголовкаResults Philip Bonhard , Clare Harries , John McCarthy , M. Angela Sasse, Accounting for taste: using profile similarity to improve recommender systems
- 53. Образец заголовкаConclusions • Rating overlap in combination with profile similarity can be a powerful source of information for a decision-maker when judging the validity of a recommendation • Participants were more confident in their choices when the recommender had a high rating overlap with them in combination with a similar profile • Decision-makers trust recommenders more when they have high rating overlap and a similar profile Philip Bonhard , Clare Harries , John McCarthy , M. Angela Sasse, Accounting for taste: using profile similarity to improve recommender systems
- 54. Образец заголовка Leveraging Tagging Systems with User Information
- 55. Образец заголовкаTagging Recommender System
- 56. Образец заголовкаTagging • The process of assigning metadata in the form of keywords to shared content by many users • An important way to provide information about resources on the Web • Enable the organization of information within personal information spaces that can be shared
- 57. Образец заголовкаCollaborative Tagging Systems • Folksonomies • Allow users to tag documents, share their tags, and search for documents based on these tags • Collaborative tagging – tagging of a collection of documents commonly accessible to a large group • Social bookmarking – tagging contents located all over the Web
- 58. Образец заголовкаTag Recommendation • Recommend relevant tags for an untagged user resource • Integrative models that leverage all three dimensions of a social annotation system (users, resources, tags) produce superior results • Various purposes: – Increase the chances of getting a resource annotated – Remind users what a resource is about – Lazy annotation – …
- 59. Образец заголовка Benefits of Collaborative Tagging Systems • Lowers costs – no complicated, hierarchically organized nomenclature to learn • Respond quickly to changes and innovations in the way users categorize content – inherently open-ended • Allow a user to search for the content that the user has tagged using a personal vocabulary • Assist navigation by providing dynamic hyperlinks among tags, documents and users
- 60. Образец заголовка Challenges of Collaborative Tagging Systems • Too much freedom of choice of tags – Polysemy: words having multiple related meanings – Synonymy: multiple words having the same or similar meanings • Challenges in support knowledge management activities in an organization • Challenges in identifying communities of common interest • Challenges in identifying information leaders or domain experts • Lack of a document hierarchy prevents it from being widely adopted by enterprises – Organizations need systematic mechanisms of storing and retrieving documents
- 61. Образец заголовка A Personalized Recommender System Based on Users’ Information In Folksonomies Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo
- 62. Образец заголовкаMotivation • Success of social bookmarking sharing systems – Flickr, Bibsonomy, Youtube, etc. • The users of a folksonomy have different profiles and expectations depending on their motivations • Personalization provides solutions to help users solve the information overload issue Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 63. Образец заголовка Personalized Recommendation in Folksonomies • Extend the folksonomy • Combine both shared tags/resources – quadratic concepts – bring maximal shared sets of users, tags and resources • Personalize tags/resources recommendations – Users’ profile as a new dimension – look for both users’ profile and tagging history before making recommendation Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 64. Образец заголовкаQuadratic Concepts Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 65. Образец заголовкаSteps • Inputs: a set of frequent quadri-concepts, a user u with its profile p and optionally a resource r to annotate • Outputs: a set of proposed users, suggested tags and recommended resources • User Proposition Step – seeks for quadri- concepts whose users have the same profile • Tag Suggestion Step – suggest personalized tags to a target user that share a resource in the p-folksonomy • Resource Recommendation Step – propose a personalized list of resources to a targeted user that is susceptible to be in accordance with its interests Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 66. Образец заголовкаAlgorithm Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 67. Образец заголовкаEvaluation • MovieLens dataset – with examples of extracted quadri-concepts following different profiles of folksonomy’ users • 50,000 users • 95,580 tags applied to 10,681 movies by 71,567 users • Additional user information available: – Gender, profession, age • Training set/Test set – 80% as training set – 20% as validation data Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 68. Образец заголовкаResults Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 69. Образец заголовкаResults and Conclusions • In an average of 38% outperforms the precision of the approach of Liang et al., which is between 24% and 30% • Best performances obtained with k=5 • Quadratic concepts improves the recommendations by suggesting tags and resources the more specific to users’ needs Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Nguifo, A Personalized Recommender System Based on Users’ Information In Folksonomies
- 70. Образец заголовка Hybrid tag recommendation for social annotation system Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke
- 71. Образец заголовкаData Model • Record of a user labeling a resource with one or more tags • Collection of annotations results in a complex network of interrelated users, resources and tags • Social annotation system – Can be described as a four-tuple: U, R, T, A – Can be viewed as a three dimensional matrix: U, R, T • U: a set of users • R: a set of resources • T: a set of tags • A: a set of annotations Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 72. Образец заголовка Linear Weighted Hybrid Tag Recommender • Aggregates the results of several component recommenders in linear combination • View each component of a tag recommendation system as a function • To produce a ranked list of suggested tags for a particular user given a specific resource: • Relevance score for a tag is calculated using several component tag recommenders Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 73. Образец заголовка Linear Weighted Hybrid Tag Recommender • Specializes in only a few available dimensions of the data • Focus on relatively simple component recommenders due to their speed and scrutability • Discussed components: – Popularity Models – User-Based Collaborative Filtering – Item-Based Collaborative Filtering Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 74. Образец заголовкаComponent 1: Popularity Models • Recommend the most popular tags • Strictly resource dependent • Does not take into account the tagging habits of the user • Serve as a baseline and may benefit the hybrid • Require little online computation • Easily built offline and can be incrementally updated Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 75. Образец заголовкаComponent 1: Popularity Models • Resource based popularity recommender • User based popularity recommender Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 76. Образец заголовкаComponent 2: User-based CF • Works under the assumption that users who have agreed in the past are likely to agree in the future • Relies on the collaboration of other users • Only recommends tags applied to the query resource • Narrows the focus of the recommendation regardless of the diversity in the user profile • Advantages: – Personalization • Disadvantages: – Cannot recommend tags that do not appear in a neighbor’s profile – Lacks the ability to reflect the habits and patterns of the larger crowd Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 77. Образец заголовкаComponent 3: Item-Based CF • Relies on discovering similarities among resources rather than among users • Similarity metrics only calculated with resources in the user profile • Constructs a neighborhood of resources from the user profile most similar to the query resource • Effectively ignores parts of the user profile not relevant to the recommendation task • Advantages: – Computation can be quickly done in real time – Similarities can be calculated offline for large user profile Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 78. Образец заголовкаEvaluation • Datasets – Bibsonomy, Citeulike, MovieLens, Delicious, Amazon, LastFM • Methodology 1. Each user’s annotations were divided equally among five folds 2. The recommenders are evaluated on their ability to recommend tags given a user-resource pair 3. Evaluate returned tags against the tags in the holdout annotation Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 79. Образец заголовкаResults • Integrative approach can exploit multiple dimensions of the data • Hybrid outperforms a state-of-the-art model- based algorithm based on tensor factorization (PITF) – particularly when the user profiles are diverse • Social annotation systems vary in how users interact with the system • The differences between datasets make the performance of individual recommenders unpredictable Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 80. Образец заголовкаAdvantages of the Proposed Hybrid System • More efficient, scalable, extensible and explainable than PITF • The proposed linear weighted hybrid inherits the capacity to focus on specific aspects of the user profile • Constructed from simple yet fast components • Offers a highly scalable and easily updatable solution for tag recommendation Jonathan Gemmell, Thomas Schimoler, Bamshad Mobasher, Robin Burke, Hybrid tag recommendation for social annotation system
- 81. Образец заголовка The Benefit of Using Tag-Based Profiles Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu
- 82. Образец заголовкаMotivation • Tags are used to enable the organization of information within personal information spaces that can also be shared • Tag distributions stabilize over time and can be used to improve search on the Web • Question: How tags can characterize the user and enable personalized recommendations? Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 83. Образец заголовкаExperiment • Dataset: Last.fm • Crawled subset of the Last.fm website, including pages corresponding to tags, music tracks and user profiles • Used track-based and tag-based profiles to evaluate different algorithms for producing music recommendations – Track-based user profiles: collections of music tracks with associated preference scores, describing users’ musical tastes – Tag-based user profiles: collections of tags together with corresponding scores representing the user’s interest in each of these tags Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 84. Образец заголовкаNotations Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 85. Образец заголовкаAlgorithms • 7 algorithms based on the type of profile and the technique used for getting the recommendations • three categories: – Collaborative Filtering based on Tracks – Collaborative Filtering based on Tags – Search based on Tags • Tag-based recommendation algorithms: – CF based on Track-Tags with ITF (CFTTI) – CF based on Track-Tags No-ITF (CFTTN) – CF based on Tags (CFTG) • Tag-Based Search algorithms – Search based on Track-Tags with ITF (STTI) – Search based on Track-Tags No-ITF (STTN) – Search based on Tags (STG) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 86. Образец заголовка CF based on Track-Tags with ITF (CFTTI) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 87. Образец заголовка CF based on Track-Tags No-ITF (CFTTN) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles • Differs from CFTTI by computing the tag based profiles without the IT F parameter in the formula corresponding to tags’ preference
- 88. Образец заголовкаCF based on Tags (CFTG) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 89. Образец заголовка Search based on Track-Tags with ITF (STTI) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 90. Образец заголовка Search based on Track-Tags No-ITF (STTN) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles • Remove the ITF parameter in the preference formula
- 91. Образец заголовкаSearch based on Tags (STG) Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 92. Образец заголовкаEvaluation • 18 subjects: B.Sc., Ph.D., and Post- Doc students in different areas of computer science and education • They installed the desktop application to extract their user profiles, then ran all 7 variants of the described algorithms • For each of the recommended tracks, the users provide two different scores: – how well the recommended track matches their music preferences – the novelty of the track Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 93. Образец заголовкаResults Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 94. Образец заголовкаResults Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 95. Образец заголовкаResults Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 96. Образец заголовкаResults Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 97. Образец заголовкаResults • All Collaborative Filtering algorithms based on tags (CFTG, CFTTI, CFTTN) performed worse than the baseline, as standard User- Item CF techniques already show high precision • All search algorithms show quite substantial improvements over track based CF • STG recommends much less popular tracks than our CFTR baseline, but still of higher quality Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 98. Образец заголовкаResults • A first set of algorithms, using collaborative filtering on tag profiles that were extracted from tracks, proved to be less successful than the baseline. • A second set of tag-based search algorithms however improved results’ quality significantly. • In addition to a 44% increase in quality for the best algorithm, search-based methods are also much faster than collaborative filtering and do not suffer from the cold start problem Claudiu S. Firan, Wolfgang Nejdl, Raluca Paiu, The Benefit of Using Tag-Based Profiles
- 99. Образец заголовка Harvesting social knowledge from folksonomies Harris Wu, Mohammad Zubair, Kurt Maly
- 100. Образец заголовкаMotivation • Enhance collaborative tagging systems to meet some key challenges: – community identification – user and document recommendation – ontology generation Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 101. Образец заголовкаCommunity Identification • Existing community identification techniques: – Spectral: identify all major communities in a large collection – Bibliometrics: determine the pair-wise affinity among users – Network flow based: identify broader communities containing a known existing community Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 102. Образец заголовкаUser and Document Recommendation • HITS (Kleinberg 1999) algorithm • Experiment different link weighting mechanisms and combinations with hyperlink analysis to improve the algorithm • Pair-wise similarities between the given document and the rest of the documents • Pair-wise similarities between a given user and the rest of the users Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 103. Образец заголовкаUser and Document Recommendation • HITS (Kleinberg 1999) algorithm • Experiment different link weighting mechanisms and combinations with hyperlink analysis to improve the algorithm • Pair-wise similarities between the given document and the rest of the documents • Pair-wise similarities between a given user and the rest of the users Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 104. Образец заголовкаOntology Generation • An ontology is one of the most efficient structures for navigation – any document can be reached with o(log(n)) • Hierarchical clustering problem • Different clustering techniques use different pair-wise similarity measures Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 105. Образец заголовкаOntology Generation Algorithm 1. identifies the set of documents for which the hierarchy needs to be generated, 2. identifies all tags associated with these documents. 3. constructs a document-tag matrix, denoted by A – Aij = 1 iff document i is tagged by tag j 4. constructs a tag-tag matrix to store the semantic similarities between tags 5. Multiplied A by the tag-tag matrix 6. Each document is now represented by a row vector Ai Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 106. Образец заголовкаEvaluation • Offline studies as pre-tests of the design concepts • Collect data through paper-based questionnaires and face-to-face interviews • Use test websites to evaluate selective modules of the proposed design solutions • Use pilot systems to evaluate the proposed design in large knowledge creation environments • Simulate large amounts of user input data to test the scalability Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 107. Образец заголовкаConclusions • Collaborative tagging systems have the potential of becoming a technological infrastructure for harvesting social knowledge • There are many challenges • The proposed designed prototypes enhance social tagging systems to meet some of the key challenges • Preliminary results show promise Harris Wu, Mohammad Zubair, Kurt Maly, Harvesting social knowledge from folksonomies
- 109. Образец заголовкаRecap • Recommender systems are widely used in the web – Facebook, Amazon, Netflix, … • There are many different recommender algorithms • Tradition recommender algorithms has pros and cons • Hybrid approaches combines multiple recommender algorithms • User profile is useful for personalized recommendations • Leveraging Tagging Systems with User Information can improve results
- 110. Образец заголовкаTake-Aways • Shared tags can improve resource discovery • Using quadratic concepts of users, tags, resources and profiles maximize sets of users sharing resources with the same tags. They can be used to find a personalized choice of tags and resources when suggestions are made following the users’ profiles • Hybrid tagging recommender system can cover more dimensions of the data by different components • Using tag-based search algorithms can significantly improve the quality of results • Collaborative tagging systems have many challenges, but can be enhanced by using with other components
- 111. Образец заголовкаFuture Works • Current project at work: – There are a lot of files coming into the enterprise file distribution system daily – Files are tagged “automatically” based on file name and a set of predefined rules – Users subscribe to particular files based on predefined subscriptions • Problems: – File name contains file metadata, so it must be a certain format – Difficult to manually manage all predefined rules and subscriptions – Some files might be useful for analysts, but they didn’t subscribe
- 112. Образец заголовкаFuture Works • Implement algorithm to automatically suggest tags to a file • Implement algorithm to recommend public files to user based on their roles and interests
- 113. Образец заголовкаAcknowledgements • Daniar Asanov, Algortihms and Methods in Recommender Systems, 2011 • Robin Burke, Hybrid Recommender Systems: Survey and Experiments, User Modeling and User-Adapted Interaction, v.12 n.4, p.331-370, November 2002 • Mohamed Nader Jelassi, Sadok Ben Yahia, Engelbert Mephu Ngui, A Personalized Recommender System Based on Users’ Information In Folksonomies, Proceedings of the 22nd International Conference on World Wide Web, May 2013 • Kerstin Bischoff , Claudiu S. Firan , Wolfgang Nejdl , Raluca Paiu, Can all tags be used for search?, Proceedings of the 17th ACM conference on Information and knowledge management, October 26-30, 2008, Napa Valley, California, USA • Jonathan Gemmell , Thomas Schimoler , Bamshad Mobasher , Robin Burke, Hybrid tag recommendation for social annotation systems, Proceedings of the 19th ACM international conference on Information and knowledge management, October 26-30, 2010, Toronto, ON, Canada
- 114. Образец заголовкаAcknowledgements • Harris Wu , Mohammad Zubair , Kurt Maly, Harvesting social knowledge from folksonomies, Proceedings of the seventeenth conference on Hypertext and hypermedia, August 22-25, 2006, Odense, Denmark • Hao Ma , Dengyong Zhou , Chao Liu , Michael R. Lyu , Irwin King, Recommender systems with social regularization, Proceedings of the fourth ACM international conference on Web search and data mining, February 09- 12, 2011, Hong Kong, China • Philip Bonhard , Clare Harries , John McCarthy , M. Angela Sasse, Accounting for taste: using profile similarity to improve recommender systems, Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, April 22-27, 2006, Montréal, Québec, Canada • Claudiu S. Firan , Wolfgang Nejdl , Raluca Paiu, The Benefit of Using Tag- Based Profiles, Proceedings of the 2007 Latin American Web Conference, p.32-41, October 31-November 02, 2007 • Mohsen Jamali , Martin Ester, A matrix factorization technique with trust propagation for recommendation in social networks, Proceedings of the fourth ACM conference on Recommender systems, September 26-30, 2010, Barcelona, Spain
Editor's Notes
- #10: User-based: Recommendations are given to user based on evaluation of items by other users sharing common preferences Item-based: Predictions are calculated based on the similarity of ratings given by users for the items More apt for offline preprocessing of large rating matrix
- #11: This approach predicts the relevance of items for users based on user history, such as items previously purchased, viewed or liked by the visitor. The system compares one user’s history to others’ user journeys and based on this data, it creates a list of recommended items for the user. The collaborative filtering method suffers from the cold start problem, meaning that it cannot recommend items without historical data. They can be further classified into model-based and memory-based algorithms.
- #12: User-based: Recommendations are given to user based on evaluation of items by other users sharing common preferences Item-based: Predictions are calculated based on the similarity of ratings given by users for the items More apt for offline preprocessing of large rating matrix
- #24: Content-based filtering (CBF) algorithms recommend items whose metadata are similar to the metadata of items the user has interacted with in the past. For instance, in the case of product recommendations, the product description, category, price, physical parameters, etc. are content metadata. Unlike the collaborative filtering approach, CBF does not suffer from new-item and cold-start problems.
- #25: Goal: recommend items similar to those the user liked