slide1 n.
Skip this Video
Loading SlideShow in 5 Seconds..
Flickr Tag Recommendation based on Collective Knowledge B. Sigurbjörnsson and Roelof van Zwol. PowerPoint Presentation
Download Presentation
Flickr Tag Recommendation based on Collective Knowledge B. Sigurbjörnsson and Roelof van Zwol.

Loading in 2 Seconds...

play fullscreen
1 / 30

Flickr Tag Recommendation based on Collective Knowledge B. Sigurbjörnsson and Roelof van Zwol. - PowerPoint PPT Presentation

  • Uploaded on

Flickr Tag Recommendation based on Collective Knowledge B. Sigurbjörnsson and Roelof van Zwol. Proceeding of the 17th international conference on World Wide Web (WWW '08) Paper presentation by Sowmya Ramesh 04/18/2011. What is Flickr?. Flickr is an image hosting and video hosting website

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about 'Flickr Tag Recommendation based on Collective Knowledge B. Sigurbjörnsson and Roelof van Zwol.' - asta

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

Flickr Tag Recommendation based on Collective

  • Knowledge
  • B. Sigurbjörnsson and Roelof van Zwol.
  • Proceeding of the 17th international conference on World Wide Web (WWW '08)

Paper presentation by Sowmya Ramesh


what is flickr
What is Flickr?
  • Flickr is an image hosting and video hosting website
  • Search and organize/share your photos online
  • Users are encouraged to upload photos, then to name, describe, tag, geotag, comment on, and group their photos
what are tags
What are tags?
  • Keyword or category label
  • Describe the contents of the photo or provide additional contextual and semantical information
  • Provide meaningful descriptors of the objects, and allow the user to organize and index her content
how are tags helpful
How are tags helpful?
  • Proper labels associated with each image on the web would allow
    • More accurate image search; Increases the accuracy of current systems
    • Help you find photos and videos which have something in common
    • Organizing content for future navigation, filtering or search
    • For accessibility purposes, visually impaired individuals surfing the web need textual descriptions of images to be read aloud
    • Inappropriate content filtering (e.g., pornographic)
tag based search
Tag Based Search
  • Tagging is like filtering; out of all the items that are tagged, a filter (i.e. a tag) returns only those items tagged with that tag
  • From a user perspective, navigating a tag system is similar to conducting keyword-based searches; regardless of the implementation, users are providing salient, descriptive terms in order to retrieve a set of applicable items
    • E.g. : Flickr photo of La SagradaFamilia - a massive Roman Catholic basilica under construction in Barcelona - is described by its owner using the tags SagradaFamilia, and Barcelona. If associated with other relevant tags: Gaudi, Spain, Catalunya, architecture, and church it provides a richer semantical description of the photo and can be used to retrieve the photo for a larger range of keyword queries
  • Recent user studies show that users are willing to provide semantic context through manual annotations
    • Self
      • Solely for the traditional organization and self retrieval benefits
      • Personal memory; aid future recall of the situation it depicts
    • Social
      • benefit of known others, such as friends or family and strangers
      • communicate contextual information to others about the image
  • First dimension, “sociality,” relates to whether the tag’s intended usage is by the individual who took and uploaded the photo or by others, including friends/family and strangers.
  • Second dimension, “function,” refers to a tag’s intended uses
other systems for digital image retrieval
Other Systems For Digital Image Retrieval
  • Metadata based retrieval system
    • Images retrieved based on descriptions associated with the image such as creator, dates, genre, (source) image type, size, file name, title, caption text surrounding embedded images
  • Content based retrieval system
    • Search will analyze the actual contents of the image rather than the metadata
    • Content in this context refers to colors, shapes, textures and feature extraction
contribution of the paper
Contribution of the Paper
  • Contribution is twofold
    • How users tag photos and what kind of tags they provide
      • Analyze a representative snapshot of Flickr and present the results by means of a tag characterization focusing on how users tags photos and what information is contained in the tagging
    • How to assist users in the tagging phase
      • Based on the analysis, provide effective tag recommendation strategies to support the user in the photo annotation task by recommending a set of tags that can be added to the photo
key applications
Key Applications
  • Two potential applications for the recommendation strategies
    • Recommendations are presented to the user, who can select the relevant tags and add them to the photo
    • Recommended tags can be directly used to enrich the index of an image retrieval system
related work
Related Work
  • (Semi-)automatically annotate photographs
    • Suggest labels for unlabelled images based on visual features using image processing and machine learning algorithms
  • ESP game
    • Users play the game by suggesting tags for photos that appear on their screen and earn points suggesting same tags as another player
  • ZoneTag
    • Mobile photo upload tool which provides tag suggestion based on personal history, geographic location and time
evaluating tag behavior in flickr
Evaluating Tag Behavior in Flickr
  • For Research described in this paper, they have used a random snapshot from Flickr of 52 million publicly available photos with annotations that contain about 188 million tags in total, and about 3.7 million unique tags. Each photo has at least one user-defined tag.
general tag characteristics
General Tag Characteristics
  • Distribution of the Tag Frequency in Flickr
    • Probability of a tag having tag frequency x is proportional to x-1.15
    • Head of the power law contains tags that would be too generic to be useful as a tag suggestion. E.g.: Top 5 most frequent occurring tags are: 2006, 2005, wedding, party, and 2004
    • Tail of the power law contains the infrequent tags that typically can be categorized as incidentally occurring words. E.g. ambient vector
  • Distribution of the number of tags per photo in Flickr
    • Probability of having x tags per photo is proportional to x-0.33
    • Head of the power law contains photos that are already exceptionally exhaustively annotated, as there are photos that have more than 50 tags defined
    • Tail of the power law consists of more than 15 million photos with only a single tag annotated and 17 million photos having only 2 or 3 tags. Together this already covers 64% of the photos.
general tag characteristics1
General Tag Characteristics
  • Both follow a power law distribution
    • A power law applies to a system when large is rare and small is common. The distribution of individual wealth is a good example of this: there are a very few rich men and lots & lots of poor folks. A familiar way to think about power laws is the 80/20 rule: 80% of the wealth is controlled by 20% of the population
tag categorization what are users tagging
Tag Categorization: What are users tagging?
  • Mapped Flickr tags onto the WordNet broad categories
  • From chart below, we can conclude that users do not only tag the visual contents of the photo, but to a large extent provide a broader context in which the photo was taken, such as, location, time, and actions.
  • Distribution of Flickr tags over the most common WordNet categories
    • Unclassified : 48%
    • Classified : 52%
      • Locations – 28%
      • Artifacts or Objects – 16%
      • People or Groups – 13%
      • Time – 7%
      • Other – 27%
tag recommendation system
Tag Recommendation System
  • Given a photo with user-defined tags, an ordered list of m candidate tags is derived for each of the user-defined tags, based on tag co-occurrence. The lists of candidate tags are then used as input for tag aggregation and ranking, which ultimately produces the ranked list of n recommended tags. Number of tags being recommended depends on the relevancy of the tags.
different types of tags
Different Types of Tags
  • Three different types of tags:
    • User-defined tags: U refers to the set of tags that the user assigned to a photo
    • Candidate tags: Cu is the ranked list with the top m most co-occurring tags, for a user-defined tag u ∈ U. We denote C to refer to the union of all candidate tags for each user-defined tag u ∈ U
    • Recommended tags: R is the ranked list of n most relevant tags produced by the tag recommendation system
tag recommendation strategies
Tag Recommendation Strategies
  • Uses collective knowledge
    • Tag Co-occurrence
      • co-occurrence between two tags is defined as the number of photos [in the collection] where both tags are used in the same annotation
    • Tag Aggregation and Promotion
      • Tag aggregation step is needed to merge the candidate tags for each of the user-defined tags into single ranking. For a given set of candidate tags (C) a tag aggregation step is needed to produce the final list of recommended tags (R), whenever there is more than one user-defined tag.
      • Re-ranking procedure is implemented that promotes candidate tags having certain properties
tag co occurrence
Tag Co-occurrence
  • Two normalization methods are used to normalize the co-occurrence count with the overall frequency of the tag
    • Symmetric measures:Jaccard coefficient is used to normalize the co-occurrence of two tags ti and tj by calculating:
      • Jaccard coefficient is know to be useful to measure the similarity between two objects or sets. In this case to induce whether two tags have a similar meaning. Always assigns a number between 0 and 1.
    • Asymmetric measures: Tag co-occurrence can be normalized using the frequency of one of the tags. This gives probability of a photo being annotated with tag tj given that it was annotated with tag ti
difference between symmetric and asymmetric co occurrence measures
Difference Between Symmetric and Asymmetric Co-occurrence Measures
  • Consider the tag Eiffel Tower
    • For the symmetric measure we find that the most co-occurring tags are (in order): Tour Eiffel, Eiffel, Seine, La Tour Eiffel and Paris.
    • For the asymmetric measure the most co-occurring tags are (in order): Paris, France, Tour Eiffel, Eiffel and Europe.
  • Conclusion:
    • It shows that the Jaccard symmetric coefficient is good at identifying equivalent tags, like Tour Eiffel, Eiffel, and La Tour Eiffel, or picking up a close by landmark such as the Seine.
    • Based on this observation, it is more likely that asymmetric tag co-occurrence will provide a more suitable diversity of candidate tags than its symmetric opponent.
tag aggregation strategies
Tag Aggregation Strategies
  • Two aggregation strategies
    • Voting strategy : does not take the co-occurrence values of the candidate tags into account
tag aggregation strategies1
Tag Aggregation Strategies
  • Summing strategy: uses the co-occurrence values to produce the final ranking. The function P(c|u) calculates the asymmetric co-occurrence value
  • Promotion function to promote more descriptive tags for recommendation. From tag frequency distribution presented we learnt that both the head and the tail of the power law would probably not contain good tags for recommendation.
    • Stability-promotion: User-defined tags with very low collection frequency are less reliable than tags with higher collection frequency, we want to promote those tags for which the statistics are more stable.
      • |u| is the collection frequency of the tag u and Ks is a parameter in this function, which is determined by training
    • Descriptiveness-promotion: Promote the descriptiveness by damping the contribution of candidate tags with a very high-frequency
      • Kd is parameter in this function, and is configured by training
    • Rank-promotion : does not look at the co-occurrence value, but at the position r of the candidate tag c ∈ Cu for a given user-defined tag u. In the equation above, kr is a damping parameter
  • The combined promotion function we apply on a tag pair (u, c) is the following:

promotion(u, c) := rank(u, c) . stability(u) . descriptive(c)

  • Score function is update as presented below for the voting case
system tuning
System Tuning
  • For the evaluation 331 photos from Flickr which are evenly distributed over different tag classes was selected. Divided the photo pool in a training set and a test set. For training we used 131 photos and the test set consists of 200 photos. Training set of 131 photos to tune the parameters of the system
evaluation metrics
Evaluation Metrics
  • Mean Reciprocal Rank (MRR) - MRR measures where in the ranking the first relevant i.e., descriptive - tag is returned by the system, averaged over all the photos. This measure provides insight in the ability of the system to return a relevant tag at the top of the ranking
  • Success at rank k (S@k) - We report the success at rank k for two values of k: S@1 and S@5. The success at rank k is defined as the probability of finding a good descriptive tag among the top k recommended tags
  • Precision at rank k (P@k) - We report the precision at rank 5 (P@5). Precision at rank k is defined as the proportion of retrieved tags that is relevant, averaged over all photos
system evaluation
System Evaluation
  • Proposed strategies are effective
    • Good recommendation for 70% of photos at the first position in the ranking
    • Good recommendation for 94% of the photos among the top 5 ranks
    • P@5 : Precision - on an average more than half of our recommendations are good
    • Promotion function has a positive effect on the performance in particular on the precision at rank 5
    • Best strategy (vote+) has a stable performance over different classes of photos.
system evaluation1
System Evaluation

System is particularly good at recommending locations, artifacts and objects, both in terms of volume and acceptance ratio.

  • Collective knowledge of the community as a whole could be used to help user extend the annotations of individual photos
  • Novel and generic method for recommending tags approach was proposed which uses collective knowledge that resides in Flickr without introducing tag-class specific heuristics
  • System evaluation confirmed that the proposed strategies are effective
  • Scott A. Golder and Bernardo A. Huberman. 2006. Usage patterns of collaborative tagging systems. J. Inf. Sci. 32, 2 (April 2006), 198-208
  • Morgan Ames and MorNaaman. 2007. Why we tag: motivations for annotation in mobile and online media. In Proceedings of the SIGCHI conference on Human factors in computing systems(CHI '07)
  • L. von Ahn and L. Dabbish. Labeling images with a computer game. In CHI '04: Proceedings of the SIGCHI conference on Human factors in computing systems, pages 319{326, New York, NY, USA, 2004.ACM Press.