This page should contain all modules created (or planned to be created) in WP3. Please list their descriptions, endpoints, inputs, outputs, etc.
Contact information of partners involved in the creation of the modules and available for questions/support.
Users Repository
Enables user- and event-specific metadata retrieval.
WADL: http://pharos.l3s.uni-hannover.de:8090/application.wadl
Methods - USERS:
Method for retrieving the user details for the specified user id
IN: {ID} = user ID in the repository [given as path parameter]
OUT: Information (both generated in GLOCAL and crawled from external sources, e.g. Flickr, LastFM, Upcoming) about the user and similar users
Example: http://pharos.l3s.uni-hannover.de:8090/users/11668
Method for retrieving the user friendship and neighbor connections for the specified user id
IN: {ID} = user ID in the repository [given as path parameter]
OUT: Information about the user‟s friends and neighbors (similar users)
Example: http://pharos.l3s.uni-hannover.de:8090/users/selectUserConnections/11668
Methods - EVENTS:
- List Events - /events (LUH)
Lists 30 events from the database - for testing purposes for now
IN: null
OUT: list of event URIs and labels
Example: http://pharos.l3s.uni-hannover.de:8090/events
- Similar Events by Tags - /events/tags (LUH)
Lists events matching the given list of tags. Offline computation required first from "User/Event Analyzer - Identify Candidate Events for Media"
IN: inTags=<list of tags separated by ":">
OUT: list of Upcoming IDs of the matching events (eg the event info from the Upcoming ID 1268524 can be retrieved from http://upcoming.yahoo.com/event/1268524)
Example: http://pharos.l3s.uni-hannover.de:8090/events/tags/?inTags=Round%20Table...
- Event by ID - /events/{eventid} (LUH)
Gets the information about a given event ID from the database
IN: event ID given in path, not as parameter
OUT: Informaion (from Upcoming) about the event; similar events and event media items
Example: http://pharos.l3s.uni-hannover.de:8090/events/2904719
- DB Query - /events/search (LUH)
Query the events DB by user selected queries
IN: query=<SQL clauses separated by ":" in the format "field=value">
OUT: list of event URIs and labels matching the query
Examples: http://pharos.l3s.uni-hannover.de:8090/events/search?query=name=New%20Yo...
http://pharos.l3s.uni-hannover.de:8090/events/search?query=UpcomingID=10...
Importer (LUH)
Imports data from different online sources into the Users Repository
Data sources:
- Flickr (image metadata: users, photos, tags, title, description)
- LastFM (social events: tags, users)
- Upcoming (event metadata: users / attendees, tags, time, place)
Connections from Flickr to Upcoming or LastFM can be done through machine tags on Flickr of the form "upcoming:*" or "lastfm:*"
User/Event Analyzer
Offline analysis of users and communities centered around events.
Methods:
- Identify Event Communities (LUH)
Analyzes users’ interaction and user generated metadata to mine implicit event-communites.
- Identify Highly Similar Events (LUH)
Near duplicate detection for events.
- Identify Candidate Events for Media (LUH)
Given set of tags describing a medium, find photos in the DB with similar tags and return the event where most found photos are part of. Online interface is "Users Repository - Similar Events by Tags".
Event Provenance Service [EPS] (iSOCO)
It aims at publishing provenance-related metadata about event life-cycle (e.g. creation, modification...) including the algorithm/component through which each new event of data is created/manipulated (e.g. manual creation, component X ...). This annotation of events is based on semantic provenance vocabularies.
Methods:
- Set provenance data for a given event
- Provenance data by event ID: Get provenance information about an event
- Search: query the triple store
New Event Detection (CERTH)
Given a large set of tagged media with geo-location information identify clusters of media where each cluster can be considered to represent a new event. (see http://www.clusttour.gr/glocal).
Methods:
Media Representation
- Visual-based representation
The image is represented as a vector of visual words.
In: MediaID, representationType
Out: Occurrence distribution of 500 pre-specified visual words in the image.
e.g., http://160.40.50.139:8085/compareMediaWP3/extractRepresentation?mediaId=...
The image is represented as a vector of tag words.
In: MediaID, , representationType
Out: Occurrence distribution of 1000 pre-specified tag words in the image.
e.g., http://160.40.50.139:8085/compareMediaWP3/extractRepresentation?mediaId=...
- Media-based representation
The image is represented as a vector of latent words.
In: MediaID, , representationType
Out: probability distribution of 50 pre-specified latent words.
e.g., http://160.40.50.139:8085/compareMediaWP3/extractRepresentation?mediaId=...
Measure similarity distance between media
Measures the euclidean, cosine, l1norm and jaccard distance between the visual-based representations of two media.
In: MediaID1, MediaID2, distanceType
Out: Visual similarity distance between MediaID1, MediaID2
e.g., http://160.40.50.139:8085/compareMediaWP3/measureVisualDistance?mediaId1=481447114&mediaId2=503120923&distanceType=cosine
Measures the euclidean, cosine, l1norm and jaccard distance between the tag-based representations of two media.
In: MediaID1, MediaID2, distanceType
Out: Tag similarity distance between media MediaID1, MediaID2
e.g., http://160.40.50.139:8085/compareMediaWP3/measureTagDistance?mediaId1=48...
Measures the euclidean, cosine and linorm distance between the media-based representations of two media
In: MediaID1, MediaID2, distanceType
Out: Media similarity distance between media MediaID1, MediaID2
e.g., http://160.40.50.139:8085/compareMediaWP3/measureMediaDistance?mediaId1=...
Group media into clusters & expose event-media relations
Offline process:
- Perform a clustering procedure that organizes media into groups based on their similarity distances.
- Use the media geo-location information to filter out non-interesting clusters.
- Extract a rough description for each of the identified clusters based on its most characteristic tags.
Expose event-media relations through rest services: (Tentative, subject to changes)
- Return a list of all identified clusters.
- Return a list of clusters that contain the input tag.
- Return a list of clusters contained in the provided bounding rectangle.
- Return a list of clusters associated with the particular place.
- Filter the photos of the clusters based on the usage license.
- Return a list of clusters that are contained between the provided time interval (e.g., begin_date and end_date).
Event Matching (UNITN)
- based on the metadata repository: http://glocal-dkm.science.unitn.it/services/api_help.html
- Get a similarity score between two events (plans to get a more structured output)
- Get matching Users for one User -- find a list of users that have similar events to a given user
- Get matching Events for one Event -- find a list of events that are similar to a given user
User Interface (DFKI) (only WP3-related features)
Available at:
- Last release (Y1-Review Meeting / version 2): http://www.dfki.de/glocal/GlocalUiPg2
- Latest inoffical version: http://www.dfki.de/glocal/latest
Suggest Events for Media
- A user can import all media of a Flickr account and the system will visualize events for each media item. The suggestions themselves are retrieved by using service The user can choose which media items she would like to import and one event for each medium.
- IN: a set of media from Flickr (each Flickr medium is referred to by its square thumbnail URL, e.g.: http://farm6.static.flickr.com/5219/5384616740_d31f77c7bc_s.jpg)
- OUT: a set of medium-event-links
Suggest Media Alignment to a Given Set of Events
- While browsing or organizing an event, the user can choose to add new media to that particular event and its sub-events. The for each given medium, the systems retrieves suitable events filtered by the currently visible event and shows a temporary represenation for the medium aligned to that event. The user will then decided whether the suggestions are correct or wrong, finally, when all suggestions are confirmed or rejected (the user might also accept or refuse all at once), the alignment becomes permanant.
- The server-side services are dummy implementations so far
- IN: set of media
- OUT: a set of updated events with media aligned to them
Connected Web Services
- Retrieve event description (XML) by URI ("Event by ID - /events/{eventid} (LUH)")
- Search events by keywords ("DB Query - /events/search (LUH)")
- Retrieve similar events ("Similar Events by Tags - /events/tags (LUH)")
- Get suitable events for a given set of media (Not mentioned in the description above... (?))
Services to Be Connected
- Retrieve medium description (XML) by URI (the medium URI can not be used as url for the XML so far)
- Search events by keywords ("DB Query - /events/search (LUH)")
- Get comments by URI (as soon as there are comments in the event XML of L3S - see UniTN service)
- Services of the user repository like getFriendsOf(userId) or suggestFriendsFor(userId)
