Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 1 , Johanna Geiß 1 , Michael Gertz 1 , Stefan Hagedorn 2 and Kai-Uwe Sattler 2 1 Heidelberg University, Institute of Computer Science Database Systems Research Group, Heidelberg 2 Technical University of Ilmenau Databases and Information Systems Group, Ilmenau { spitz, geiss, gertz } @informatik.uni-heidelberg.de { stefan.hagedorn, kus } @tu-ilmenau.de 10th GIR Workshop San Francisco, , 2016
Motivation Event Networks Event Refinement Evaluation Summary Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 1 of 23
Motivation Event Networks Event Refinement Evaluation Summary What is an Event? Event “The Jimi Hendrix Experience toured Germany in 1967.” Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 2 of 23
Motivation Event Networks Event Refinement Evaluation Summary What is an Event? Event “The Jimi Hendrix Experience toured Germany in 1967.” Definition: Event “Something that happens at a given place and time between a group of actors .” [All02] Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 2 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Triangles Intuition: • Events correspond to triangular structures of entities • Participating entities can be used to extract events • Linguistic, sentence-based event detection can improve results Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 3 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Extraction Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 4 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Extraction Generating weights for events: • Similarity of entities [GSG15]: � � − dist ( r,r ′ ) φ ( r, r ′ ) := exp 2 Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 4 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Extraction Generating weights for events: • Similarity of entities [GSG15]: � � − dist ( r,r ′ ) φ ( r, r ′ ) := exp 2 • Weight of individual events: ω ( e i ) = min { φ ( l, t ) , φ ( l, a ) , φ ( t, a ) } Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 4 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Extraction Generating weights for events: • Similarity of entities [GSG15]: � � − dist ( r,r ′ ) φ ( r, r ′ ) := exp 2 • Weight of individual events: ω ( e i ) = min { φ ( l, t ) , φ ( l, a ) , φ ( t, a ) } • Weight of aggregated events: ω ( e ) := � k i =1 ω ( e i ) Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 4 of 23
Motivation Event Networks Event Refinement Evaluation Summary Geographic Background Network Locations can be connected in a location network G L based on: • Geographic containment • Geographic neighbourhood • Reachability • Semantic similarity • Context-dependent relations Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 5 of 23
Motivation Event Networks Event Refinement Evaluation Summary Temporal Background Network Dates are inherently connected in a temporal network G T : • Temporal containment is straightforward • Heterogeneous granularity levels are possible Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 6 of 23
Motivation Event Networks Event Refinement Evaluation Summary Social Background Network Actors form a social network G A and are connected by: • Acquaintance or Relation • Collaboration • Organization membership • Context-dependent similarities Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 7 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Hypergraph Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 8 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Hypergraph Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 8 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Hypergraph Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 8 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Refinement Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 9 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Refinement Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 10 of 23
Motivation Event Networks Event Refinement Evaluation Summary Granularity Refinement Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 11 of 23
Motivation Event Networks Event Refinement Evaluation Summary Neighbourhood Refinement Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 12 of 23
Motivation Event Networks Event Refinement Evaluation Summary Neighbourhood Refinement Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 12 of 23
Motivation Event Networks Event Refinement Evaluation Summary Event Refinement: Stratification Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 13 of 23
Motivation Event Networks Event Refinement Evaluation Summary Evaluation Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 14 of 23
Motivation Event Networks Event Refinement Evaluation Summary Extraction of Events from Wikipedia Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 15 of 23
Motivation Event Networks Event Refinement Evaluation Summary Entities and Events by Window Size · 10 5 · 10 8 4 8 number of entities number of events 3 6 2 4 1 2 0 1 2 4 6 8 10 window size w | A | | L | | T | |E| Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 16 of 23
Motivation Event Networks Event Refinement Evaluation Summary Background Network Construction Location Network: • From Wikidata: continents, countries and cities • Containment edges between hierarchy levels • Neighbourhood edges between adjacent countries • Neighbourhood edges between close cities Temporal Network: • Temporal Tagging with Heideltime [SG13] • Containment edges between years, months and days • Neighbourhood edges within granularity layers Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 17 of 23
Motivation Event Networks Event Refinement Evaluation Summary Ground Truth Event Queries We obtain events by • Named Entity Recognition in online news articles • Extraction of the entity triples of events by hand We generate queries by making events less certain in the combination of dimensions • Location L • Time T • Granularity g • Neighbourhood n Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 18 of 23
Motivation Event Networks Event Refinement Evaluation Summary Evaluation Results (Precision) 1 T g 0 . 8 L g T g L g 0 . 6 T n MAP L n T n L n 0 . 4 T g L n T n L g 0 . 2 orig 0 0 1 2 4 6 8 10 window size w Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 19 of 23
Motivation Event Networks Event Refinement Evaluation Summary Evaluation Results (Recall) 1 T g L g T g L g 0 . 8 T n Recall L n T n L n 0 . 6 T g L n T n L g orig 0 . 4 0 1 2 4 6 8 10 window size w Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 20 of 23
Motivation Event Networks Event Refinement Evaluation Summary Summary New method for event representation: • As a hypergraph model • Backed by underlying entity networks • Compatible with any entity-based definition of event Graph-based event refinement offers: • Spatio-temporal refinement in two dimensions: neighbourhood and granularity • Efficient computation due to localized queries • Language independence Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 21 of 23
Motivation Event Networks Event Refinement Evaluation Summary Ongoing Work Directions for ongoing event refinement research: • Include measures of granularity and neighbourhood in social background networks • Include hierarchical or organization networks • Add event structures beyond triangles Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 22 of 23
Motivation Event Networks Event Refinement Evaluation Summary The event network and the background networks are available for download. http://dbs.ifi.uni-heidelberg.de/index.php?id=data Refining Imprecise Spatio-temporal Events: A Network-based Approach Andreas Spitz 23 of 23
Recommend
More recommend
