CLEOPATRA

HapPenIng: Happen, Predict, Infer

— Event Series Completion in a Knowledge Graph

Simon Gottschalk & Elena Demidova L3S Research Center, Leibniz Universität Hannover

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 1 Challenge: Knowledge Graph Completion

● Knowledge Graphs (KGs) such as DBpedia and Wikidata are incomplete by nature ⇨ Need for KG completion

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 2 Challenge: Knowledge Graph Completion

● Knowledge Graphs (KGs) such as DBpedia and Wikidata are incomplete by nature ⇨ Need for KG completion ● Existing approaches ● Link prediction

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 3 Challenge: Knowledge Graph Completion

● Knowledge Graphs (KGs) such as DBpedia and Wikidata are incomplete by nature ⇨ Need for KG completion ● Existing approaches ● Link prediction ● Utilize knowledge from external sources

……… ……… ……… …..

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 4 Challenge: Knowledge Graph Completion

● Knowledge Graphs (KGs) such as DBpedia and Wikidata are incomplete by nature ⇨ Need for KG completion ● Existing approaches ● Link prediction ● Utilize knowledge from external sources

How to enrich KGs by adding new nodes, without relying on external knowledge?

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 5 Solution: KG Completion in Event Series

● KG completion for event series, such as ● Wimbledon Championships ● US presidential elections ● International Semantic Web Conference

● Event series are often covered incompletely ● focus on recent and current events

● Event series editions often follow similar patterns ⇨ use structural similarity to infer new events

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 6 Wikidata Example: Wimbledon Championships

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 7 Wikidata Example: Wimbledon Championships

Wimbledon Championships 2019

Men's Men's Women's Women's Mixed Singles Doubles Singles Doubles Doubles

Final Final Final Final Final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 8 History of the Wimbledon Championships

1877 1885

2018 2019

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 9 Incomplete Knowledge Graph: Missing Edge

1877 1885

2018 2019

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 10 Incomplete Knowledge Graph: Missing Node

1877 1885

2018 2019

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 11 Approach

1. Definition of event series graphs ● Event series and sub-event relations

2. Sub-event relation prediction ● Find new sub-event relations ● Binary classification problem

3. Event inference ● Infer new events ● Add event description ○ Label, time and location

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 12 Wimbledon Example: Event Series Graph

2008 Wimbledon 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 13 Wimbledon Example: Event Series Graph Event Series: Wimbledon Championships

follow-up 2008 Wimbledon relation 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 14 Wimbledon Example: Event Series Graph Event Series: Wimbledon Championships

follow-up 2008 Wimbledon relation 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

sub-event relation

2008 Wimbledon “2009 Wimbledon 2010 Wimbledon Championships – Championships – Championships – Men's Singles Men's Singles” Men's Singles

Event Series: Wimbledon Championships – Men's Singles

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 15 Wimbledon Example: Event Series Graph Event Series: Wimbledon Championships

follow-up 2008 Wimbledon relation 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

sub-event relation

2008 Wimbledon “2009 Wimbledon 2010 Wimbledon Championships – Championships – Championships – Men's Singles Men's Singles” Men's Singles

2009 Wimbledon Championships – Men's Singles final

Event Series: Wimbledon Championships Event Series: Wimbledon Championships – Men's Singles – Men's Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 16 Wimbledon Example: Event Series Graph Event Series: Wimbledon Championships

follow-up 2008 Wimbledon relation 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

sub-event relation

2008 Wimbledon “2009 Wimbledon 2010 Wimbledon Championships – Championships – Championships – Men's Singles Men's Singles” Men's Singles

2008 Wimbledon Championships – 2009 Wimbledon Men's Singles final Championships – Men's Singles final

Event Series: Wimbledon Championships Event Series: Wimbledon Championships – Men's Singles – Men's Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 17 Wimbledon Example: Sub-Event Relation Prediction Event Series: Wimbledon Championships

follow-up 2008 Wimbledon relation 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

sub-event relation

2008 Wimbledon “2009 Wimbledon 2010 Wimbledon Championships – Championships – Championships – Men's Singles Men's Singles” Men's Singles

Predicted sub-event relation 2008 Wimbledon 2009 Wimbledon Championships – Championships – Men's Singles final Men's Singles final

Event Series: Wimbledon Championships Event Series: Wimbledon Championships – Men's Singles – Men's Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 18 Wimbledon Example: Event Inference Event Series: Wimbledon Championships

follow-up 2008 Wimbledon relation 2009 Wimbledon 2010 Wimbledon Championships Championships Championships

sub-event relation

2008 Wimbledon “2009 Wimbledon 2010 Wimbledon Championships – Championships – Championships – Men's Singles Men's Singles” Men's Singles

Predicted sub-event relation 2008 Wimbledon 2009 Wimbledon 2010 Wimbledon Championships – Championships – Championships – Inferred Men's Singles final Men's Singles final Men's Singles final event

Event Series: Wimbledon Championships Event Series: Wimbledon Championships – Men's Singles – Men's Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 19 Sub-Event Relation Prediction I

ep: 2008 Wimbledon ● Binary classification problem Championships – Men's Singles ○ Given an event pair (es, ep), we aim to predict

whether es is a sub-event of ep

es: 2008 Wimbledon ● Features Championships – ○ STransE embedding score Men's Singles final ○ Textual features based on labels and template labels - Template of "2008 Wimbledon Championships": "Wimbledon Championships" ○ Spatio-temporal features - Time overlap, time containment, time equality - Location overlap

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 20 Sub-Event Relation Prediction II

ep: 2008 Wimbledon ● Binary classification problem Championships – Men's Singles ○ Given an event pair (es, ep), we aim to predict

whether es is a sub-event of ep

es: 2008 Wimbledon ● For each event in a series that does not have a Championships – sub-event: Find potential sub-events Men's Singles final ● Repeat that search until all leaf nodes were tested

● Training based on a balanced set of positive and difficult negative examples

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 21 Event Inference 2010 Wimbledon ● If an event is repeated in most editions of an Championships – event series, but is missing in a particular Men's Singles edition, we can infer that missing event

2010 Wimbledon ● Label, time and location generation for each Championships – Men's Singles final new event

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 22 Event Inference — Example

e.series: Wimbledon Championships – Men's Singles

2008 Wimbledon 2009 Wimbledon e: 2010 Wimbledon Championships – Men's Championships – Men's Championships – Men's Singles Singles Singles

2008 Wimbledon 2009 Wimbledon Championships – Men's Championships – Men's Singles final Singles final

m ∈ M: Wimbledon Championships – Men's Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 23 Event Inference — Example

e.series: Wimbledon Championships – Men's Singles

2008 Wimbledon 2009 Wimbledon e: 2010 Wimbledon Championships – Men's Championships – Men's Championships – Men's Singles Singles Singles

2008 Wimbledon 2009 Wimbledon New Event: Championships – Men's Championships – Men's ??? Singles final Singles final

m ∈ M: Wimbledon Championships – Men's Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 24 2009 Wimbledon 2010 Wimbledon Championships – Men's Championships – Men's Singles Singles Event Inference — Label Generation

Example 2009 Wimbledon Championships – Men's ??? Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 25 2009 Wimbledon 2010 Wimbledon Championships – Men's Championships – Men's Singles Singles Event Inference — δ Label Generation

Example 2009 Wimbledon 2010 Wimbledon Championships – Men's Championships – Men's Singles final Singles final

step δ.op δ.text event label

init 2009 WC - Men’s Singles final

1 DELETE 2009 WC - Men’s Singles final

2 INSERT 2010 2010 WC - Men’s Singles final

3 EQUAL WC - Men’s Singles 2010 WC - Men’s Singles final

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 26 Event Inference — Constraints

● Time Evolution (EVO) ○ the event has happened in a previous edition ?

● Window (WIN) ○ the event has happened within a previous and b later editions

● Evolution coverage window (ECW) ○ the event has happened before and in ɑ (%) of a previous and b later editions

● ...

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 27 Datasets & Results

● Two events graphs ● Wikidata ● DBpedia (English) ● Enriched with time and location information of EventKG

Wikidata DBpedia

Events 352,235 92,523

Event Series 9,007 1,871

New Sub-event relations 89,492 6,306

New Events 5,010 1,364

Data and results available at: http://eventkg.l3s.uni-hannover.de/event_series_files/index.html

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 28 Example Results (Wikidata)

New Sub-events relations Parent Event (Wikidata ID) Sub Event (Wikidata ID) Azerbaijan at the (Q16152206) Azerbaijan at the (Q38089620) Canarian parliamentary election (Q8774363) Canarian parliamentary election, 1999 (Q8774378) WCT World Doubles (Q1134214) 1973 WCT World Doubles (Q3851605)

New Events

Parent Event (Wikidata ID) New Sub Event Start Date End Date Location 2013 Alpine Skiing World Cup 2013 Alpine Skiing World Cup – 2012-01-01 2012-12-31 - (Q14564) Men's Super G Democratic Party leadership election, New Brunswick New Democratic 2013-01-01 2013-12-31 - 2013 (Q15107411) Party leadership election, 2013 2004 Wimbledon Championships 2004 Wimbledon Championships – 2004-01-01 2004-12-31 - (Q268871) Mixed Doubles at the 2019 European 2019 European Games (Q17112170) 2019-01-01 2019-12-31 Baku Crystal Hall Games

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 29 Parent Event Evaluation US presidential election, 1892 ● Manual and automated evaluation to measure New Event ○ Sub-event relation prediction accuracy US presidential ○ Ability to reconstruct corrupted event graphs election in ○ Correctness of inferred events Illinois, 1892 ● Sub-event relation prediction: ○ Textual and spatio-temporal features improve solely embedding-based binary classification by 21% ● Event inference: ○ More than 60% of removed events can be reconstructed ○ ECW constraint infers 70% correct new events

X New sub-event relations and inferred events are valuable inputs for KG completion

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 30 Thank You!

Questions?

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 31 Back-up Slides

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 32 Sub-Event Relation Prediction — Features

ep: 2008 Wimbledon ● Textual features: Championships – ○ Label containment, LCS fraction, unigram Men's Singles similarity, label cosine similarity ○ Template containment, template LCS fraction, es: 2008 Wimbledon Template unigram similarity Championships – ○ Parent event label length, sub-event label length Men's Singles final

● Spatio-temporal features ○ Time overlap, time containment, time equality ○ Location overlap

● Embedding features ○ Embedding score

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 33 Motivation: Knowledge Graph Completion

● Type Assertion rdf:type Roger Tennis Federer Player

● Link Prediction

Zürich Roger dbo:birthPlace Basel Federer Bern

So far: Knowledge graph completion finds new connections between existing nodes ⇒ Goal: generate new nodes without external resources Approach: detect patterns in event series

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 34 Event Inference — Example I e.series: Event Series: Wimbledon Championships

“2008 “2009 e: “2010 Wimbledon Wimbledon Wimbledon Champion- Champion- Champion- ships” ships” ships”

“2008 “2009 “2010 Wimbledon Wimbledon Wimbledon Championships Championships Championships – Men's – Men's – Men's Singles” Singles” Singles”

m ∈ M: Event Series: sub-event of e Wimbledon Championships – Men's Singles and part of m

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 35 Sub-Event Prediction: Evaluation I (Examples)

● Positive Examples: ○ 1989 African Cup of Champions Clubs Final → 1989 African Cup of Champions Clubs ○ European Parliament election of 2009 in the Netherlands → European Parliament election, 2009

● Negative Example: ○ at the 2015 Summer Universiade → Tennis at the 2015 Summer Universidade

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 36 Event Inference - Evaluation I (Examples)

● Positive Examples: ○ FIBA EuroBasket 2017 Group A → FIBA EuroBasket 2017 ○ Argentine legislative presidential election, 1993 → Argentine legislative election, 1993

● Negative Examples: ○ Jody Wil2013liams: A realistic vision for world peace → TEDWomen 2013 ○ 1988 Brazil at the Winter Olympics → 1988 Winter Olympics

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 37 Sub-Event Relation Prediction ep: “2008 Wimbledon Champion- ships – Men's ● Binary classification problem Singles”

○ Given an event pair (es, ep), we aim to predict

whether es is a sub-event of ep

es: “2008 ● For each event in a series that does not have a Wimbledon Champion- ships – Men's sub-event: Find potential sub-events Singles final” ● Repeat that search until all leaf nodes were tested

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 38 “2010 Event Inference Wimbledon Champion- ships – Men's ● Identify similar patterns in the different editions of an Singles” event series

● If an event is repeated in most editions of an event “2010 Wimbledon Champion- series, but is missing in a particular edition, we can ships – Men's infer that missing event Singles final” ● Label, time and location generation for each new event

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 39 Sub-event relation prediction: Feature Groups

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 40 “2010 Event Inference — Constraints Wimbledon Champion- ships – Men's ● Naive/no constraints (NAI) Singles” ● Time Evolution (EVO)

○ the event has happened in a previous edition “2010 Wimbledon ● Interval ( ) Champion- INT ships – Men's ○ the event has happened in a previous and later edition Singles final” ● Window (WIN) ○ the event has happened within a previous and b later editions ● Coverage (COV) ○ the event has happened in ɑ (%) other editions ● Coverage window (CWI) ○ the event has happened in ɑ (%) of a previous and b later editions ● Evolution coverage window (ECW) ○ the event has happened before and in ɑ (%) of a previous and b later editions

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 41 Event Inference - Reconstruction Evaluation

Wikidata DBpedia

Corruption Factor

Constraints 5% 10% 15% 5% 10% 15%

Baseline BSL 6.81 63.13 61.83 39.58 38.40 38.17

EVO 53.63 54.70 53.12 31.04 31.32 30.12

INT 46.68 47.89 46.39 24.58 24.04 23.46

WIN 46.06 47.45 45.94 22.71 22.27 21.93 HapPenIng COV 45.49 45.65 43.64 11.46 11.03 9.30

CWI 53.36 53.93 51.32 23.96 21.96 19.43

ECW 48.89 49.17 47.03 21.67 20.71 18.18

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 42 Event Inference - Manual Evaluation

Wikidata DBpedia Inferred Events Inferred Events Relations Relations Constraints Number Precision Number Precision Baseline BSL 114,077 0.26 16,877 31,410 0.24 3,42 EVO 28,846 0.47 10,045 11,295 0.35 1,170 INT 5,256 0.57 5,376 2,115 0.67 3,419 WIN 3,363 0.56 4,547 936 0.71 783 HapPenIng COV 7,297 0.54 2,712 1,313 0.45 417 CWI 7,965 0.59 4,442 1,965 0.61 718 ECW 5,010 0.70 3,687 1,364 0.70 655

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 43 Event Inference - Manual Evaluation

● Manual evaluation of the correctness of inferred events ● For each constraint, 100 newly inferred events were randomly sampled and judged as correct or not

● Without sub-event prediction, 30% less new events are inferred ● 93.84% of the inferred events are assigned a happening time ● 176 of the 8, 334 inferred events were assigned a location

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 44 Times & Locations ● Happening times: 99.36% of the inferred events are assigned a happening time. 0.38% of them were inferred by the first, 81.52% by the second and 18.10% by the third rule: ● If the happening time of each event s ∈ m equals its parent event’s happening time, also e adopts its happening time directly from its parent event; ● If the happening time of each event s ∈ m is modelled as a whole year, the happening time of e is also modelled as the same year as any of its (transitive) parent events; ● If the event label contains a year expression, that part is transformed into its happening time. ● Locations: Only 79 of the 5,010 inferred events were assigned a location under this strict condition: ● "If there is a location assigned to every event s ∈ m, this location is also assigned to e."

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 45 History of the Wimbledon Championships

1877 1887

1888 1892

2017 2019

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 46 Event Inference - Manual Evaluation Parent Event US presidential ● Manual evaluation of inferred events election, 1892 ● For each constraint, 100 newly inferred New Event US presidential events were randomly sampled and judged election in as correct or not Illinois, 1892 Wikidata DBpedia Inferred Events Inferred Events Relations Relations Constraints Number Precision Number Precision Baseline BSL 114,077 0.26 16,877 31,410 0.24 3,42 EVO 28,846 0.47 10,045 11,295 0.35 1,170 HapPenIng WIN 3,363 0.56 4,547 936 0.71 783

ECW 5,010 0.70 3,687 1,364 0.70 655 ● Without sub-event prediction, 30% less new events are inferred ● 93.84% of the inferred events are assigned a happening time ● 176 of the 8, 334 inferred events were assigned a location

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 47 Sub-Event Prediction: Classifier Evaluation

● 10-fold cross-validation of the sub-event prediction using different classifiers and all the introduced features

Wikidata DBpedia

Method TP TN FP FN Acc. Acc. Baseline STransE 46,479 43,143 6,949 13,859 0.81 0.50 LOG 54,345 46,605 3,487 5,993 0.91 0.87 HapPenIng SVM 55,958 48,825 1,267 4,380 0.95 0.92 RF 58,649 49,497 595 1,689 0.98 0.97

● Manual judgement ○ RF finds 85,805 new sub-event relations with 61% precision ○ STransE finds 46,807 new sub-event relations with 9% precision

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 48 Event Inference - Reconstruction Evaluation

● We remove a set of events and attempt to reconstruct them

X X ● Unconstrained approach is able to reconstruct > 60% of events ● The more incomplete a knowledge graph is, the harder it is to reconstruct its information Wikidata DBpedia Corruption Factor Constraints 5% 10% 15% 5% 10% 15% Baseline BSL 61.81 63.13 61.83 39.58 38.40 38.17 EVO 53.63 54.70 53.12 31.04 31.32 30.12 HapPenIng WIN 46.06 47.45 45.94 22.71 22.27 21.93 ECW 48.89 49.17 47.03 21.67 20.71 18.18

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 49 Event Inference - Manual Evaluation Parent Event ● Manual evaluation of inferred events US presidential ● For each constraint, 100 newly inferred election, 1892 events were randomly sampled and New Event judged as correct or not US presidential election in Illinois, 1892

Wikidata DBpedia Inferred Events Inferred Events Relations Relations Constraints Number Precision Number Precision Baseline BSL 114,077 0.26 16,877 31,410 0.24 3,42 EVO 28,846 0.47 10,045 11,295 0.35 1,170 HapPenIng WIN 3,363 0.56 4,547 936 0.71 783 ECW 5,010 0.70 3,687 1,364 0.70 655

HapPenIng: Happen, Predict, Infer — Event Series Completion in a Knowledge Graph Simon Gottschalk, Elena Demidova 50