# Temporal Knowledge Graph

## Paper Reviews

### Temporal Knowledge Graph

> #### [Know-Evolve: Deep Temporal Reasoning for Dynamic Knowledge Graphs](https://arxiv.org/pdf/1705.05742.pdf)
>> 1. Dynamic Evolving Knowledge Graphs contain temporal information for each edge.
>> 2. Reasoning over time
>> 3. Method: A deep evolutionary knowledge network that learns non-linearly evolving entity representations over time.
>> 
>>
>>Keywords: Dynamic knowledge graph
>>GitHub: https://github.com/rstriv/Know-Evolve
>>Link: http://blog.openkg.cn/%E8%AE%BA%E6%96%87%E6%B5%85%E5%B0%9D-know-evolve-deep-temporal-reasoning-for-dynamic-kg/

### Temporal Graph Convolutional Networks
> #### *[Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition](https://arxiv.org/abs/1801.07455)
>> 1. This model is formulated on top of a sequence of skeleton graphs, where
each node corresponds to a joint of the human body.
>> 2. There are two types of edges, namely the spatial edges that conform to the natural connectivity of joints and the temporal edges that connect the same joints across consecutive time steps.
>> 3. Each skeleton graph is a slice of temporal graphs.
>>
>> Keywords: Temporal Graph Convolutional Networks, 3D graph，Skeleton-Based Action Recognition
>> Code：https://github.com/yysijie/st-gcn
>>
>> 

> #### [Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting](https://arxiv.org/abs/1709.04875)
>> 1. Timely traffic forecast problem.
>> 2. Gated CNNs for Extracting Temporal Features
>> 3. Two real-world traffic datasets:
>> BJER4 and PeMSD7, collected by Beijing Municipal Traffic Commission and California Deportment of Transportation, respectively.
>>
>> Keywords: Temporal Graph, Gated CNN
>>
>>
>>
>>

> #### [Spatio-Temporal Graph Convolution for Skeleton Based Action Recognition](https://arxiv.org/abs/1802.09834)
>> 1. To encode dynamic graphs, the constructed multi-scale local graph convolution filters (k-neighbor receptive field), consisting of matrices of local receptive fields and signal mappings, are recursively performed on structured graph data of temporal and spatial domain.
>> 2. Simultaneously perform local convolutional filtering on temporal motions and spatial structures.
>> 3. Multi-scale convolutional filters：
The temporal convolutional filtering recursively encodes motion variations while the spatial filtering extracts more robust feature of spatial structures.
>> 4. Assumption: the slices of different time are aligned.
>>
>> Keywords: Skeleton-Based Action Recognition

> #### [Dynamic Graph Convolutional Networks](https://arxiv.org/abs/1704.06199)
>> 1. Combine Long Short-Term Memory networks (LSTM) with graph convolutional networks to learn long short-term dependencies together with graph structure.
>> 2. First paper to combine rnn with gcn.
>>
>> 3. Datasets:
>> * DBLP: co-author relationships
>> * CAD-120: Composed of 122 RGB-D videos corresponding to 10 high-level human activities. Each video is annotated with sub-activity labels, object affordance labels, tracked human skeleton joints and tracked object bounding boxes.
>>
>>Keywords: LSTM + GCN

### Deep Graph Networks
> #### [Relational inductive biases, deep learning, and graph networks](https://arxiv.org/pdf/1806.01261.pdf)
>> 1. Deepmind, published in June, 2018
>> 2. A summary of all graph methods, include GCNs.
>>
>>

>>Keywords: graph network
>>Links:
>>* https://mp.weixin.qq.com/s/iQYVyo2PHuGbEsYgdIf_oQ
>>* https://mp.weixin.qq.com/s/yYJw7gvploiRCTfwmeSwZQ

## Datasets

### GDELT
The GDELT Project is the largest, most comprehensive, and highest resolution open database of human society.
* Time: Collected from April 1, 2015 to Mar 31, 2016 (temporal granularity of 15 mins).
* People, organizations, locations. themes. counts. images and emotions and so on.
* 15 Minute Updates:
Access the world’s breaking events and reaction in near-realtime as both the GDELT Event and Global Knowledge Graph now update every 15 minutes.
* The entire underlying event and graph datasets in CSV format - over 2.5TB for last year alone.

Link: https://www.gdeltproject.org/data.html
Data Description： https://blog.csdn.net/qq_23926575/article/details/78064093
Offcial Website: https://blog.gdeltproject.org/gdelt-2-0-our-global-world-in-realtime/
Codebook: http://data.gdeltproject.org/documentation/GDELT-Global_Knowledge_Graph_Codebook-V2.1.pdf

### ICEWS
Event data consists of coded interactions between socio-political actors.
* Publication Date 2015-03-27
* Author
Boschee, Elizabeth (BBN Technologies)
Lautenschlager, Jennifer (Lockheed Martin)
Shellman, Steve (Strategic Analysis Enterprises)
Shilliday, Andrew (Lockheed Martin)
(Producer: Lockheed Martin et. al. http://www.icews.com)
* Collected from Jan 1, 2014 to Dec 31, 2014 (temporal granularity of 24 hrs).
* Interactions between socio-political actors (i.e., cooperative or hostile actions between individuals, groups, sectors and nation states)
* Contain both named individuals or groups, known as actors, and generic individuals or groups, known as agents. Actors are known by a specific name, such as 'Free Syrian Army' or 'Goodluck Johnathan', while agents are known by a generic improper noun, such as 'insurgents' or 'students'.
* Time: Both actors and agents have time-dependent affiliations with another actor
* Update: More than 15 million events extracted from news stories.
* Approximately 50,000 new events are uploaded monthly.

Link: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/28075
ICEWS program can be found at: https://www.lockheedmartin.com/en-us/news/features/2016/ICEWs-10000-dataset-download.html

Datasets used in paper "know-evolve":

### EventKG

* Existing knowledge graphs, with popular examples including DBpedia, YAGO and Wikidata, focus mostly on entity-centric information and are insufficient in terms of their coverage and completeness with respect to events and temporal relations.
* EventKG is a multilingual event-centric temporal knowledge graph that addresses this gap.
* 690 thousand contemporary and historical events and over 2.3 million temporal relations extracted from several large-scale knowledge graphs.
* Extracted from several largescale entity-centric knowledge graphs (i.e. Wikidata, DBpedia in five language editions and YAGO), as well as WCEP and Wikipedia event lists in five languages.


Paper: [EventKG: A Multilingual Event-Centric Temporal Knowledge Graph](https://arxiv.org/abs/1804.04526)
Link: http://eventkg.l3s.uni-hannover.de/