Causation in its simplest form can be expressed as the pairing of a cause event with an effect event. Grouping and relating common past event pairs can be a key to predicting an effect event for a previously unseen cause event. This paper presents a project that (i) constructs an abstraction tree based on several semantic web datasets to represent all past events and then (ii) uses the abstraction tree to make such a prediction. To construct an abstraction tree, similarities between each pair of cause-effect events are measured using ConceptNet, DBpedia, and YAGO semantic web data and the measured similarity values are used by a clustering algorithm to group similar cause-effect event pairs together. To predict an effect event, the abstraction tree is traversed to find similar past cause events and then the paired effect events are extracted, which are then generalized using ConceptNet to form a new effect event. A user study shows favorable ratings on new predicted events compared with known baseline events.
[1]
Hossein Shirazi,et al.
News events prediction using Markov logic networks
,
2018,
J. Inf. Sci..
[2]
Shaul Markovitch,et al.
Learning to Predict from Textual Data
,
2012,
J. Artif. Intell. Res..
[3]
Shaul Markovitch,et al.
Learning Causality from Textual Data
,
2011
.
[4]
Kira Radinsky,et al.
Learning causality for news events prediction
,
2012,
WWW.
[5]
Juan-Zi Li,et al.
What Happens Next? Future Subevent Prediction Using Contextual Hierarchical LSTM
,
2017,
AAAI.
[6]
Eric Horvitz,et al.
Mining the web to predict future events
,
2013,
WSDM.