Abstract:Applying link prediction (LP) methods over knowledge graphs (KG) for tasks such as causal event prediction presents an exciting opportunity. However, typical LP models are ill-suited for this task as they are incapable of performing inductive link prediction for new, unseen event entities and they require retraining as knowledge is added or changed in the underlying KG. We introduce a case-based reasoning model, EvCBR, to predict properties about new consequent events based on similar cause-effect events present in the KG. EvCBR uses statistical measures to identify similar events and performs path-based predictions, requiring no training step. To generalize our methods beyond the domain of event prediction, we frame our task as a 2-hop LP task, where the first hop is a causal relation connecting a cause event to a new effect event and the second hop is a property about the new event which we wish to predict. The effectiveness of our method is demonstrated using a novel dataset of newsworthy events with causal relations curated from Wikidata, where EvCBR outperforms baselines including translational-distance-based, GNN-based, and rule-based LP models.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve the problems encountered in the task of event prediction based on knowledge graphs (KG). Specifically, the author points out that traditional link prediction (LP) methods have limitations when dealing with newly - emerging and unseen event entities. These methods are usually unable to perform inductive link prediction for new event entities, and when the knowledge in the underlying knowledge graph is added or changed, the model needs to be retrained. These problems limit the application of LP methods in causal event prediction tasks. To overcome these limitations, the author proposes a case - based reasoning (CBR) - based model - EvCBR for predicting the properties of new events. EvCBR identifies similar causal events through statistical metrics and uses the path prediction method to make predictions without a training step. This method is especially suitable for dynamically changing knowledge graphs such as Wikidata. ### Specific problem description 1. **Limitations of traditional LP methods**: - **Challenges in inductive link prediction**: Most LP methods can only predict the relationships between entities that already exist in the KG and cannot handle completely new entities. - **Impact of frequent updates**: Some KGs (such as Wikidata) are frequently updated and changed, which may lead to a decline in model performance because the training data becomes obsolete. 2. **Event prediction tasks**: - **Objective**: Predict the properties of new events based on causal relationships. - **Task form**: Represent the event prediction task as a 2 - hop link prediction task. For example, starting from a cause event, assume that there is a causal relationship connected to a new result event, and then predict the properties of that result event. ### Solutions The author proposes a case - based reasoning method - EvCBR. The main features of this method are as follows: 1. **No training required**: EvCBR does not rely on the training step. Instead, it uses statistical metrics and subclass hierarchies to identify similar entities and makes predictions through paths in the KG. 2. **2 - hop link prediction**: Represent the event prediction task as a 2 - hop link prediction task, starting from the cause event and predicting the properties of the result event. 3. **Adapt to dynamic KG**: EvCBR is especially suitable for frequently changing KGs such as Wikidata because it does not need to be retrained. ### Experimental verification The author uses a novel dataset extracted from Wikidata, which contains news events with causal relationships and their local connections. The experimental results show that EvCBR performs excellently in the 2 - hop inductive link prediction task and outperforms baseline methods including translation - distance - based, graph neural network (GNN) - based, and rule - based LP models. ### Summary This paper solves the limitations of traditional LP methods in dealing with new event entities by proposing the EvCBR model, especially in dynamically changing knowledge graphs. EvCBR successfully predicts the properties of new events through the case - based reasoning method, providing an effective solution for event prediction tasks.

Event Prediction using Case-Based Reasoning over Knowledge Graphs

CausalLP: Learning causal relations with weighted knowledge graph link prediction

Location Based Link Prediction for Knowledge Graph

HyperCausalLP: Causal Link Prediction using Hyper-Relational Knowledge Graph

Probabilistic Case-based Reasoning for Open-World Knowledge Graph Completion

Influence of Backdoor Paths on Causal Link Prediction

A Benchmark for Semi-Inductive Link Prediction in Knowledge Graphs

Causal Inference KGC: A Multi-Hop Reasoning Framework Over Sparse Knowledge Graph

A Multi-Hop Link Prediction Approach Based on Reinforcement Learning in Knowledge Graphs.

Generalized Relation Learning with Semantic Correlation Awareness for Link Prediction

Causal Lifting and Link Prediction

Spatial-Temporal Knowledge Graph Network for Event Prediction

Explainable Link Prediction for Emerging Entities in Knowledge Graphs

LinkLogic: A New Method and Benchmark for Explainable Knowledge Graph Predictions

Predicting the Co-Evolution of Event and Knowledge Graphs

Causal Inference for Knowledge Graph Based Recommendation.

Causal Graph Representation Learning for Outcome-Oriented Link Prediction

Causal Knowledge Guided Societal Event Forecasting

Causal Reinforcement Learning for Knowledge Graph Reasoning

Link Prediction of Knowledge Graph Based on Bayesian Network

CausalKG: Causal Knowledge Graph Explainability using interventional and counterfactual reasoning