Know-Evolve: Deep Temporal Reasoning for Dynamic Knowledge Graphs (1705.05742v3)

Abstract: The availability of large scale event data with time stamps has given rise to dynamically evolving knowledge graphs that contain temporal information for each edge. Reasoning over time in such dynamic knowledge graphs is not yet well understood. To this end, we present Know-Evolve, a novel deep evolutionary knowledge network that learns non-linearly evolving entity representations over time. The occurrence of a fact (edge) is modeled as a multivariate point process whose intensity function is modulated by the score for that fact computed based on the learned entity embeddings. We demonstrate significantly improved performance over various relational learning approaches on two large scale real-world datasets. Further, our method effectively predicts occurrence or recurrence time of a fact which is novel compared to prior reasoning approaches in multi-relational setting.

• The paper introduces a temporal point process framework that models events as multi-dimensional processes influenced by evolving entity embeddings.
• It dynamically updates entity embeddings over time to capture recent interactions and temporal behavior.
• The approach outperforms state-of-the-art methods in link prediction and event timing on large-scale real datasets.

Know-Evolve: Deep Temporal Reasoning for Dynamic Knowledge Graphs

The paper "Know-Evolve: Deep Temporal Reasoning for Dynamic Knowledge Graphs" addresses the critical need for handling temporal dynamics within multi-relational data, a need that arises from the availability of large-scale, event-based interaction data. The authors introduce a novel framework named "Know-Evolve," which advances the capability of traditional knowledge graphs by incorporating temporal reasoning, thereby offering a more comprehensive understanding of evolving knowledge systems.

Key Contributions

1. Temporal Point Process Framework: The paper introduces a framework for temporal reasoning over knowledge graphs by leveraging the temporal point process. This approach models the occurrence of a fact as a multi-dimensional temporal point process, with its conditional intensity function influenced by the relationship score, which in turn is modulated by dynamically evolving entity embeddings.
2. Dynamic Embeddings: One of the significant innovations in this work is the development of an architecture that evolves entity embeddings over time. This architecture learns from incoming facts and updates the embeddings of entities based on their recent interactions and temporal behavior. This dynamic nature allows for a more nuanced understanding of entity behavior over time.
3. Predictive Capabilities: Beyond predicting the occurrence of facts, the architecture uniquely predicts the time when an event is likely to occur, addressing a gap unserved by previous relational learning approaches. This capability is critical for applications requiring timely insights, such as forecasting potential conflicts or alliances.
4. Open World Assumption: The model supports the Open World Assumption, which considers missing links as potentially occurring in the future. It also supports prediction over unseen entities, emphasizing its robustness and scalability in accommodating new and evolving data.
5. Experimental Validation: Large-scale experiments validate the framework's superiority. Using two real-world datasets, GDELT and ICEWS, the Know-Evolve framework demonstrates significantly improved link prediction performance compared to state-of-the-art methods that do not account for temporal dynamics.

Implications and Future Directions

The introduction of deep temporal reasoning into knowledge graphs has profound theoretical and practical implications. Practically, this research can enhance systems that depend on understanding and forecasting complex entity interactions over time, such as in social network analysis, intelligence gathering, and dynamic recommendation systems. Theoretically, it paves the way for more advanced models that can seamlessly integrate temporal contexts into knowledge representation and reasoning tasks.

Furthermore, by successfully applying a temporal point process framework to dynamic knowledge graphs, the paper opens new avenues for combining relational learning with temporal data analysis. Future research could explore more sophisticated temporal dynamics, such as alternating periods of activity and dormancy or hierarchical temporal structures within multi-relational contexts.

In conclusion, the Know-Evolve framework represents a substantial step forward in the field of knowledge graph augmentation with temporal reasoning. Its ability to model the evolution of knowledge effectively over time not only fills a crucial gap but also sets the stage for further exploration and innovation in dynamic knowledge-based systems.