Abstract:Event sequence models have been found to be highly effective in the analysis and prediction of events. Building such models requires availability of abundant high-quality event sequence data. In certain applications, however, clean structured event sequences are not available, and automated sequence extraction results in data that is too noisy and incomplete. In this work, we explore the use of Large Language Models (LLMs) to generate event sequences that can effectively be used for probabilistic event model construction. This can be viewed as a mechanism of distilling event sequence knowledge from LLMs. Our approach relies on a Knowledge Graph (KG) of event concepts with partial causal relations to guide the generative language model for causal event sequence generation. We show that our approach can generate high-quality event sequences, filling a knowledge gap in the input KG. Furthermore, we explore how the generated sequences can be leveraged to discover useful and more complex structured knowledge from pattern mining and probabilistic event models. We release our sequence generation code and evaluation framework, as well as corpus of event sequence data.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
The paper aims to solve the problem of the lack of high - quality event - sequence data in some applications. Specifically, in some fields, such as news event analysis, automatically - extracted event sequences are often too messy and incomplete to be directly used for constructing effective event models. Therefore, the author explores the use of large - language models (LLMs) to generate structured event sequences to fill the knowledge gaps in these fields and further use the generated sequences for pattern mining and probability - event - model learning.
### Background and motivation
1. **Importance of event - sequence models**:
- Event - sequence models have wide applications in many fields such as finance, healthcare, and cybersecurity. For example, in the finance field, event models can predict stock - market trends; in the healthcare field, they can identify patterns in patient data to improve diagnosis and treatment plans; in the cybersecurity field, they can detect and prevent potential cyber - attacks by analyzing the event sequences leading to security vulnerabilities.
2. **Existing challenges**:
- In some applications, such as news event analysis, automatically - extracted event sequences usually contain multiple topics and participants, and many related or unrelated events may occur simultaneously or nearly simultaneously, which makes it difficult to automatically construct coherent event sequences.
- Existing methods, such as narrative extraction, topic detection and tracking, and timeline summarization, although having achieved certain success in some applications, usually generate short and noisy sequences, which are not suitable for constructing complex event models.
### Solutions
1. **Using large - language models to generate event sequences**:
- The author proposes a new mechanism to use large - language models (LLMs) to generate structured event sequences. By using event concepts and partial causal relationships in the knowledge graph (KG) to guide the generation process, the generated sequences can be used to construct probability - event models.
2. **Knowledge distillation**:
- The generated event sequences can be regarded as the event - sequence knowledge distilled from LLMs. These sequences not only fill the knowledge gaps in the input knowledge graph but also can be further used for pattern mining and probability - event - model learning, thereby extracting more complex and structured knowledge.
### Main contributions
1. **New iterative - context - prompt strategy**:
- A new iterative - context - prompt strategy is proposed for generating high - quality event sequences. To the author's knowledge, this is the first time that LLMs are used to generate structured event sequences for various event - model analyses.
2. **Generation of high - quality event sequences**:
- A set of high - quality event sequences is compiled using the generation mechanism, based on a carefully selected set of high - level event concepts (categories) from Wikidata.
3. **Evaluation framework**:
- An evaluation framework is developed, and through experiments, the value of the event sequences generated by LLM in replicating and enhancing the knowledge in structured representations (such as knowledge graphs) is demonstrated.
4. **Practical applications**:
- The practical uses of the generated event sequences in downstream pattern mining and probability - event models are further demonstrated.
### Conclusion
By using large - language models to generate structured event sequences, the paper provides a new method to solve the problem of the lack of high - quality event - sequence data in some applications. The generated sequences can not only be used to construct probability - event models but also can extract more complex and structured knowledge through pattern mining and probability - model learning.