Temporal Knowledge Graph (TKG) is an extension of regular knowledge graph by
attaching the time scope. Existing temporal knowledge graph question answering
(TKGQA) models solely approach simple questions, owing to the prior assumption
that each question only contains a single temporal fact with explicit/implicit
temporal constraints. Hence, they perform poorly on questions which own
multiple temporal facts. In this paper, we propose textbf{underline{J}}oint
textbf{underline{M}}ulti textbf{underline{F}}acts
textbf{underline{R}}easoning textbf{underline{N}}etwork (JMFRN), to jointly
reasoning multiple temporal facts for accurately answering emph{complex}
temporal questions. Specifically, JMFRN first retrieves question-related
temporal facts from TKG for each entity of the given complex question. For
joint reasoning, we design two different attention (ie entity-aware and
time-aware) modules, which are suitable for universal settings, to aggregate
entities and timestamps information of retrieved facts. Moreover, to filter
incorrect type answers, we introduce an additional answer type discrimination
task. Extensive experiments demonstrate our proposed method significantly
outperforms the state-of-art on the well-known complex temporal question
benchmark TimeQuestions.
In this article, the authors discuss the limitations of existing temporal knowledge graph question answering (TKGQA) models and propose a new approach called Joint Multi Facts Reasoning Network (JMFRN) to address these limitations. The authors highlight the importance of considering multiple temporal facts in complex temporal questions and argue that existing models perform poorly on such questions due to the assumption that each question only contains a single temporal fact.
The JMFRN model aims to overcome this limitation by jointly reasoning multiple temporal facts for accurately answering complex temporal questions. The model first retrieves question-related temporal facts from the Temporal Knowledge Graph (TKG) for each entity mentioned in the question. To facilitate joint reasoning, the authors design two attention modules – one entity-aware and one time-aware – to aggregate information from the retrieved facts. These attention modules help the model effectively capture and incorporate entities and timestamps in the reasoning process.
Additionally, the authors introduce an answer type discrimination task to filter incorrect answers. This task helps improve the accuracy of the model by ensuring that it not only provides correct answers but also answers that are of the correct type.
The proposed method is evaluated on the TimeQuestions benchmark and compared against state-of-the-art models. The experimental results demonstrate that JMFRN significantly outperforms other models in answering complex temporal questions.
From a multidisciplinary perspective, this research combines concepts from natural language processing, knowledge graphs, and temporal reasoning. By incorporating temporal information into the knowledge graph, the authors enhance the capability of existing question answering models to handle complex temporal queries. This work has implications for various domains where understanding and reasoning about time-based information is crucial, such as historical analysis, event prediction, and planning.
In conclusion, the JMFRN model represents a promising approach for addressing the challenges of answering complex temporal questions in the context of knowledge graphs. The attention modules and answer type discrimination task contribute to improved performance, and the experimental results validate the effectiveness of the proposed method. This research highlights the importance of considering multiple temporal facts and the multi-disciplinary nature of incorporating temporal reasoning into question answering systems.
Read the original article