History Semantic Graph Enhanced Conversational KBQA with Temporal Information Modeling

Hao Sun,Yang Li,Liwei Deng,Bowen Li,Binyuan Hui,Binhua Li,Yunshi Lan,Yan Zhang,Yongbin Li

DOI: https://doi.org/10.48550/arXiv.2306.06872

2023-06-12

Abstract:Context information modeling is an important task in conversational KBQA. However, existing methods usually assume the independence of utterances and model them in isolation. In this paper, we propose a History Semantic Graph Enhanced KBQA model (HSGE) that is able to effectively model long-range semantic dependencies in conversation history while maintaining low computational cost. The framework incorporates a context-aware encoder, which employs a dynamic memory decay mechanism and models context at different levels of granularity. We evaluate HSGE on a widely used benchmark dataset for complex sequential question answering. Experimental results demonstrate that it outperforms existing baselines averaged on all question types.

Computation and Language

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively model long - range semantic dependencies in Conversational KBQA (Conversational Knowledge Base Question Answering) while maintaining a low computational cost. Existing methods usually assume the independence between utterances and model them in isolation, which results in the inability to fully utilize the information in the historical conversation when dealing with multi - turn conversations, especially when the user asks questions involving coreference and ellipsis phenomena. To solve these problems, the paper proposes a Conversational Knowledge Base Question Answering model based on History Semantic Graph Enhancement (HSGE), which can effectively model long - range semantic dependencies in the conversation history and improve the efficiency and performance of the model through the dynamic memory attenuation mechanism and multi - granularity context modeling.

Specifically, HSGE solves the problem in the following ways:

1. **History Semantic Graph (HSG)**: Convert the logical forms of previous conversation turns into a history semantic graph, where the nodes are the entities mentioned in the conversation history and the edges are the relationships between these entities. By applying a Graph Neural Network (GNN) on the history semantic graph, the model can capture the complex interactions between entities, thus better understanding the conversation history.

2. **Context - aware Encoder**: Design a context - aware encoder. By introducing Temporal Embedding to deal with the phenomenon of the shift of the user conversation focus, the model can aggregate context information at different granularities, including the word level and the sentence level.

3. **Grammar - Guided Decoder**: Use the stacked Masked Attention mechanism to generate the logical form in a sequence format, and calculate the probability distribution of each word through a linear layer and softmax.

4. **Entity Recognition Module**: Include an entity detection module and an entity linking module to fill the entity slots in the predicted logical form.

5. **Concept - aware Attention Module**: Model the complex interactions between entity types and predicates and predict the correct entity type or predicate of the logical form.

Through these methods, HSGE was evaluated on the widely - used CSQA dataset, and the experimental results show that its average performance on all question types is better than that of the existing baseline models.