Complex Knowledge Base Question Answering: A Survey

Abstract

Knowledge base question answering (KBQA) aims to answer a question over a knowledge base (KB). Early studies mainly focused on answering simple questions over KBs and achieved great success. However, their performances on complex questions are still far from satisfactory. Therefore, in recent years, researchers propose a large number of novel methods, which looked into the challenges of answering complex questions. In this survey, we review recent advances in KBQA with the focus on solving complex questions, which usually contain multiple subjects, express compound relations, or involve numerical operations. In detail, we begin with introducing the complex KBQA task and relevant background. Then, we present two mainstream categories of methods for complex KBQA, namely semantic parsing-based (SP-based) methods and information retrieval-based (IR-based) methods. Specifically, we illustrate their procedures with flow designs and discuss their difference and similarity. Next, we summarize the challenges that these two categories of methods encounter when answering complex questions, and explicate advanced solutions as well as techniques used in existing work. After that, we discuss the potential impact of pre-trained language models (PLMs) on complex KBQA. To help readers catch up with SOTA methods, we also provide a comprehensive evaluation and resource about complex KBQA task. Finally, we conclude and discuss several promising directions related to complex KBQA for future research.

Research and Applications Analysis of Knowledge Base Question Answering

“…(3) Question representation It is to get semantical information of the question by representing it as reasoning instructions through using neural network together with other methods [5].…”

“…The input natural language question is transformed into the form of tokens; thus, it could be understood by computers. The subject is recognized and is executed against the topic entity that is the counterpart of the subject to figure out the answer derived from the neighborhood of topic entity [5]. The principle of approaches in KBQA will be discussed in Section 4.…”

“…However, considering the real applications of question answering systems, most questions are multihop, so that the question consists of multiple subjects and multiple relations such as "When was the founder of Harvard University born?". The entity of the answer is often multiple hops or even multiple aggregation away from the topic entity [5].…”

“…There are two main approaches for KBQA known as Information Retrieval-based (IR-based) methods and Semantic Parsing-based (SP-based) methods [5]. The former one works as generating a subgraph related to the natural language question from the knowledge graph, retrieving entities of candidate answers from subgraph, and applying ranking generator to get the entity with the highest rank, which is obtained as the predicated answer [5].…”

“…There are two main approaches for KBQA known as Information Retrieval-based (IR-based) methods and Semantic Parsing-based (SP-based) methods [5]. The former one works as generating a subgraph related to the natural language question from the knowledge graph, retrieving entities of candidate answers from subgraph, and applying ranking generator to get the entity with the highest rank, which is obtained as the predicated answer [5]. The latter one works as transferring the unstructured natural language question into a query language [7] that can be executed against knowledge base to output the answer [5,8].…”