A New Concept of Knowledge based Question Answering (KBQA) System for Multi-hop Reasoning

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4007 - 4017 July 10-15, 2022 ©2022 Association for Computational Linguistics A New Concept of Knowledge based Question Answering (KBQA) System for Multi-hop Reasoning Yu Wang Vijay Srinivasan Hongxia Jin Samsung Research America {yu.wang1,v.srinivasan,hongxia.jin}@samsung.com Abstract Knowledge based question answering (KBQA) is a complex task for natural language understanding. Many KBQA ap- proaches have been proposed in recent years, and most of them are trained based on labeled reasoning path. This hinders the system’s performance as many correct reasoning paths are not labeled as ground truth, and thus they cannot be learned. In this paper, we introduce a new concept of KBQA system which can leverage multiple reasoning paths’ information and only requires labeled answer as supervision. We name it as Mutliple Reasoning Paths KBQA System (MRP- QA). We conduct experiments on several benchmark datasets containing both single- hop simple questions as well as muti-hop complex questions, including WebQues- tionSP (WQSP), ComplexWebQuestion-1.1 (CWQ), and PathQuestion-Large (PQL), and demonstrate strong performance. 1 Introduction Knowledge-based question answering (KBQA) is the task of finding answers to questions by pro- cessing a structured knowledge base KB. A KB consists of a set of entities E, a set of relations R, and a set of literals S. A knowledge base fact is defined as (h, r, t), where h ∈ E is the head en- tity, t ∈ E � S is the tail entity/literal, and r ∈ R is the directed relation between h and t. To an- swer a simple single-relation question (i.e. a 1-hop question) such as: “Who is the president of the United States?”, a typical KBQA system first iden- tifies the entity (i.e. United States) and the rela- tion (i.e. “president”) asked in the question, and then searches for the answer entity by matching the entity-relation tuple <United States, president, ?> over KB. While a single-hop question can be answered by searching a predicate relation in KB, it is much harder to answer more complex multi-hop ques- tions containing multiple entities and relations with constraints. For instance, for complex composi- tional questions, it is not easy to extract all the relations correctly together with their head and tail entities in the right order. For complex conjunction questions that requires a conjunction of multiple ev- idences, it is even more difficult to correctly extract all the reasoning paths included. Most prior works on multi-hop KBQA focus on learning a single given ground truth reasoning path for each question, and outputting the most possible reasoning path during prediction (Zhou et al., 2018; Zhang et al., 2018; Yu et al., 2018; Lan et al., 2019). However, it is common that KB has many alterna- tive paths leading to the correct answer, of various reasoning qualities. These alternative reasoning paths are usually not provided as ground truth by the human annotators. For example, Figure 1 shows 7 reasoning paths pn = en 0 → rn …

Content truncated to 500 words