Multi-Hop Question Answering(MHQA)tasks require retrieving and reasoning over multiple relevant supporting facts to answer a question.However,existing MHQA models often rely on a single entity or fact to provide an an...Multi-Hop Question Answering(MHQA)tasks require retrieving and reasoning over multiple relevant supporting facts to answer a question.However,existing MHQA models often rely on a single entity or fact to provide an answer,rather than performing true multi-hop reasoning.Additionally,during the reasoning process,models may be influenced by multiple irrelevant factors,leading to broken reasoning chains and even incorrect answers.In recent years,causal inference-based methods have gained widespread attention in bias removal research.But existing models still perform poorly when dealing the complex causal biases hidden in multi-hop evidence.To address these challenge,we propose CausalBridgeQA,a novel method that integrates multi-hop question answering with causal relationships,effectively mitigating feature spurious correlations and the problem of broken reasoning chains.Specifically,we first extract causal relationships from the input text context,then compile these relationships into causal questions containing higher-level semantic information and feed them into MHQA reasoning system.Finally,we design a knowledge compensation mechanism in the reading comprehension module of the MHQA system to specifically address questions that are difficult to answer or frequently answered incorrectly,significantly improving the performance of MHQA tasks.Finally,a series of experiments conducted on three real-world QA datasets verified the effectiveness of our proposed method.展开更多
With more and more knowledge provided by WWW, querying and mining the knowledge bases have attracted much research attention. Among all the queries over knowledge bases, which are usually modelled as graphs, a keyword...With more and more knowledge provided by WWW, querying and mining the knowledge bases have attracted much research attention. Among all the queries over knowledge bases, which are usually modelled as graphs, a keyword query is the most widely used one. Although the problem of keyword query over graphs has been deeply studied for years, knowledge bases, as special error-tolerant graphs, lead to the results of the traditional defined keyword queries out of users＇ satisfaction. Thus, in this paper, we define a new keyword query, called confident r-clique, specific for knowledge bases based on the r-clique definition for keyword query on general graphs, which has been proved to be the best one. However, as we prove in the paper, finding the confident r-cliques is #P-hard. We propose a filtering-and-verification framework to improve the search efficiency. In the filtering phase, we develop the tightest upper bound of the confident r-clique, and design an index together with its search algorithm, which suits the large scale of knowledge bases well. In the verification phase, we develop an efficient sampling method to verify the final answers from the candidates remaining in the filtering phase. Extensive experiments demonstrate that the results derived from our new definition satisfy the users＇ requirement better compared with the traditional r-clique definition, and our algorithms are efficient.展开更多
基金supported by the National Natural Science Fundation of China(Grant No.62472027)Beijing Natural Science Foundation(No.U2241213).
文摘Multi-Hop Question Answering(MHQA)tasks require retrieving and reasoning over multiple relevant supporting facts to answer a question.However,existing MHQA models often rely on a single entity or fact to provide an answer,rather than performing true multi-hop reasoning.Additionally,during the reasoning process,models may be influenced by multiple irrelevant factors,leading to broken reasoning chains and even incorrect answers.In recent years,causal inference-based methods have gained widespread attention in bias removal research.But existing models still perform poorly when dealing the complex causal biases hidden in multi-hop evidence.To address these challenge,we propose CausalBridgeQA,a novel method that integrates multi-hop question answering with causal relationships,effectively mitigating feature spurious correlations and the problem of broken reasoning chains.Specifically,we first extract causal relationships from the input text context,then compile these relationships into causal questions containing higher-level semantic information and feed them into MHQA reasoning system.Finally,we design a knowledge compensation mechanism in the reading comprehension module of the MHQA system to specifically address questions that are difficult to answer or frequently answered incorrectly,significantly improving the performance of MHQA tasks.Finally,a series of experiments conducted on three real-world QA datasets verified the effectiveness of our proposed method.
基金Yu-Rong Cheng and Guo-Ren Wang are supported by the National Natural Science Foundation of China （NSFC） under Grant Nos. 61332006, 61332014, 61328202 and U1401256. Ye Yuan is supported by the NSFC under Grant No. 61572119 and the Fundamental Research Fudnds for the Central Universities of China under Grant Nos. N150402005 and N130504006. Lei Chen is supported by the NSFC under Grant No. 61328202.
文摘With more and more knowledge provided by WWW, querying and mining the knowledge bases have attracted much research attention. Among all the queries over knowledge bases, which are usually modelled as graphs, a keyword query is the most widely used one. Although the problem of keyword query over graphs has been deeply studied for years, knowledge bases, as special error-tolerant graphs, lead to the results of the traditional defined keyword queries out of users＇ satisfaction. Thus, in this paper, we define a new keyword query, called confident r-clique, specific for knowledge bases based on the r-clique definition for keyword query on general graphs, which has been proved to be the best one. However, as we prove in the paper, finding the confident r-cliques is #P-hard. We propose a filtering-and-verification framework to improve the search efficiency. In the filtering phase, we develop the tightest upper bound of the confident r-clique, and design an index together with its search algorithm, which suits the large scale of knowledge bases well. In the verification phase, we develop an efficient sampling method to verify the final answers from the candidates remaining in the filtering phase. Extensive experiments demonstrate that the results derived from our new definition satisfy the users＇ requirement better compared with the traditional r-clique definition, and our algorithms are efficient.
