A Semantic Parsing and Reasoning-Based Approach to Knowledge Base Question Answering
Open accessAbstract
Knowledge Base Question Answering (KBQA) is a task where existing techniques have faced significant challenges, such as the need for complex question understanding, reasoning, and large training datasets. In this work, we demonstrate Deep Thinking Question Answering (DTQA), a semantic parsing and reasoning-based KBQA system. DTQA (1) integrates multiple, reusable modules that are trained specifically for their individual tasks (e.g. semantic parsing, entity linking, and relationship linking), eliminating the need for end-to-end KBQA training data; (2) leverages semantic parsing and a reasoner for improved question understanding. DTQA is a system of systems that achieves state-of-the-art performance on two popular KBQA datasets.
AI snapshot
Written from the paper’s full text.
TL;DR
DTQA is a semantic parsing and reasoning-based KBQA system that parses questions into AMR, aligns AMR to KB triples, converts them to logic forms, and uses a Logical Neural Network reasoner. It reports state-of-the-art performance on QALD-9 and LC-QuAD 1.0.
Key findings
- DTQA reports state-of-the-art performance on QALD-9 and LC-QuAD 1.0, outperforming graph-driven approaches such as gAnswer, WDAqua, and QAmp.
- On QALD-9, DTQA reports P=31.41, R=32.16, F=30.88, and Macro-F1-QALD=45.33; gAnswer reports P=29.34, R=32.68, F=29.81, F1=42.96; WDAqua reports P=26.09, R=26.7, F=24.99, F1=28.87.
- On LC-QuAD 1.0, DTQA reports P=33.94, R=34.99, F=33.72; QAmp reports P=25.00, R=50.00, F=33.00; WDAqua reports P=22.00, R=38.00, F=28.00.
- DTQA integrates reusable modules trained for individual tasks—semantic parsing, entity linking, and relationship linking—eliminating the need for end-to-end KBQA training data.
Methods
- DTQA is a pipeline: AMR parsing, entity extraction/linking with BERT-based mention detection and BLINK, path-based triple generation, relation extraction/linking with SLING, rule-based AMR-to-Logic conversion, and reasoning with Logical Neural Networks (LNN) using SPARQL integration.
- Evaluation uses QALD-9 (150 test questions) and LC-QuAD 1.0 (1,000 test questions), comparing against GAnswer, QAmp, and WSDAqua-core1.
- Metrics are precision, recall, F-score, and Macro-F1-QALD.
Limitations
- The paper does not state a limitations section.
- Evident from the text: evaluation is limited to two datasets and three baselines, with no ablation or error analysis reported.
- The reported table shows QAmp has higher recall on LC-QuAD 1.0 (50.00 vs. DTQA’s 34.99), despite the paper’s claim that DTQA achieves state-of-the-art on all metrics.
- The system relies on multiple modules and manually added generic axioms for geographic reasoning.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
8 of 11 cited sources checked · 5 supported · 2 flagged · accuracy not yet measured
- Premisecitation
Existing KBQA systems do not have a modular reasoner to perform logical reasoning using information available in the KB.
“However, existing KBQA systems (Diefenbach et al. 2020;Vakulenko et al. 2019;Sun et al. 2020) do not have a modular reasoner to perform logical reasoning using the information available in the knowledge base.”
Quote in the paperSubstantial claim2 of 3 sources checked- cites Diefenbach et al. 2020
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites Vakulenko et al. 2019 source
The Introduction (p. 2) says a core aspect of QAmp is disentangling reasoning from question interpretation. Section 3.2 (p. 4) describes a separate answer-inference process that traverses the knowledge graph and passes messages. This contradicts the claim insofar as it presents this cited system as lacking a modular reasoner, although the paper describes its reasoning as handling uncertainty rather than as formal logical reasoning. (LLM, conf=0.81)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured - cites Sun et al. 2020 source
The Introduction (p. 2) calls EmQL a module in a KBQA system. Section 4.2 (p. 7) explicitly describes embedding EmQL as a reasoning component and says an earlier KBQA model also had a reasoner, ReifKB. The source therefore contradicts the claim that existing KBQA systems lack a modular reasoner. (LLM, conf=0.90)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured
- Rests on an assumption: Some questions in KBQA datasets require reasoning.
- Findingown data
DTQA achieves state-of-the-art performance on two popular KBQA datasets (QALD-9 and LC-QuAD-1).
“DTQA is a system of systems that achieves state-of-the-art performance on two popular KBQA datasets.”
- Also stated in: § Experimental Evaluation · plain
“DTQA achieves stateof-the-art performance on all metrics, outperforming existing graph-driven approaches such as gAnswer, WDAqua, and QAmp on both datasets.”
- Also stated in: § Demonstration and Conclusion · plain
“DTQA is a system of systems trained for different subproblems that has achieved the state-of-the art performance on two prominent KBQA datasets.”
Quote in the paperSubstantial claim - Also stated in: § Experimental Evaluation · plain
- Interpretationreasoning
DTQA demonstrates an interpretable methodology for KBQA.
“This demonstration shows an interpretable methodology for KBQA that achieves state-of-the-art results on two prominent datasets (QALD-9, LC-QuAD-1).”
- Also stated in: § Demonstration and Conclusion · boosted
“The system demonstrates an interpretable methodology for understanding the natural language question and reasoning to infer the answer from the KB.”
Quote in the paperSubstantial claim - Also stated in: § Demonstration and Conclusion · boosted
- Interpretationreasoning
By utilizing AMR to get the semantic representation of the question, DTQA is able to generalize to sentence structures that come from very different distributions.
“By utilizing AMR to get the semantic representation of the question, DTQA is able to generalize to sentence structures that come from very different distributions, and achieve state-of-the-art performance on both datasets.”
Quote in the paperSubstantial claim - Interpretationreasoning
DTQA uses task-general semantic parsing via AMR.
“DTQA is a system of systems that demonstrates: (1) the use of task-general semantic parsing via AMR;”
Quote in the paperSubstantial claim- Rests on an assumption: AMR enables task-independent language understanding.
- Interpretationreasoning
The path-based triple generation finds paths between the question amr-unknown and every linked entity, avoiding unnecessary triples.
“The main idea of this module is to find paths between the question amr-unknown to every linked entity and hence avoid generating unnecessary triples.”
Quote in the paperSubstantial claim - Interpretationreasoning
DTQA uses a neuro-symbolic reasoner for the KBQA task.
“(3) the use of a neuro-symbolic reasoner for the KBQA task.”
Quote in the paperSubstantial claim- Rests on an assumption: Some questions in KBQA datasets require reasoning.
- Interpretationnone given
The AMR to Logic module is a rule-based system that transforms KB-aligned AMR paths to a logical formalism supporting binary predicates and higher-order functional predicates.
“The AMR to Logic module is a rule-based system that transforms the KB-aligned AMR paths to a logical formalism to support binary predicates and higher-order functional predicates (e.g. for aggregation and manipulation of sets).”
Quote in the paperSubstantial claim - Interpretationreasoning
The path-based triple generation collapses multiple consecutive predicates into a single edge to address granularity mismatch.
“To address this granularity mismatch, the path-based triple generation collapses multiple predicates occurring consecutively into a single edge.”
Quote in the paperSubstantial claim - Interpretationreasoning
DTQA uses a novel Path-based Triple Generation module that transforms the AMR graph with linked entities into triples where each has one-to-one correspondence with constraints in the SPARQL query.
“In order to transform the n-ary predicates to binary, we use a novel 'Path-based Triple Generation' module that transforms the AMR graph (with linked entities) into a set of triples where each has one to one correspondence with constraints in the SPARQL query.”
Quote in the paperSubstantial claim - Interpretationnone given
This logic formalism has a deterministic mapping to SPARQL by introducing SPARQL constructs such as COUNT, ORDER BY, and FILTER.
“This logic formalism has a deterministic mapping to SPARQL by introducing SPARQL constructs such as COUNT, ORDER BY, and FILTER.”
Quote in the paperNot a substantial claim - Premisereasoning
Aligning the AMR graph to the underlying KB is a challenge due to structural and granularity mismatch between AMR and KBs such as DBpedia.
“Aligning the AMR graph to the underlying KB is a challenge, particularly due to structural and granularity mismatch between AMR and KBs such as DBPedia.”
Quote in the paperSubstantial claim- Rests on an assumption: AMR parses are KB independent.
- Interpretationreasoning
DTQA enables opportunities for reasoning by transforming text to a logic form that can be used by LNN.
“DTQA enables opportunities for reasoning by transforming text to a logic form that can be used by LNN (a neuro-symbolic reasoner).”
Quote in the paperSubstantial claim- Rests on an assumption: Some questions in KBQA datasets require reasoning.
- Interpretationreasoning
DTQA aligns AMR to KB triples.
“(2) an approach that aligns AMR to KB triples;”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 7
- May combine two claims. 5
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationSLING's most relevant and novel approach is statistical mapping of an AMR predicate to its semantically equivalent relation from the underlying KB using a distant supervision dataset.
“The most relevant and novel approach in SLING is its statistical mapping of an AMR predicate to its semantically equivalent relation from the underlying KB using a distant supervision dataset.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites Mihindukulasooriya et al. 2020 source
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationDTQA uses SLING, the state-of-the-art REL system for KBQA.
“For this task, DTQA uses SLING (Mihindukulasooriya et al. 2020); the state-of-the-art REL system for KBQA.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites Mihindukulasooriya et al. 2020 source
Classified as probably not a substantial claim.
PremisecitationLNN is a neural network architecture in which neurons model a rigorously defined notion of weighted fuzzy or classical first-order logic.
“LNN is a neural network architecture in which neurons model a rigorously defined notion of weighted fuzzy or classical first-order logic.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Riegel et al. 2020 source
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationSLING relies on an ensemble of complementary relation linking approaches to produce a top-k list of KB relations for each AMR predicate.
“It relies on an ensemble of complementary relation linking approaches to produce a top-k list of KB relations for each AMR predicate.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites Mihindukulasooriya et al. 2020 source
May combine two claims.
PremisecitationExisting KBQA approaches face challenges including complex question understanding, scarcity of end-to-end training data, and the need for reasoning.
“Knowledge Base Question Answering (KBQA) is a task where existing techniques have faced significant challenges, such as the need for complex question understanding, reasoning, and large training datasets.”
- Also stated in: § Introduction · plain
“Both categories still suffer from a number of challenges, including complex question understanding, scarcity of end-to-end training data, and the need for reasoning.”
Quote in the paperSubstantial claim2 of 3 sources checked- cites Diefenbach et al. 2020
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites Vakulenko et al. 2019 source
- cites Sun et al. 2020 source
Section 4.2 (p. 7) describes KBQA as taking natural-language questions and using a reasoning component; §4.1 (p. 7) discusses difficulties with complex queries. The paper reports training-data sizes and a benefit from pre-training (§4.2, p. 9), but does not characterize scarce end-to-end data or a need for large training datasets as a challenge for existing KBQA techniques. (LLM, conf=0.82)
Experimental — accuracy not yet measured
- Also stated in: § Introduction · plain
May combine two claims.
Findingown dataDTQA outperforms existing graph-driven approaches (gAnswer, WDAqua, QAmp) on both QALD-9 and LC-QuAD 1.0 on all reported metrics.
“DTQA achieves stateof-the-art performance on all metrics, outperforming existing graph-driven approaches such as gAnswer, WDAqua, and QAmp on both datasets.”
Quote in the paperSubstantial claimMay combine two claims.
InterpretationreasoningLNN retrieves predicate groundings via its granular SPARQL integration and performs type-based and geographic reasoning based on ontological information available in the KB.
“LNN retrieves predicate groundings via its granular SPARQL integration and performs type-based, and geographic reasoning based on the ontological information available in the knowledge base.”
Quote in the paperSubstantial claim- Rests on an assumption: Some questions in KBQA datasets require reasoning.
May combine two claims.
InterpretationreasoningDTQA integrates multiple reusable modules trained specifically for individual tasks (semantic parsing, entity linking, and relationship linking), eliminating the need for end-to-end KBQA training data.
“DTQA (1) integrates multiple, reusable modules that are trained specifically for their individual tasks (e.g.semantic parsing, entity linking, and relationship linking), eliminating the need for end-to-end KBQA training data;”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
PremisecitationAMR is a semantic parse representation that solves the ambiguity of natural language by representing syntactically different sentences with the same underlying meaning in the same way.
“AMR is a semantic parse representation that solves the ambiguity of natural language by representing syntactically different sentences with the same underlying meaning in the same way.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Banarescu et al. 2013
Probably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningLNN allowed DTQA to manually add generic axioms for geographic reasoning scenarios.
“LNN allowed us to manually add generic axioms for the geographic reasoning scenarios.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based).
Premisenone givenAMR parse provides relationship information that is generally more fine-grained than the KB.
“Furthermore, AMR parse provides relationship information that is generally more fine-grained than the KB.”
Quote in the paperNot a substantial claimMay combine two claims.
Interpretationnone givenDTQA is the first system that successfully harnesses a generic semantic parser with a neuro-symbolic reasoner for a KBQA task and a novel path-based approach to map AMR to the underlying KG.
“To the best of our knowledge, DTQA is the first system that successfully harnesses a generic semantic parser with a neuro-symbolic reasoner for a KBQA task and a novel path-based approach to map AMR to the underlying KG.”
Quote in the paperNot a substantial claim
The paper relies on this without support.
AMR parses are KB independent.
“AMR parses are KB independent, however an essential task to solve KBQA using AMR is to align the AMR parses with the knowledge base.”
The paper relies on this without support.
AMR enables task-independent language understanding.
“The use of AMR enables task-independent language understanding, which DTQA leverages to answer structurally complex questions.”
The paper relies on this without support.
Some questions in KBQA datasets require reasoning.
“Some questions in KBQA datasets require reasoning.”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
DTQA does not lead on recall in either dataset
MajorThe claim that DTQA outperforms the other systems on all metrics conflicts with Table 1. DTQA's reported recall is 32.16 versus gAnswer's 32.68 on QALD, and 34.99 versus QAMP's 50.00 and WDAqua's 38.00 on LC-QuAD.
“DTQA achieves stateof-the-art performance on all metrics, outperforming existing graph-driven approaches such as gAnswer, WDAqua, and QAmp on both datasets.”
“| gAnswer | QALD | 29.34 32.68 29.81 42.96 | | | | | DTQA | QALD | 31.41 32.16 30.88 45.33 | | | | | WDAqua LC-QuAD 22.00 38.00 28.00 - | | | | | | | QAMP | LC-QuAD 25.00 50.00 33.00 - | | | | | | DTQA | LC-QuAD 33.94 34.99 33.72 - | | | | |”
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 48,293 · cited 2× · see citing works
“In DTQA, we trained a BERT-based neural mention detection and used BLINK (Devlin et al. 2018) for disambiguation of named entities.”
MethodGlove: Global Vectors for Word Representation
Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 1× · see citing works
“This is captured by a neural-model and an unsupervised semantic similarity module that measures similarity of each AMR predicate with a list of possible relations using GloVe (Pennington, Socher, and Manning 2014) embeddings.”
MethodDBpedia: A Nucleus for a Web of Open Data
Auer, S., Bizer, C., Kobilarov, G., Lehmann, J., Richard Cyganiak, A., & Ives, Z. G. (2007) · Lecture notes in computer science · doi:10.1007/978-3-540-76298-0_52 · Cited by 4,847 · cited 5× · see citing works
“Therefore, in the next three modules we align the AMR parse to a DBpedia sub-graph that can be transformed into a SPARQL query.”
BackgroundInfluentialVrandečić, D., & Krötzsch, M. (2014) · Communications of the ACM · doi:10.1145/2629489 · Cited by 3,367 · see citing works
(2013) · Cited by 1,117 · see citing works
LC-QuAD: A Corpus for Complex Question Answering over Knowledge Graphs
Trivedi, P., Maheshwari, G., Dubey, M., & Lehmann, J. (2017) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-68204-4_22 · Cited by 261 · cited 1× · see citing works
“To evaluate DTQA , we use two prominent KBQA datasets; QALD - 9 (Ngomo 2018) and LC-QuAD 1.0 (Trivedi et al. 2017).”
MethodInfluentialNatural language question answering over RDF
Zou, L., Huang, R., Wang, H., Yu, J. X., He, W., & Zhao, D. (2014) · ACM SIGMOD International Conference on Management of Data (SIGMOD) · doi:10.1145/2588555.2610525 · Cited by 239 · cited 1× · see citing works
“We compare DTQA against three KBQA systems: 1) GAn-swer (Zou et al. 2014) is a graph-driven approach and the state-of-the-art on QALD dataset, 2) QAmp (Vakulenko et al. 2019) is another graph-driven approach based on message passing.”
MethodUsbeck, R., Röder, M., Ngomo, A.-C. N., Baron, C., Both, A., Brümmer, M., Ceccarelli, D., Cornolti, M., Cherix, D., Eickmann, B., et al. (2015) · International World Wide Web Conference (WWW) · doi:10.1145/2736277.2741626 · Cited by 191 · cited 1× · see citing works
“We also report Macro-F1-QALD (Usbeck et al. 2015), a recommended metric to use for evaluating QALD dataset.”
MethodTowards a question answering system over the Semantic Web
Diefenbach, D., Both, A., Singh, K., & Maret, P. (2019) · Semantic Web · doi:10.3233/sw-190343 · Cited by 118 · cited 3× · see citing works
“QAmp is also state-of-the-art on LC-QuAD dataset, and 3) WSDAqua-core1 (Diefenbach et al. 2020) which to the best of our knowledge, is the only approach evaluated on both QALD and LC-QuAD.”
BackgroundMethodRiegel, R., Gray, A., Luus, F., Khan, N., Makondo, N., Akhalwaya, I. Y., Qian, H., Fagin, R., Barahona, F., Sharma, U., et al. (2020) · arXiv · doi:10.48550/arxiv.2006.13155 · Cited by 95 · cited 2× · see citing works
“Finally, the logic forms are used by Logical Neural Network (LNN) (Riegel et al. 2020), a neuro-symbolic reasoner that enables DTQA to perform different types of reasoning.”
MethodSelf-citationRewarding Smatch: Transition-Based AMR Parsing with Reinforcement Learning
Naseem, T., Shah, A., Wan, H., Florian, R., Roukos, S., & Ballesteros, M. (2019) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p19-1451 · Cited by 60 · cited 1× · see citing works
“We use the current state-of-the-art system for AMR that leverage transition-based (Naseem et al. 2019) parsing approach, parameterized with neural networks and trained in an end-to-end fashion.”
MethodSelf-citationFaithful Embeddings for Knowledge Base Queries
Sun, H., Arnold, A. O., Bedrax-Weiss, T., Pereira, F., & Cohen, W. W. (2020) · arXiv · doi:10.48550/arxiv.2004.03658 · Cited by 51 · see citing works
Leveraging Semantic Parsing for Relation Linking over Knowledge Bases
Mihindukulasooriya, N., Rossiello, G., Kapanipathi, P., Abdelaziz, I., Ravishankar, S., Yu, M., Gliozzo, A., Roukos, S., & Gray, A. (2020) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-62419-4_23 · Cited by 24 · cited 4× · see citing works
“Apart from this statistical co-occurrence information, the rich lexical information in the AMR predicate and question text is also leveraged by SLING.”
MethodInfluentialSelf-citationA Thematic Hierarchy for Efficient Generation from Lexical-Conceptual Structure
Dorr, B. J., Habash, N. Y., & Traum, D. R. (1998) · Lecture notes in computer science · doi:10.1007/3-540-49478-2_30 · Cited by 22 · cited 1× · see citing works
“The input to DTQA is the question text, which is first parsed using the Abstract Meaning Representation (AMR) parser (Banarescu et al. 2013; Dorr, Habash, and Traum 1998).”
MethodMessage Passing for Complex Question Answering over Knowledge Graphs
Vakulenko, S., Fernandez Garcia, J. D., Polleres, A., de Rijke, M., & Cochez, M. (2019) · Proceedings of the 28th ACM International Conference on Information and Knowledge Management · doi:10.1145/3357384.3358026 · Cited by 14 · cited 4× · see citing works
“QAmp is also state-of-the-art on LC-QuAD dataset, and 3) WSDAqua-core1 (Diefenbach et al. 2020) which to the best of our knowledge, is the only approach evaluated on both QALD and LC-QuAD.”
MethodInfluentialA Thematic Hierarchy for Efficient Generation from Lexical-Conceptual Structure
Dorr, B. J., Habash, N. Y., & Traum, D. R. (1998) · doi:10.21236/ada458673 · Cited by 8 · see citing works
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805
Rewarding Smatch: Transition-Based AMR Parsing with Reinforcement Learning
Naseem, T., Shah, A., Wan, H., et al. (2019) · arXiv · doi:10.48550/arxiv.1905.13370
Flagged first, then by citations
Cited by
Sorted by citations
A Survey on Deep Learning Event Extraction: Approaches and Applications
Li, Q., Li, J., Sheng, J., Cui, S., Wu, J. X., Hei, Y., Peng, H. M., Guo, S., Wang, L., Beheshti, A., et al. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3213168 · Cited by 105 · see citing works
“From a technical perspective, EE can depend on some other foundational natural language processing (NLP) tasks such as named entity recognition (NER) [26], [27], [28], semantic parsing [29], [30], [31], and relationship extraction [32], [33], [34].”
BackgroundLeveraging Abstract Meaning Representation for Knowledge Base Question Answering
Kapanipathi, P., Abdelaziz, I., Ravishankar, S., Roukos, S., Gray, A., Astudillo, R. F., Chang, M., Cornelio, C., Dana, S., Fokoue, A., et al. (2021) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2021.findings-acl.339 · Cited by 85 · see citing works
“KBQA requires a system to answer a natural language question based on facts available in a Knowledge Base (KB) (Zou et al., 2014; Vakulenko et al., 2019; Diefenbach et al., 2020; Abdelaziz et al., 2021).”
BackgroundSelf-citationA Toolkit for Generating Code Knowledge Graphs
Abdelaziz, I., Dolby, J., McCusker, J., & Srinivas, K. (2021) · Proceedings of the 11th Knowledge Capture Conference · doi:10.1145/3460210.3493578 · Cited by 46 · see citing works
“These knowledge graphs have provided significant advantages in a number of different application areas, such as semantic parsing [20], recommendation systems [13], information retrieval [14], and question answering [1, 34, 36].”
BackgroundSelf-citationQuestion Answering over Knowledge Bases by Leveraging Semantic Parsing and Neuro-Symbolic Reasoning.Preprint
Kapanipathi, P., Abdelaziz, I., Ravishankar, S., Roukos, S., Gray, A., Astudillo, R. F., Chang, M., Cornelio, C., Dana, S., Fokoue, A., et al. (2020) · arXiv (Cornell University) · Cited by 24 · see citing works
Self-citationA Two-Stage Approach towards Generalization in Knowledge Base Question Answering
Ravishankar, S., Thai, D., Abdelaziz, I., Mihindukulasooriya, N., Naseem, T., Kapanipathi, P., Rossiello, G., & Fokoue, A. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.408 · Cited by 22 · see citing works
“In particular GAnswer, NSQA, and QAmp are specific to DBpedia and do not evaluate their approaches on any other KGs.”
BackgroundMethodInfluentialSelf-citationInductive Relation Prediction with Logical Reasoning Using Contrastive Representations
Pan, Y., Liu, J., Zhang, L., Zhao, T., Lin, Q., Hu, X., & Wang, Q. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.286 · Cited by 22 · see citing works
A knowledge inference model for question answering on an incomplete knowledge graph
Guo, Q., Wang, X., Zhu, Z., et al. (2022) · Applied Intelligence · doi:10.1007/s10489-022-03927-0 · Cited by 19 · see citing works
Implicit Event Argument Extraction With Argument-Argument Relational Knowledge
Wei, K., Sun, X., Zhang, Z., Jin, L., Zhang, J., Jianwei, L. V., & Guo, Z. M. (2022) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2022.3218830 · Cited by 15 · see citing works
“[5] knowledge base construction [6], [7], and event graph [8], [9], etc.”
BackgroundPay More Attention to Relation Exploration for Knowledge Base Question Answering
Cao, Y., Li, X., Liu, H., Dai, W., Chen, S., Wang, B., Chen, M., & Hershcovich, D. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.133 · Cited by 11 · see citing works
Lv, J., Zhang, Z., Jin, L., Li, S., Li, X., Xu, G., & Sun, X. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.109480 · Cited by 8 · see citing works
Chen, Y.-H., Lu, E. J., & Cheng, K.-H. (2025) · Journal of Web Semantics · doi:10.1016/j.websem.2025.100869 · Cited by 7 · see citing works
Entity neighborhood awareness and hierarchical message aggregation for inductive relation prediction
Zeng, D., Huang, T., Zhang, Z., & Jiang, L. (2024) · Information Processing & Management · doi:10.1016/j.ipm.2024.103737 · Cited by 5 · see citing works
Deep Learning Schema-based Event Extraction: Literature Review and Current Trends.Preprint
Li, Q., Peng, H. M., Li, J., Hei, Y., Sun, R. Q., Sheng, J., Guo, S., Wang, L., & Yu, P. S. (2021) · arXiv (Cornell University) · Cited by 5 · see citing works
Generative Event Extraction via Internal Knowledge-Enhanced Prompt Learning
Song, H., Zhu, Q., Yu, Z., Liang, J., & He, H. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-44192-9_8 · Cited by 3 · see citing works
Sataer, Y., Gao, Z., Fan, Y., Li, B., Gao, M., & Shi, C. (2023) · Applied Intelligence · doi:10.1007/s10489-023-05120-3 · Cited by 3 · see citing works
“AMR has been proven useful for several downstream tasks of natu-ral language processing, including text summarization [2], information extraction [3], machine translation [4], dialogue systems [5] and question answering [6].”
Longwell, J., Alavi, M. A. A., Zarrinkalam, F., et al. (2024) · Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region (SIGIR-AP) · doi:10.1145/3673791.3698426 · Cited by 3 · see citing works
“We compare the performance of our methods with the following state-of-the-art KGQA baselines: QAmp [25], TQAS [9], DTQA [1], EQP [7], SPARQL-QA-v2 [4], Slot-Matching Dot [18], AQGG [30], SParseQA [2], and QAmp [25].”
Enhancing SPARQL Query Performance With Recurrent Neural Networks
Chen, Y.-H., Lu, E. J., & Lin, J.-D. (2023) · IEEE Access · doi:10.1109/access.2023.3308691 · Cited by 2 · see citing works
“, QAMP [33] and DTQA [34], respectively.”
InfluentialVisKoP: Visual Knowledge oriented Programming for Interactive Knowledge Base Question Answering
Yao, Z., Chen, Y., Lv, X., Cao, S., Xin, A., Yu, J., Jin, H., Xu, J., Zhang, P., Hou, L., et al. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-demo.17 · Cited by 2 · see citing works
Automatic annotation for accurate text-to-SPARQL translation using hybrid encoder–decoder models
Chen, Y.-H., Lu, E. J., & Hsu, C. (2025) · The Journal of Supercomputing · doi:10.1007/s11227-025-08127-4 · Cited by 1 · see citing works
“Table 12 presents a comparison between the proposed method and several state-of-the-art end-to-end question answering systems evaluated on the LC-QuAD 1.0 dataset [15, 20, 22, 26, 27].”
InfluentialEnhancing Large Language Models Reasoning via Multi-Path Optimization on Knowledge Graph
Liao, J., Liu, C., Ding, Y., Wang, H., Tang, Z., Li, K., & Li, K. (2026) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2026.3682046 · Cited by 1 · see citing works
“KGQA as a typical task for evaluatingthecollaborativereasoningofLLMsandKGs,mainly adoptstwoparadigms:Semanticparsingapproaches[15],[16] thatconvertnaturallanguagequestionsintostructuredqueries, andretrieval-augmentedmethods[7],[10]thatassistLLM…”
A Knowledge Graph Question-Answering Platform Trained Independently of the Graph.
Omar, R., Dhall, I., Sheikh, N., et al. (2021) · Cited by 1 · see citing works
Chinese Sentence Matching with Multiple Alignments and Feature Augmentation
Zuo, Y., Peng, X., Lü, W., Wang, S., Li, Z., Zhang, W., & Zhai, Y. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892521 · Cited by 1 · see citing works
“Chinese sentence matching plays an essential role in various down-stream tasks, such as question answering [1], [2], machine reading comprehension [3], [4] and natural language inference [5], [6], and so on.”
BackgroundA Comprehensive Survey on Schema-based Event Extraction with Deep LearningPreprint
Li, Q., Peng, H. M., Li, J., Hei, Y., Sun, R. Q., Sheng, J., Guo, S., Wang, L., Wu, J. X., Beheshti, A., et al. (2021) · arXiv (Cornell University) · Cited by 1 · see citing works
A LLM Enhanced Two-Stage Approach for Multi-hop Knowledge Graph Reasoning
Han, D., Zhao, Z., & Liu, C. (2025) · Communications in computer and information science · doi:10.1007/978-981-96-2376-1_23
Event-Structured LLM Planning for Continuous-Parameter Industrial Workflows
Wei, H., & Zhang, Z. (2026) · International Conference on Mechatronics Technology and Intelligent Manufacturing (ICMTIM) · doi:10.1109/icmtim69588.2026.11525771
“Technically, event extraction relies on several core NLP tasks, including named entity recognition (NER)[11], [12], semantic parsing [13], and relation extraction [14], [15].”
Leveraging Abstract Meaning Representation for Knowledge Base Question Answering
Kapanipathi, P., Abdelaziz, I., Ravishankar, S., Roukos, S., Gray, A., Astudillo, R., Chang, M., Cornelio, C., Dana, S., Fokoue, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2012.01707 · Cited by 9 · see citing works
“KBQA requires a system to answer a natural language question based on facts available in a Knowledge Base (KB) (Zou et al., 2014; Vakulenko et al., 2019; Diefenbach et al., 2020; Abdelaziz et al., 2021).”
BackgroundSelf-citation
- Authors
- Abdelaziz, I., Ravishankar, S., Kapanipathi, P., Roukos, S., & Gray, A.
- Affiliations
- IBM (United States) (US)
- Venue
- Proceedings of the AAAI Conference on Artificial Intelligence
- Published
- 2021-05-18
- Volume
- 35
- Issue
- 18
- Pages
- 15985-15987
- ISSN
- 2159-5399 (ISSN-L)
- 2374-3468
- OpenAlex
- W3176061953
- Cited by
- 26
- References
- 18
- Citations by year
- 2026: 2 · 2025: 3 · 2024: 2 · 2023: 5 · 2022: 7 · 2021: 5 · 2020: 2
- FWCI
- 2.19
- Type
- conference paper
- Language
- en
- Open access
- diamond
- doi.org
- Locations
- Proceedings of the AAAI Conference on Artificial Intelligence (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Semantic Web and Ontologies
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- knowledge base question answering, semantic parsing, question comprehension, entity linking, knowledge base reasoning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-23.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-23.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
