%0 Conference Paper %1 10.1007/978-3-031-43458-7_43 %A Purohit, Disha %B The Semantic Web: ESWC 2023 Satellite Events %C Cham %D 2023 %E Pesquita, Catia %E Skaf-Molli, Hala %E Efthymiou, Vasilis %E Kirrane, Sabrina %E Ngonga, Axel %E Collarana, Diego %E Cerqueira, Renato %E Alam, Mehwish %E Trojahn, Cassia %E Hertling, Sven %I Springer Nature Switzerland %K myown %P 244--254 %R https://doi.org/10.1007/978-3-031-43458-7_43 %T Semantic and Efficient Symbolic Learning over Knowledge Graphs %X In recent years, the rise of large Knowledge Graphs (KGs), which capture knowledge in machine-driven formats, has arisen broadly. KGs are the convergence of data and knowledge, and may be incomplete due to the Open World Assumption (OWA). Inductive Logic Programming (ILP) is a popular traditional approach for mining logical rules to complete the KGs. ILP approaches derive logical rules from ground facts in knowledge bases. Deducing new information or adding missing information to the KGs, identifying potential errors, and understanding the data more substantially can be accomplished by mining logical rules. Inference can be used to deduce new facts and complete KGs. To discover meaningful insights, traditional rule mining approaches first ignore axiomatic systems defining the semantics of the predicates and classes available in KGs. Second, most rule miners measure the impact of mined rules in terms of correlation rather than causation, and they are overwhelmed by the volume of data. Finally, existing frameworks implement blocking methods that require the processing of complete KGs to generate the mined rules. In this Ph.D. proposal, an outline of a rule-mining model explicitly tailored to mine Horn rules encapsulating semantics on top of KGs is reported. Additionally, the rule-mining approach is based on reliably estimating the cause-effect relationships and discovering new facts in the KGs considering data and metadata. Our approach follows an iterative process to inductively mine rules incorporating semantics to enhance completeness. Our experimental results suggest that by combining entailment regimes and querying KGs on demand, our approach outperforms the state-of-the-art in terms of accuracy. A publicly available Jupyter notebook that executes a demonstration is available (https://mybinder.org/v2/gh/SDM-TIB/DIGGER-ESWC2023Demo/HEAD?labpath=Mining\%20Symbolic\%20Rules\%20To\%20Explain\%20Lung\%20Cancer\%20Treatments.ipynb). %@ 978-3-031-43458-7

@inproceedings{10.1007/978-3-031-43458-7_43, abstract = {In recent years, the rise of large Knowledge Graphs (KGs), which capture knowledge in machine-driven formats, has arisen broadly. KGs are the convergence of data and knowledge, and may be incomplete due to the Open World Assumption (OWA). Inductive Logic Programming (ILP) is a popular traditional approach for mining logical rules to complete the KGs. ILP approaches derive logical rules from ground facts in knowledge bases. Deducing new information or adding missing information to the KGs, identifying potential errors, and understanding the data more substantially can be accomplished by mining logical rules. Inference can be used to deduce new facts and complete KGs. To discover meaningful insights, traditional rule mining approaches first ignore axiomatic systems defining the semantics of the predicates and classes available in KGs. Second, most rule miners measure the impact of mined rules in terms of correlation rather than causation, and they are overwhelmed by the volume of data. Finally, existing frameworks implement blocking methods that require the processing of complete KGs to generate the mined rules. In this Ph.D. proposal, an outline of a rule-mining model explicitly tailored to mine Horn rules encapsulating semantics on top of KGs is reported. Additionally, the rule-mining approach is based on reliably estimating the cause-effect relationships and discovering new facts in the KGs considering data and metadata. Our approach follows an iterative process to inductively mine rules incorporating semantics to enhance completeness. Our experimental results suggest that by combining entailment regimes and querying KGs on demand, our approach outperforms the state-of-the-art in terms of accuracy. A publicly available Jupyter notebook that executes a demonstration is available (https://mybinder.org/v2/gh/SDM-TIB/DIGGER-ESWC2023Demo/HEAD?labpath=Mining{\%}20Symbolic{\%}20Rules{\%}20To{\%}20Explain{\%}20Lung{\%}20Cancer{\%}20Treatments.ipynb).}, added-at = {2024-02-15T15:08:47.000+0100}, address = {Cham}, author = {Purohit, Disha}, biburl = {https://www.bibsonomy.org/bibtex/2753f647f91a507f1464487cf20d83a83/gabydler}, booktitle = {The Semantic Web: ESWC 2023 Satellite Events}, doi = {https://doi.org/10.1007/978-3-031-43458-7_43}, editor = {Pesquita, Catia and Skaf-Molli, Hala and Efthymiou, Vasilis and Kirrane, Sabrina and Ngonga, Axel and Collarana, Diego and Cerqueira, Renato and Alam, Mehwish and Trojahn, Cassia and Hertling, Sven}, interhash = {adcf9977d616ed1bb01169474d7085fa}, intrahash = {753f647f91a507f1464487cf20d83a83}, isbn = {978-3-031-43458-7}, keywords = {myown}, pages = {244--254}, publisher = {Springer Nature Switzerland}, timestamp = {2024-02-15T15:08:47.000+0100}, title = {Semantic and Efficient Symbolic Learning over Knowledge Graphs}, year = 2023 }