Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Computing and Explaining Query Answers over Inconsistent DL-Lite Knowledge Bases
Authors: Meghyn Bienvenu, Camille Bourgaux, François Goasdoué
JAIR 2019 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we empirically study the efficiency of our query answering and explanation framework using a benchmark we built upon the well-established LUBM benchmark. ... 7. Experimental Evaluation |
| Researcher Affiliation | Academia | Meghyn Bienvenu EMAIL CNRS, Universit e Montpellier 2 Campus St Priest, 161 rue Ada 34090 Montpellier, France; Camille Bourgaux EMAIL T el ecom Paris Tech 46 rue Barrault F-75634 Paris Cedex 13, France; Fran cois Goasdou e EMAIL Univ Rennes, CNRS, IRISA 6 rue de Kerampont, CS 80518, 22305 Lannion Cedex, France |
| Pseudocode | Yes | ALGORITHM 1: Compute Conflicts; ALGORITHM 2: Compute Causes; ALGORITHM 3: Classify Query |
| Open Source Code | Yes | The CQAPri system and benchmark can be downloaded from https://www.lri.fr/~bourgaux/CQAPri. |
| Open Datasets | Yes | Finally, we empirically study the efficiency of our query answering and explanation framework using a benchmark we built upon the well-established LUBM benchmark. ... The CQAPri system and benchmark can be downloaded from https://www.lri.fr/~bourgaux/CQAPri. |
| Dataset Splits | No | We generated ABoxes of increasing sizes with the Extended University Data Generator (EUGen) provided with LUBM 20 by setting its data completeness parameter (i.e. the percentage of individuals from a given concept for which roles describing this concept are indeed filled) to its default value of 95%, which seems realistic from the application viewpoint. All the generated ABoxes were found consistent w.r.t. our enriched TBox, which suggests that the added disjointness constraints were faithful to the reused benchmark. The size of these ABoxes ranges from 75,663 to 9,938,139 assertions, which corresponds to 1 to 100 universities in EUGen settings, and each ABox is included in the larger ones: the smallest corresponds to university 0, the largest to universities 0 to 99. Inconsistencies were introduced by reviewing all of the assertions of the consistent ABox, and contradicting the presence of an individual in a concept assertion with probability p, and the presence of each individual in a role assertion with probability p/2. The paper discusses data generation and how inconsistencies were introduced, but it does not specify training, testing, or validation splits for machine learning experiments. |
| Hardware Specification | Yes | All experiments reported in this work were run on an Intel Xeon X5647 at 2.93 GHz with 16 GB of RAM, running Cent OS 6.8. |
| Software Dependencies | Yes | CQAPri is built on top of the relational database server Postgre SQL v9.3.2 (www.postgresql.org), the Rapid v1.0 query rewriting engine for DL-Lite (Chortaras, Trivela, & Stamou, 2011), and the SAT4J v2.3.4 SAT solver (Berre & Parrain, 2010). |
| Experiment Setup | Yes | All these building blocks are used with their default settings. ... We generated ABoxes of increasing sizes with the Extended University Data Generator (EUGen) provided with LUBM 20 by setting its data completeness parameter ... to its default value of 95% ... For each of the 100 universities that constitute our consistent ABoxes, we set p = 0.002 and generated 50 batches of conflicting assertions using the method described above. |