Solving and Explaining Analogy Questions Using Semantic Networks

Authors: Adrian Boteanu, Sonia Chernova

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on two datasets totaling 600 analogy questions. Our results show reliable performance and low false-positive rate in question answering; human evaluators agreed with 96% of our analogy explanations.
Researcher Affiliation Academia Adrian Boteanu, Sonia Chernova Worcester Polytechnic Institute 100 Institute Road Worcester, MA 01609 aboteanu@wpi.edu, soniac@wpi.edu
Pseudocode Yes Algorithm 1 Sequence similarity. s1 and s2 have the same length, each connecting a pair of concepts in different contexts. ... Algorithm 2 Generating a human-readable explanation from the best similarity sequence pair, which have the same length.
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes 373 questions used in SAT US college admittance tests. This dataset was also used in previous work on answering analogies (Turney and Littman 2005); ... 227 questions from a public domain website1 targeted for grades 1-12. ... Footnote 1: Section Unit 2: Read Theory Word Pair Analogies from http://www.englishforeveryone.org/Topics/Analogies.htm
Dataset Splits No The paper does not provide specific dataset split information (e.g., percentages, counts) for training, validation, and test sets. It evaluates on pre-existing datasets of analogy questions.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions software like Concept Net, Word Net, Verb Net, DBPedia, and Divisi toolkit, but does not provide specific version numbers for these or other ancillary software components.
Experiment Setup Yes In our analysis, we limited the semantic context subgraphs to have a maximum geodesic distance of two. ... We conducted our survey through the Crowd Flower crowdsourcing market using the 74 explanations (60 direct, 14 one-hop) produced by SSE from the correct answers selected when ignoring Related To and Conceptually Related To edges.