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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Discovering Opinion Intervals from Conflicts in Signed Graphs
Authors: Peter Blohm, Florian Chen, Aristides Gionis, Stefan Neumann
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Next, we experimentally evaluate our algorithms. Our code is available in a Git Hub repository.2 We aim to answer the following research questions: (RQ1) Does BEST INTERVAL APPROXIMATION yield a substantial increase in expressiveness compared to CORRELATION CLUSTERING? (RQ2) How computationally efficient and scalable are our proposed algorithms? (RQ3) What is the trade-off between solution quality and the number of intervals? (RQ4) Are the solutions produced by our method interpretable? (RQ5) Are our algorithms able to recover ground-truth interval structures? |
| Researcher Affiliation | Academia | Peter Blohm Aalto University Espoo, Finland EMAIL Florian Chen University of Oxford Oxford, UK EMAIL Aristides Gionis KTH Royal Institute of Technology Digital Futures Stockholm, Sweden EMAIL Stefan Neumann TU Wien Vienna, Austria EMAIL |
| Pseudocode | Yes | Algorithm 1: Greedy Agreement Interval Assignment (GAIA) Input: Signed graph G = (V, E+ E ), intervals I1, . . . , Ik Output: Interval assignment (C1, . . . , Ck) where Cℓare the vertices assigned to interval Iℓ |
| Open Source Code | Yes | Next, we experimentally evaluate our algorithms. Our code is available in a Git Hub repository.2 |
| Open Datasets | Yes | We evaluate our algorithms on real-world datasets from SNAP [46] and KONECT [39]. We further provide a novel dataset based on voting data from the German Bundestag (parliament) between the years of 2012 and 2025 and make it available in our repository. |
| Dataset Splits | No | We fix the 8-Chain and generate a graph with n = 800 vertices as follows. We assign n/8 vertices to each interval, and we introduce edges with signs corresponding to the interval structure for d n/2 random pairs of vertices, where d ∈ [0, 1] is the desired density of the graph. Each edge obtains a correct edge sign based on the interval structure with probability 1 − p and we flip the sign with probability p. In our experiments, we measure the relative change of the objective function achieved by our algorithms compared to the ground-truth assignment in percent, agree(G,ground truth)−agree(G,ALG) / |E| · 100, and we also report the accuracy with which vertices are assigned to their corresponding interval. |
| Hardware Specification | Yes | All our algorithms are implemented in the Rust programming language. The experiments were run on a system with two AMD EPYC 9124 CPUs, 500 GB of RAM, and an NVIDIA RTX 4000 Ada Generation GPU. |
| Software Dependencies | No | All our algorithms are implemented in the Rust programming language. |
| Experiment Setup | Yes | For VENUS, we use an initial temperature of 100 and a decay factor of 2/3, applied every 5 epochs. Both are run with 10 batches for vertex reassignment. For the interval structure that our algorithms receive as input, unless stated otherwise, we use a chain-like structure of 8 intervals, where each interval overlaps with the next, e.g., [0, 1], [1, 2], . . . , [7, 8], and we call this interval structure an 8-Chain. |