Reveal hidden links through shared neighbor evidence. Tune directed or weighted graphs with robust options. Export results quickly, then validate decisions using clear reports.
The resource allocation index estimates similarity between two nodes using their shared neighbors.
Higher scores indicate stronger similarity through informative common neighbors.
Example edge list (unweighted, undirected) and a sample pair score.
| Edge | Node A | Node B |
|---|---|---|
| 1 | A | B |
| 2 | A | C |
| 3 | B | C |
| 4 | B | D |
| 5 | C | D |
| 6 | C | E |
| 7 | D | E |
| Pair | Common neighbors | RAI |
|---|---|---|
| A–D | B, C | 1/3 + 1/4 = 0.583333 |
| B–E | C, D | 1/4 + 1/3 = 0.583333 |
| A–E | C | 1/4 = 0.25 |
Many physics workflows can be modeled as networks: transport pathways, interaction graphs, and correlation structures. The resource allocation index (RAI) rates how plausible a connection is by counting shared neighbors while downweighting hubs. It is widely used in complex-network link prediction. In physics datasets, it can highlight likely interactions missing from sparse measurements.
If two sites share intermediates, resources can be routed through them in diffusion or flow analogies. A neighbor with few connections provides focused support because its influence is less split. RAI sums the inverse degree or strength of each common neighbor.
The calculator applies RAI(u, v) = Σz∈Γ(u)∩Γ(v) 1/k(z). Γ(u) is u’s neighbor set and k(z) is the degree of z. With weighted mode, k(z) becomes node strength computed as the sum of incident weights. In pairwise mode, the contribution table lists each 1/k(z) term so you can see which common neighbor drives the score.
For directed data, the tool uses out‑neighbors so direction stays consistent. Weighted mode fits edges that represent intensity, frequency, conductance, or coupling magnitude. Use positive weights on a consistent scale across the dataset.
RAI is a relative similarity score, so compare pairs within the same network. Larger values indicate more low‑degree common neighbors and stronger candidate links. Batch mode ranks pairs so you can prioritize follow‑up work.
Use one edge per line and keep labels consistent. Remove self‑loops; duplicates are merged by summing weights in weighted mode. For large graphs, restrict to a component or time window. Match the directed option to causality or flow. Lines starting with # are treated as comments and ignored.
RAI supports link prediction in interaction networks, helps suggest missing couplings, and ranks candidate correlations in complex systems. Use it alongside other measures, then filter results using physics constraints, units, and measurement limits. Examples include contact networks, lattice defect interactions, power‑grid couplings, and photonic or molecular interaction graphs.
CSV export suits spreadsheets, scripts, and reproducible notebooks. The PDF report is helpful for sharing results. Pairwise mode also lists each neighbor’s contribution, clarifying why a score is high.
A higher score means the pair shares more informative common neighbors. Neighbors with low degree or low strength contribute more, so the score favors specific intermediates rather than broad hubs.
No. It is a similarity score for ranking candidate links within the same dataset. You can calibrate scores to probabilities using historical links or a validation set.
Enable it when edge weights represent meaningful intensity, such as coupling strength, interaction frequency, or conductance. The calculator then uses node strength instead of degree in the inverse term.
Directed mode uses out‑neighbors to define each node’s neighbor set. This is useful for causal, flow, or influence networks where direction matters for similarity interpretation.
Hubs connect to many nodes, so their support is diffuse. The inverse degree term reduces their influence, prioritizing rare shared neighbors that often carry more specific information.
The score becomes zero and the contribution table is empty. In batch mode, those pairs can be filtered out by setting a minimum common-neighbor threshold above zero.
Be careful. Network size, density, and weighting scales change score ranges. For cross‑dataset comparisons, normalize inputs or evaluate rankings using the same validation methodology.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.