#!/usr/bin/env python3 """ Co-authorship network for the AI corpus. Two explicit thresholds keep the picture readable rather than a hairball, and both are stated in the figure caption: MIN_ITEMS an author must appear on at least this many corpus items MIN_LINKS ...and must have at least this many co-author links after that Authors are identified by surname + first initial. The journal database stores names as published with NO disambiguation, so two people who share a surname and an initial are merged here; treat author-level counts as approximate. pip install matplotlib networkx python3 make_network.py """ import json, math, os from collections import Counter, defaultdict import matplotlib matplotlib.use("Agg") import matplotlib.pyplot as plt import networkx as nx HERE = os.path.dirname(__file__) NET = os.path.join(HERE, "..", "data", "coauthorship.json") FIGS = os.path.join(HERE, "..", "figures") MIN_ITEMS = 2 # author must appear on >= this many corpus items MIN_LINKS = 3 # ...and have >= this many co-author links EMP, REC, COM = "#4a3aa7", "#eb6834", "#1baf7a" INK, INK2, MUT, EDGE, SURF = "#18181B", "#3F3F46", "#71717A", "#C9C7C0", "#FFFFFF" plt.rcParams.update({"svg.fonttype": "none", "font.family": "sans-serif", "font.sans-serif": ["IBM Plex Sans", "system-ui", "Arial"], "text.color": INK}) def main(): D = json.load(open(NET)) pc, counts, labels = D["pc"], Counter(D["counts"]), D["labels"] keep = {a for a, c in counts.items() if c >= MIN_ITEMS} w = Counter() for pid, ns in pc.items(): ks = sorted(a for a in ns if a in keep) for i, x in enumerate(ks): for y in ks[i + 1:]: w[(x, y)] += 1 # prune iteratively: dropping a node removes its links, which can push a # neighbour below the threshold. Repeat until every surviving node really # has MIN_LINKS links *within the displayed graph* (a k-core). nodes = set(keep) while True: edges = {k: v for k, v in w.items() if k[0] in nodes and k[1] in nodes} deg = Counter() for (x, y) in edges: deg[x] += 1; deg[y] += 1 drop = {a for a in nodes if deg[a] < MIN_LINKS} if not drop: break nodes -= drop # each author's dominant category, for node colour mix = defaultdict(Counter) for pid, ns in pc.items(): lab = labels.get(pid) for a in ns: if a in nodes and lab: mix[a][lab] += 1 colour = {a: {"empirical": EMP, "reception": REC, "commentary": COM}[mix[a].most_common(1)[0][0]] for a in nodes} G = nx.Graph(); G.add_nodes_from(nodes) for (x, y), n in edges.items(): G.add_edge(x, y, weight=n) order = sorted(nodes, key=lambda a: (-counts[a], a)) num = {a: i + 1 for i, a in enumerate(order)} pos = nx.spring_layout(G, seed=9, k=3.4 / math.sqrt(len(nodes)), iterations=900) P = {n: [pos[n][0], pos[n][1]] for n in G} MIN_SEP = 0.17 for _ in range(1200): # push overlapping nodes apart moved = False ns = list(P) for i, a in enumerate(ns): for b in ns[i + 1:]: dx, dy = P[a][0] - P[b][0], P[a][1] - P[b][1] d = math.hypot(dx, dy) if d < MIN_SEP: push = (MIN_SEP - d) / 2 / (d + 1e-9) P[a][0] += dx * push; P[a][1] += dy * push P[b][0] -= dx * push; P[b][1] -= dy * push moved = True if not moved: break fig = plt.figure(figsize=(13.4, 7.8)); fig.patch.set_facecolor(SURF) ax = fig.add_axes([0.0, 0.085, 0.615, 0.85]); ax.set_facecolor(SURF); ax.axis("off") for (a, b), n in edges.items(): ax.plot([P[a][0], P[b][0]], [P[a][1], P[b][1]], color=EDGE, lw=0.7 + 0.8 * n, zorder=1, solid_capstyle="round") for a in G: c = counts[a] ax.scatter([P[a][0]], [P[a][1]], s=190 + 62 * c, color=colour[a], edgecolors="#111", linewidths=1.1, zorder=3) ax.text(P[a][0], P[a][1], str(num[a]), ha="center", va="center", fontsize=8.6, color="#fff" if colour[a] != COM else "#06301e", fontweight="bold", zorder=4) ax.margins(0.09) lg = fig.add_axes([0.625, 0.085, 0.375, 0.85]); lg.axis("off") lg.set_xlim(0, 1); lg.set_ylim(0, 1) lg.text(0, 0.995, "Node colour — the author's most common category", fontsize=9.2, fontweight="bold", va="top", color=INK) for i, (lab, col) in enumerate((("empirical", EMP), ("reception", REC), ("commentary", COM))): lg.scatter([0.022], [0.952 - i * 0.038], s=95, color=col, edgecolors="#111", linewidths=1) lg.text(0.065, 0.952 - i * 0.038, lab, fontsize=8.8, color=INK2, va="center") top_y = 0.952 - 3 * 0.038 - 0.032 lg.text(0, top_y, "Authors, by number of corpus items", fontsize=9.2, fontweight="bold", va="top", color=INK) half = math.ceil(len(order) / 2) step = min(0.042, (top_y - 0.05) / max(half, 1)) for col, chunk in enumerate((order[:half], order[half:])): for row, a in enumerate(chunk): y = top_y - 0.042 - row * step lg.text(col * 0.5, y, f"{num[a]}.", fontsize=8.5, color=MUT, va="center", ha="left") lg.text(col * 0.5 + 0.055, y, f"{a} ({counts[a]})", fontsize=8.5, color=INK2, va="center", ha="left") fig.text(0.008, 0.988, f"Who works with whom — {len(nodes)} authors, {len(edges)} collaborations", fontsize=13, fontweight="bold", color=INK, va="top") fig.text(0.008, 0.012, f"Inclusion: authors on {MIN_ITEMS}+ corpus items with {MIN_LINKS}+ co-author links " f"({len(nodes)} of {len(counts)} authors in the corpus). Node size and the number in the legend " f"are that author's item count;\nedge width is joint items. Position carries no meaning beyond " f"connectivity. Names are surname + first initial: the journal database does not disambiguate " f"authors, so same-initial namesakes merge.", fontsize=7.9, color=MUT, va="bottom") fig.savefig(f"{FIGS}/network.svg", bbox_inches="tight", pad_inches=0.28, facecolor=SURF) fig.savefig("/tmp/_network_check.png", dpi=155, bbox_inches="tight", pad_inches=0.28, facecolor=SURF) print(f"wrote network.svg — {len(nodes)} nodes, {len(edges)} edges " f"(every node has >= {MIN_LINKS} links within the graph)") comps = [c for c in nx.connected_components(G) if len(c) > 1] print(f"components: {len(comps)}, largest {max((len(c) for c in comps), default=0)}") if __name__ == "__main__": main()