SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2310.04668 · ICLR · 2024

Label-free Node Classification on Graphs with Large Language Models (LLMS)

Jiliang Tang, Wei Jin, Haitao Mao, Hongzhi Wen, Hui Liu, Haoyu Han, Haiyang Zhang, Zhikai Chen

arXiv · PDF · Open in the Atlas

Code that ran

22 of 29 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 16 failures of our instrument; 7 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2310.04668.json · badge: /api/ran/2310.04668.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
pmernyei/wiki-cs-dataset canonical (classifier label)
classifier basis: census
17 ran · 2 no recorded run
CurryTang/LLMGNN canonical (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
currytang/llmgnn canonical (classifier label)
classifier basis: census
1 ran · 5 no recorded run
FunctionStatusWhere it lives
calc_ratios Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/calc_sim_neighbours.py
code served (permissive licence) · get_code("6061fb3161a49f36")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import networkx as nx

# Create a small graph with nodes and edges
graph = nx.Graph()
graph.add_edges_from([(0, 1), (0, 2), (1, 2), (2, 3), (3, 4)])

# Create ys dictionary mapping node indices to class labels
ys = {0: 0, 1: 0, 2: 1, 3: 1, 4: 0}

ARGS = {
    "graph": graph,
    "ys": ys
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
calculate_avg_cosine_similarities Ran
ran, honoured the contract we drafted
pmernyei/wiki-cs-dataset/data_processing/pyscripts/analyze_datasets.py
code served (permissive licence) · get_code("ddc2ead4604bc732")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (3, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import random

# Define a simple Node class that matches what the function expects
class Node:
    def __init__(self, label, vector):
        self.label = label
        self.vector = vector

# Create small test data
random.seed(42)
np.random.seed(42)

# Create nodes dictionary with nodes from 3 different classes
nodes = {}
label_list = ["class_a", "class_b", "class_c"]

# Add 5 nodes per class with small 4-dimensional vectors
node_id = 0
for label in label_list:
    for i in range(5):
        vector = np.random.randn(4).astype(np.float32)
        nodes[node_id] = Node(label=label, vector=vector)
        node_id += 1

sample_count = 3

ARGS = {
    "nodes": nodes,
    "label_list": label_list,
    "sample_count": sample_count
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 3) — shape (3, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
calculate_connectivity_stats Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/analyze_datasets.py
code served (permissive licence) · get_code("0b02ef4443cf1e77")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

class Node:
    def __init__(self, label, outlinks):
        self.label = label
        self.outlinks = outlinks

# Create sample nodes with labels and outlinks
nodes = {
    0: Node(label='A', outlinks=[1, 2]),
    1: Node(label='A', outlinks=[0, 2]),
    2: Node(label='B', outlinks=[0, 1]),
    3: Node(label='B', outlinks=[1, 3]),
    4: Node(label='C', outlinks=[0, 3]),
}

label_list = ['A', 'B', 'C']

ARGS = {
    'nodes': nodes,
    'label_list': label_list,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
calculate_milestone_tree Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/inspect_milestone_tree.py
code served (permissive licence) · get_code("f87b1e7f91458a7b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os

# Create temporary files for the function to read
temp_dir = tempfile.mkdtemp()

# Create a subcats file with category parent relations
subcats_file = os.path.join(temp_dir, 'subcats.tsv')
with open(subcats_file, 'w', encoding='utf8') as f:
    f.write('Category:Science\tArticles\n')
    f.write('Category:Physics\tCategory:Science\n')
    f.write('Category:Chemistry\tCategory:Science\n')
    f.write('Category:Biology\tCategory:Science\n')
    f.write('Category:Quantum\tCategory:Physics\n')

# Create a sizes file with precomputed category sizes
sizes_file = os.path.join(temp_dir, 'sizes.tsv')
with open(sizes_file, 'w', encoding='utf8') as f:
    f.write('category\tpages\n')
    f.write('Category:Science\t100\n')
    f.write('Category:Physics\t50\n')
    f.write('Category:Chemistry\t30\n')
    f.write('Category:Biology\t20\n')
    f.write('Category:Quantum\t15\n')

# Create a page2cat file with page-to-category mappings
page2cat_file = os.path.join(temp_dir, 'page2cat.tsv')
with open(page2cat_file, 'w', encoding='utf8') as f:
    f.write('Page1\tCategory:Science\tCategory:Physics\n')
    f.write('Page2\tCategory:Chemistry\n')
    f.write('Page3\tCategory:Biology\tCategory:Science\n')

# Create output file path
out_file = os.path.join(temp_dir, 'output.txt')

ARGS = {
    'subcats_filename': subcats_file,
    'sizes_filename': sizes_file,
    'page2cat_filename': page2cat_file,
    'out_filename': out_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_all_descendants Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/inspect_milestone_tree.py
code served (permissive licence) · get_code("0778263c5d981ef7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned set. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

# Simple tree structure for testing get_all_descendants
# The function recursively visits nodes and collects all descendants

# Create a simple tree:
#       1
#      / \
#     2   3
#    /
#   4

root_node = 1
children_map = {
    1: [2, 3],
    2: [4],
    3: [],
    4: []
}

ARGS = {
    "root": root_node,
    "children_map": children_map
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_different_num Ran
ran, honoured the contract we drafted
CurryTang/LLMGNN/src/helper/data.py
pointer only (licence: NONE) · get_code("393a0d8988f753de")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, CurryTang/LLMGNN when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_values Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/mysqldump_to_csv.py
code served (permissive licence) · get_code("2b9ee849ce40d859")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

def get_values(line):
    """
    Returns the portion of an INSERT statement containing values
    """
    return line.partition('` VALUES ')[2]

# The function takes a single parameter: line
# It expects a string containing SQL INSERT statement syntax
# The function partitions on '` VALUES ' and returns the third element (after the separator)

line = "INSERT INTO table_name (column1, column2) ` VALUES (1, 2, 3)"

ARGS = {
    "line": line
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
inject_random_noise Ran
ran; our contract draft was wrong, not the code
CurryTang/LLMGNN/src/helper/data.py
pointer only (licence: NONE) · get_code("7841874972fcbc73")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, CurryTang/LLMGNN when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

is_insert Ran
ran, violated the contract we drafted
pmernyei/wiki-cs-dataset/data_processing/pyscripts/mysqldump_to_csv.py
code served (permissive licence) · get_code("0354e7c71fe7c820")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

# Construct valid arguments for is_insert function

# The function takes a single parameter 'line' which should be a string
# It checks if the string starts with 'INSERT INTO'

ARGS = {
    "line": "INSERT INTO users (id, name) VALUES (1, 'John')"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
keep_first_n_words Ran
ran; our contract draft was wrong, not the code
CurryTang/LLMGNN/src/llm.py
pointer only (licence: NONE) · get_code("e2a07418f83e3b77")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, CurryTang/LLMGNN when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

label_set Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/process_dataset.py
code served (permissive licence) · get_code("a559c37cf21d8253")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned set. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

# Define a simple node class with a label attribute
class Node:
    def __init__(self, label):
        self.label = label

# Create a dict of nodes with distinct labels
nodes = {
    "node_0": Node(label="label_A"),
    "node_1": Node(label="label_B"),
    "node_2": Node(label="label_A"),
    "node_3": Node(label="label_C"),
}

ARGS = {
    "nodes": nodes
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_data Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/experiments/linkprediction/load_wiki.py
code served (permissive licence) · get_code("465405a2bbb6bd57")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Data. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os.path
import numpy as np
import json
import itertools
import torch
from torch_geometric.data.data import Data
import tempfile

torch.manual_seed(0)

# Create a temporary JSON file with mock data
mock_data = {
    'features': [[0.1, 0.2, 0.3, 0.4], [0.5, 0.6, 0.7, 0.8]],
    'labels': [0, 1],
    'train_masks': [[True, False], [False, True]],
    'val_masks': [[False, True], [True, False]],
    'stopping_masks': [[True, True], [False, False]],
    'test_mask': [False, True],
    'links': [[1], [0]]
}

with tempfile.NamedTemporaryFile(mode='w', suffix='.json', delete=False) as f:
    json.dump(mock_data, f)
    temp_filename = f.name

ARGS = {
    'filename': temp_filename,
    'directed': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, torch_geometric — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Data — a Data output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_glove_dict Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/process_dataset.py
code served (permissive licence) · get_code("7baee91fefde7b99")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import tempfile
import os

# Create a temporary GloVe-format file for testing
temp_dir = tempfile.mkdtemp()
glove_file = os.path.join(temp_dir, "glove_test.txt")

with open(glove_file, 'w', encoding='utf8') as f:
    f.write("the 0.1 0.2 0.3 0.4\n")
    f.write("hello 0.5 0.6 0.7 0.8\n")
    f.write("world 0.9 1.0 1.1 1.2\n")
    f.write("python 1.3 1.4 1.5 1.6\n")

# Define ARGS
ARGS = {
    "filename": glove_file,
    "relevant_words": {"the", "hello", "world"}
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_redirects Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/extract_full_data_for_dataset.py
code served (permissive licence) · get_code("04fc08ae6a6e7056")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import tempfile
import os

# Create temporary CSV files for testing
page_table_fd, page_table_filename = tempfile.mkstemp(suffix='.csv', text=True)
redirect_table_fd, redirect_table_filename = tempfile.mkstemp(suffix='.csv', text=True)

try:
    # Write page table CSV
    # Format: id, namespace, title, is_redirect, ...
    with os.fdopen(page_table_fd, 'w', encoding='utf8') as f:
        writer = csv.writer(f)
        writer.writerow(['1', '0', 'Main_Page', '0', '', '0'])
        writer.writerow(['2', '0', 'About', '0', '', '0'])
        writer.writerow(['3', '0', 'Help', '0', '', '0'])
        writer.writerow(['100', '1', 'User:Test', '0', '', '0'])

    # Write redirect table CSV
    # Format: from_id, to_namespace, to_title, ..., ...
    with os.fdopen(redirect_table_fd, 'w', encoding='utf8') as f:
        writer = csv.writer(f)
        writer.writerow(['10', '0', 'Main_Page', 'ignored', 'ignored'])
        writer.writerow(['11', '0', 'About', 'ignored', 'ignored'])
        writer.writerow(['12', '1', 'User:Test', 'ignored', 'ignored'])
        writer.writerow(['13', '0', 'Help', 'ignored', 'ignored'])

    ARGS = {
        'page_table_filename': page_table_filename,
        'redirect_table_filename': redirect_table_filename,
    }

finally:
    pass
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_wiki Ran
ran, honoured the contract we drafted
pmernyei/wiki-cs-dataset/data_processing/pyscripts/calc_shortest_paths.py
code served (permissive licence) · get_code("d3648e98b2f4d8ef")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned DiGraph. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import itertools
import networkx as nx
import os
import tempfile

# Create a temporary directory and a mock data.json file
temp_dir = tempfile.mkdtemp()
mock_data = {
    'links': [
        [1, 2],
        [0, 2],
        [0, 1]
    ],
    'features': [
        [0.1, 0.2],
        [0.3, 0.4],
        [0.5, 0.6]
    ]
}

mock_path = os.path.join(temp_dir, 'data.json')
with open(mock_path, 'w') as f:
    json.dump(mock_data, f)

ARGS = {
    'path': mock_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DiGraph — a DiGraph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DiGraph output the recorded ndim 1 and dtype family int describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_wiki Ran
ran, honoured the contract we drafted
pmernyei/wiki-cs-dataset/data_processing/pyscripts/calc_sim_neighbours.py
code served (permissive licence) · get_code("2003a5c4bbb37699")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 2, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

# Create a temporary JSON file with mock data matching the expected structure
temp_dir = tempfile.mkdtemp()
data_path = os.path.join(temp_dir, 'data.json')

mock_data = {
    'links': [
        [1, 2],      # node 0 links to nodes 1 and 2
        [0, 2],      # node 1 links to nodes 0 and 2
        [0],         # node 2 links to node 0
    ],
    'features': [
        [0.1, 0.2],
        [0.3, 0.4],
        [0.5, 0.6],
    ],
    'labels': [0, 1, 0]
}

with open(data_path, 'w') as f:
    json.dump(mock_data, f)

ARGS = {
    'path': data_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, dtype family int — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
most_common_number Ran
ran; our contract draft was wrong, not the code
CurryTang/LLMGNN/src/llm.py
pointer only (licence: NONE) · get_code("78364baf10dfdc68")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, CurryTang/LLMGNN when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

page_titles_to_ids Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/extract_full_data_for_dataset.py
code served (permissive licence) · get_code("17e9b89c587f323d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import tempfile
import os

# Create a temporary CSV file with Wikipedia page data
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.csv', encoding='utf8', newline='')
temp_filename = temp_file.name

# Write sample page table data
writer = csv.writer(temp_file)
writer.writerow(['1', '0', 'Python_(programming_language)'])
writer.writerow(['2', '1', 'Talk:Python'])
writer.writerow(['3', '0', 'Java_(programming_language)'])
writer.writerow(['4', '0', 'C++'])
writer.writerow(['5', '14', 'Category:Programming_languages'])
temp_file.close()

# Define the titles we want to look up
titles_set = {'Python_(programming_language)', 'Java_(programming_language)', 'NonexistentPage'}

ARGS = {
    'titles_set': titles_set,
    'page_table_filename': temp_filename
}

# Cleanup will happen after function execution
import atexit
atexit.register(lambda: os.unlink(temp_filename))
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
page_titles_to_labels Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/extract_full_data_for_dataset.py
code served (permissive licence) · get_code("46b893445845c5f7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import csv
import tempfile
import os

# Create a temporary CSV file with test data
temp_dir = tempfile.mkdtemp()
page2cat_filename = os.path.join(temp_dir, "page2cat.tsv")

# Write test data to the TSV file
with open(page2cat_filename, 'w', encoding='utf8', newline='') as f:
    writer = csv.writer(f, delimiter='\t')
    writer.writerow(['Page Title One', 'Category A', 'Category B'])
    writer.writerow(['Page Title Two', 'Category A'])
    writer.writerow(['Page Title Three', 'Category C'])
    writer.writerow(['Another Page', 'Category B', 'Category C'])

# Create category_label_mapping
category_label_mapping = {
    'Label1': ['Category A', 'Category B'],
    'Label2': ['Category C'],
}

ARGS = {
    'category_label_mapping': category_label_mapping,
    'page2cat_filename': page2cat_filename,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
raw_data_dict Ran
ran; our contract draft was wrong, not the code
pmernyei/wiki-cs-dataset/data_processing/pyscripts/process_dataset.py
code served (permissive licence) · get_code("0199f42427faddb5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

class Node:
    def __init__(self, id, title, label, outlinks, tokens):
        self.id = id
        self.title = title
        self.label = label
        self.outlinks = outlinks
        self.tokens = tokens

node = Node(
    id=1,
    title="Example Node",
    label="test_label",
    outlinks=[2, 3, 4],
    tokens=["hello", "world", "example"]
)

ARGS = {
    'node': node
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
replace_args_with_dict_values Ran
ran; our contract draft was wrong, not the code
currytang/llmgnn/src/helper/args.py
pointer only (licence: NONE) · get_code("ca3bd8e13278c0f7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, currytang/llmgnn when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

values_sanity_check Ran
ran, violated the contract we drafted
pmernyei/wiki-cs-dataset/data_processing/pyscripts/mysqldump_to_csv.py
code served (permissive licence) · get_code("50aabaf962cff885")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

def values_sanity_check(values):
    """
    Ensures that values from the INSERT statement meet basic checks.
    """
    assert values
    assert values[0] == '('
    return True

ARGS = {
    'values': '(column1, column2, column3)'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
bfs_component Not yet run
unverified (harvested, no recorded run)
pmernyei/wiki-cs-dataset/data_processing/pyscripts/calc_shortest_paths.py
code served (permissive licence) · get_code("640e9a0511864dc6")
How it ran

no invocation recorded.

budget_density_query Not yet run
unverified (harvested, no recorded run)
currytang/llmgnn/src/helper/active.py
pointer only (licence: NONE) · get_code("71366d46fc7695ec")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, currytang/llmgnn when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

budget_density_query2 Not yet run
unverified (harvested, no recorded run)
currytang/llmgnn/src/helper/active.py
pointer only (licence: NONE) · get_code("492a608a449a11f4")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, currytang/llmgnn when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_T_HOC Not yet run
unverified (harvested, no recorded run)
currytang/llmgnn/src/helper/hoc.py
pointer only (licence: NONE) · get_code("a8799dcc333a0cdf")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, currytang/llmgnn when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_T_P_global Not yet run
unverified (harvested, no recorded run)
currytang/llmgnn/src/helper/hoc.py
pointer only (licence: NONE) · get_code("d1b530f92e04557e")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, currytang/llmgnn when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_T_P_local Not yet run
unverified (harvested, no recorded run)
currytang/llmgnn/src/helper/hoc.py
pointer only (licence: NONE) · get_code("26a4b05c4ea2e051")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, currytang/llmgnn when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_category_sizes Not yet run
unverified (harvested, no recorded run)
pmernyei/wiki-cs-dataset/data_processing/pyscripts/inspect_milestone_tree.py
code served (permissive licence) · get_code("7d905d3619fb5a98")
How it ran

no invocation recorded.

Also present in other papers' code

Of 29 samples lifted from this paper's linked repositories, 29 carry a function identity and 0 do not. Of the 29 distinct functions with an identity, 4 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 0 later) and 25 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
get_T_P_local 2 other papers
2 dated earlier
2102.05291 (2021-02)
Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels
2110.06283 (2021-10)
Detecting Corrupted Labels Without Training a Model to Predict
get_T_HOC 1 other paper
1 dated earlier
2102.05291 (2021-02)
Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels
get_T_P_global 1 other paper
1 dated earlier
2102.05291 (2021-02)
Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels
replace_args_with_dict_values 1 other paper
1 dated earlier
2309.16595 (2023-09)
Can LLMs Effectively Leverage Graph Structural Information through Prompts, and Why?

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

In recent years, there have been remarkable advancements in node classification achieved by Graph Neural Networks (GNNs). However, they necessitate abundant high-quality labels to ensure promising performance. In contrast, Large Language Models (LLMs) exhibit impressive zero-shot proficiency on text-attributed graphs. Yet, they face challenges in efficiently processing structural data and suffer from high inference costs. In light of these observations, this work introduces a label-free node classification on graphs with LLMs pipeline, LLM-GNN. It amalgamates the strengths of both GNNs and LLMs while mitigating their limitations. Specifically, LLMs are leveraged to annotate a small portion of nodes and then GNNs are trained on LLMs' annotations to make predictions for the remaining large portion of nodes. The implementation of LLM-GNN faces a unique challenge: how can we actively select nodes for LLMs to annotate and consequently enhance the GNN training? How can we leverage LLMs to obtain annotations of high quality, representativeness, and diversity, thereby enhancing GNN performance with less cost? To tackle this challenge, we develop an annotation quality heuristic and leverage the confidence scores derived from LLMs to advanced node selection. Comprehensive experimental results validate the effectiveness of LLM-GNN on text-attributed graphs from various domains. In particular, LLM-GNN can achieve an accuracy of 74.9% on a vast-scale dataset OGBN-PRODUCTS with a cost less than 1 dollar. Our code is available from https://github.com/CurryTang/LLMGNN.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2310.04668")
get_code_for_paper("2310.04668")
have("2310.04668")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2310.04668.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2310.04668.svg)](https://syntology.ai/paper/2310.04668)

Connect an agent — have() is free.