SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.06378 · NAACL · 2021

QA-GNN: Reasoning with Language Models and Knowledge Graphs for Question Answering

Hongyu Ren, Jure Leskovec, Percy Liang, Michihiro Yasunaga, Antoine Bosselut

arXiv · PDF · Open in the Atlas

Code that ran

8 of 25 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; 1 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 4 failures of our instrument; 17 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/2104.06378.json · badge: /api/ran/2104.06378.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
michiyasunaga/qagnn canonical (classifier label)
classifier basis: census
8 ran · 13 no recorded run
HaochenLiu2000/QAP community
Papers with Code archive
0 ran · 2 no recorded run
tobias-opsahl/fact-or-fiction community
Papers with Code archive
0 ran · 2 no recorded run
FunctionStatusWhere it lives
create_hypothesis Ran
ran on a synthesized input
michiyasunaga/qagnn/utils/convert_csqa.py
code served (permissive licence) · get_code("e4b000b919d28344")
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 re

BLANK_STR = '___'

def create_hypothesis(fitb: str, choice: str, ans_pos: bool) -> str:
    if '. ' + BLANK_STR in fitb or fitb.startswith(BLANK_STR):
        choice = choice[0].upper() + choice[1:]
    else:
        choice = choice.lower()
    if not fitb.endswith(BLANK_STR):
        choice = choice.rstrip('.')
    if not ans_pos:
        try:
            hypothesis = re.sub('__+', choice, fitb)
        except:
            print(choice, fitb)
        return hypothesis
    choice = choice.strip()
    m = re.search('__+', fitb)
    start = m.start()
    length = len(choice) - 1 if fitb.endswith(BLANK_STR) and choice[-1] in ['.', '?', '!'] else len(choice)
    hypothesis = re.sub('__+', choice, fitb)
    return (hypothesis, (start, start + length))


ARGS = {
    'fitb': 'The quick ___ fox jumps over the lazy dog',
    'choice': 'brown',
    'ans_pos': False
}
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 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
create_output_dict Ran
ran on a synthesized input
michiyasunaga/qagnn/utils/convert_obqa.py
code served (permissive licence) · get_code("345456d568545c8e")
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

torch.manual_seed(0)

ARGS = {
    'input_json': {'existing_key': 'existing_value'},
    'statement': 'This is a test statement',
    'label': True
}
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
del_pos Ran
ran on a synthesized input
michiyasunaga/qagnn/utils/conceptnet.py
code served (permissive licence) · get_code("5f4eefd99c11e698")
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

# Create valid arguments for the del_pos function
ARGS = {
    "s": "example/n"
}
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
dropout_mask Ran
ran; our contract draft was wrong, not the code
michiyasunaga/qagnn/utils/layers.py
code served (permissive licence) · get_code("866c784c24ba2fa8")
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 Tensor, ndim 2, dtype family float, shape (2, 4) 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
import torch.nn as nn

torch.manual_seed(0)

# Create a small tensor to use as x
x = torch.randn(2, 4)

# Define arguments
ARGS = {
    'x': x,
    'sz': (2, 4),
    'p': 0.5
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
gelu Ran
ran, honoured the contract we drafted
michiyasunaga/qagnn/utils/layers.py
pointer only (licence: MIT); the stored copy get_code serves, from qshi95/lergv, is pointer only (licence: NONE) · get_code("b75e9c793d2a286a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from qshi95/lergv, and GitHub detected no licence in that repository 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_fitb_from_question Ran
ran; our contract draft was wrong, not the code
michiyasunaga/qagnn/utils/convert_csqa.py
code served (permissive licence) · get_code("a9a454bf4414e3fe")
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.

# Setup code for replace_wh_word_with_blank and get_fitb_from_question

# The function takes a question_text string and returns a FITB (Fill In The Blank) version
# Let's create test cases for different scenarios

ARGS = {
    "question_text": "What is the capital of France?"
}
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 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
load_cpnet_vocab Ran
ran; our contract draft was wrong, not the code
michiyasunaga/qagnn/utils/grounding.py
code served (permissive licence) · get_code("2f104114280df441")
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 tempfile
import os

# Create a temporary file with sample vocabulary
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt', encoding='utf8')
temp_file.write("concept_one\n")
temp_file.write("concept_two_word\n")
temp_file.write("another_concept\n")
temp_file.write("simple\n")
temp_file.close()

ARGS = {
    "cpnet_vocab_path": temp_file.name
}
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 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
normalize_sparse_adj Ran
ran; our contract draft was wrong, not the code
michiyasunaga/qagnn/utils/maths.py
code served (permissive licence) · get_code("5edf23a7957ba803")
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 coo_matrix. 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
from scipy import sparse

# Create a small sparse adjacency matrix
# Using a 4x3 matrix to have distinct dimensions
data = np.array([1, 2, 3, 4, 5])
row = np.array([0, 1, 2, 3, 0])
col = np.array([0, 1, 2, 2, 1])
A_sparse = sparse.coo_matrix((data, (row, col)), shape=(4, 3))

ARGS = {
    'A': A_sparse,
    'sparse_type': 'coo'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type coo_matrix — a coo_matrix output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coo_matrix 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
CustomGraphTransformerLayer Not yet run
unverified (harvested, no recorded run)
HaochenLiu2000/QAP/medqa/train_qap.py
pointer only (licence: NONE) · get_code("2d38a4d448347c3c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, HaochenLiu2000/QAP 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.

GraphTransformer Not yet run
unverified (harvested, no recorded run)
HaochenLiu2000/QAP/medqa/train_qap.py
pointer only (licence: NONE) · get_code("941efc4033b96f8c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, HaochenLiu2000/QAP 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.

QAGNN Not yet run
unverified (harvested, no recorded run)
tobias-opsahl/fact-or-fiction/models.py
pointer only (licence: NONE) · get_code("f82bdec01f779733")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, tobias-opsahl/fact-or-fiction 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.

concepts2adj Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/graph.py
code served (permissive licence) · get_code("e6a99b8687aab99c")
How it ran

no invocation recorded.

construct_graph Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/conceptnet.py
code served (permissive licence) · get_code("be8c35c9c1d52b66")
How it ran

no invocation recorded.

convert_qajson_to_entailment Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/convert_csqa.py
code served (permissive licence) · get_code("15baaf34a0e42ec6")
How it ran

no invocation recorded.

evaluate_accuracy Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/qagnn.py
code served (permissive licence) · get_code("086fa0038784db8c")
How it ran

no invocation recorded.

generate_adj_matrix_per_inst Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/graph.py
code served (permissive licence) · get_code("96ec8d520ae857f1")
How it ran

no invocation recorded.

get_bert_model Not yet run
unverified (harvested, no recorded run)
tobias-opsahl/fact-or-fiction/models.py
pointer only (licence: NONE) · get_code("216445b74badf787")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, tobias-opsahl/fact-or-fiction 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.

glove_init Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/conceptnet.py
code served (permissive licence) · get_code("9dcf00b849de8977")
How it ran

no invocation recorded.

lemmatize Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/grounding.py
code served (permissive licence) · get_code("5ff1624579da2bdc")
How it ran

no invocation recorded.

load_bert_xlnet_roberta_input_tensors Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/data_utils.py
code served (permissive licence) · get_code("b5768cf5e27d534d")
How it ran

no invocation recorded.

load_gpt_input_tensors Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/data_utils.py
code served (permissive licence) · get_code("e633666b83ca73b6")
How it ran

no invocation recorded.

load_sparse_adj_data_with_contextnode Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/data_utils.py
code served (permissive licence) · get_code("16bd49eb2e3b8970")
How it ran

no invocation recorded.

make_one_hot Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/modeling/modeling_qagnn.py
code served (permissive licence) · get_code("df5e0b02401bf941")
How it ran

no invocation recorded.

masked_softmax Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/layers.py
code served (permissive licence) · get_code("3841430a72ff67fa")
How it ran

no invocation recorded.

plain_graph_generation Not yet run
unverified (harvested, no recorded run)
michiyasunaga/qagnn/utils/graph.py
code served (permissive licence) · get_code("66135965fe362f47")
How it ran

no invocation recorded.

Also present in other papers' code

Of 25 samples lifted from this paper's linked repositories, 25 carry a function identity and 0 do not. Of the 25 distinct functions with an identity, 15 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 1 the same month, 15 later) and 10 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
gelu 51 other papers
23 dated earlier, 1 the same month, 27 later · 8 only as an assembly brick
1706.03762 (2017-06)
Attention Is All You Need
1810.04805 (2018-10)
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
1901.02860 (2019-01)
Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
and 48 more not named here
del_pos 5 other papers
1 dated earlier, 4 later
1909.02151 (2019-09)
KagNet: Knowledge-Aware Graph Networks for Commonsense Reasoning
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
and 2 more not named here
construct_graph 4 other papers
4 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
2023.findings-emnlp.540 (2023 (year only))
Open-ended Commonsense Reasoning with Unrestricted Answer Scope
and 1 more not named here
dropout_mask 4 other papers
3 later, 1 same year, order not determinable
2021.findings-emnlp.130 (2021 (year only))
Cross-lingual Transfer for Text Classification with Dictionary-based Heterogeneous Graph
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
and 1 more not named here
get_fitb_from_question 4 other papers
1 dated earlier, 3 later
1909.02151 (2019-09)
KagNet: Knowledge-Aware Graph Networks for Commonsense Reasoning
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
and 1 more not named here
glove_init 4 other papers
4 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
2023.findings-emnlp.540 (2023 (year only))
Open-ended Commonsense Reasoning with Unrestricted Answer Scope
and 1 more not named here
convert_qajson_to_entailment 3 other papers
3 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
aaai_26578 (2023 (year only))
Relation-Aware Language-Graph Transformer for Question Answering
create_output_dict 3 other papers
3 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
aaai_26578 (2023 (year only))
Relation-Aware Language-Graph Transformer for Question Answering
masked_softmax 3 other papers
3 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
aaai_26578 (2023 (year only))
Relation-Aware Language-Graph Transformer for Question Answering
create_hypothesis 2 other papers
2 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
aaai_26578 (2023 (year only))
Relation-Aware Language-Graph Transformer for Question Answering
make_one_hot 2 other papers
2 later
2022.findings-naacl.131 (2022 (year only))
Great Truths are Always Simple: A Rather Simple Knowledge Encoder for Enhancing the Commonsense Reasoning Capacity of Pre-Trained Models
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
evaluate_accuracy 1 other paper
1 later
2112.02732 (2021-12)
JointLK: Joint Reasoning with Language Models and Knowledge Graphs for Commonsense Question Answering
generate_adj_matrix_per_inst 1 other paper
1 later
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining
load_gpt_input_tensors 1 other paper
1 later
aaai_26578 (2023 (year only))
Relation-Aware Language-Graph Transformer for Question Answering
plain_graph_generation 1 other paper
1 later
2210.09338 (2022-10)
Deep Bidirectional Language-Knowledge Graph Pretraining

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

The problem of answering questions using knowledge from pre-trained language models (LMs) and knowledge graphs (KGs) presents two challenges: given a QA context (question and answer choice), methods need to (i) identify relevant knowledge from large KGs, and (ii) perform joint reasoning over the QA context and KG. Here we propose a new model, QA-GNN, which addresses the above challenges through two key innovations: (i) relevance scoring, where we use LMs to estimate the importance of KG nodes relative to the given QA context, and (ii) joint reasoning, where we connect the QA context and KG to form a joint graph, and mutually update their representations through graph-based message passing. We evaluate QA-GNN on the CommonsenseQA and OpenBookQA datasets, and show its improvement over existing LM and LM+KG models, as well as its capability to perform interpretable and structured reasoning, e.g., correctly handling negation in questions. If it is not used for hair, a round brush is an example of what? A. hair brush B. bathroom C. art supplies* D. shower E. hair salon painting AtLocation round brush hair brush R e l a t e d T o AtLocation Us ed Fo r Us ed Fo r Choice Entity Answer Question Entity QA context Knowledge graph art supply QA context Node hair Graph Connection ( §3.1) Dev Acc.

For agents

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

get_harvested_code_for_paper("2104.06378")
get_code_for_paper("2104.06378")
have("2104.06378")

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

curl https://syntology.ai/api/ran/2104.06378.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.06378.svg)](https://syntology.ai/paper/2104.06378)

Connect an agent — have() is free.