版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
#利用大语言模型
促进综合图学习能力02
大语言模型进行图学习的现状概述01
为什么应用大语言模型进行图学习
03
大语言模型促进跨领域跨任务的统一图学习04
潜在研究方向目录#01为什么应用大语言模型进行图学习为什么应用大语言模型进行图学习大语言模型的能力图数据的特征为什么应用大语言模型进行图学习大语言模型的能力LLMs
have
demonstrated
their
strong
text
encoding/decoding
ability.为什么应用大语言模型进行图学习大语言模型的能力LLMs
have
shown
newly
found
emergent
ability
(e.g.,
reasoning).为什么应用大语言模型进行图学习图数据的特征In
real
world,
text
and
graph
usually
appears
simultaneously.Text
data
are
associated
with
rich
structure
information
in
the
form
of
graphs.Graph
data
are
captioned
with
rich
textual
information.#02大语言模型进行图学习的现状概述大语言模型进行图学习的现状概述不同的图数据应用场景图任务中大语言模型的不同角色不同的图数据应用场景大语言模型进行图学习的现状概述大语言模型进行图学习的现状概述不同的图数据应用场景:Pure
GraphDefinition:
Graph
with
no
text
information
or
no
semantically
rich
text
information.
eg.
traffic
graphsor
power
transmission
graph.Problems
on
Pure
Graphs:graph
reasoning
tasks
likeconnectivityshortestpathsubgraph
matchinglogical
rule
induction…大语言模型进行图学习的现状概述不同的图数据应用场景:Pure
GraphGraph
with
no
text
information
or
no
semantically
rich
text
information.
eg.
traffic
graphs
or
powertransmission
graph.大语言模型进行图学习的现状概述不同的图数据应用场景:Text-Paired
Graph大语言模型进行图学习的现状概述不同的图数据应用场景:Text-Paired
Graph大语言模型进行图学习的现状概述不同的图数据应用场景:Text-Attributed
Graph大语言模型进行图学习的现状概述不同的图数据应用场景:Text-Attributed
Graph大语言模型进行图学习的现状概述图任务中大语言模型的不同角色LLM
as
Enhancer/EncoderLLM
as
PredictorLLM
as
Aligner大语言模型进行图学习的现状概述Eembedding-based图任务中大语言模型的不同角色:LLM
as
Enhancer/EncoderExplanation-based大语言模型进行图学习的现状概述LLM
as
Enhancer/Encoder:
Explanation-basedICLR’24Basically,
using
T
andA,to
generate
P
and
E,then
use
T,
A,
P,
Easenriched
text
feature.大语言模型进行图学习的现状概述LLM
as
Enhancer/Encoder:
Explanation-based大语言模型进行图学习的现状概述LLM
as
Enhancer/Encoder:
Embedding-basedObservation:Fine-tune-based
LLMsmay
fail
at
low
labelingrate
settings.Chen,
Z.,
Mao,
H.,
Li,
H.,
Jin,
W.,
Wen,
H.,
Wei,
X.,
Wang,
S.,
Yin,
D.,
Fan,
W.,
Liu,
H.,
&
Tang,
J.
(2024).
Exploring
thePotential
of
Large
Language
Models
(LLMs)
in
Learning
on
Graphs
(arXiv:2307.03393).
arXiv.Low
label
ratioHigh
label
ratio大语言模型进行图学习的现状概述LLM
as
Enhancer/Encoder:
Embedding-basedObservation:Under
embedding-basedstructure,
thebination
ofdeep
sentence
embedding
withGNNs
makes
a
strong
baseline.Low
label
ratioChen,
Z.,
Mao,
H.,
Li,
H.,
Jin,
W.,
Wen,
H.,
Wei,
X.,
Wang,
S.,
Yin,
D.,
Fan,
W.,
Liu,
H.,
&
Tang,
J.
(2024).
Exploring
thePotential
of
Large
Language
Models
(LLMs)
in
Learning
on
Graphs
(arXiv:2307.03393).
arXiv.High
label
ratio大语言模型进行图学习的现状概述图任务中大语言模型的不同角色:LLM
as
PredictorFlatten-basedGNN-based大语言模型进行图学习的现状概述LLM
asPredictor:
Flatten-based大语言模型进行图学习的现状概述LLM
asPredictor:
Flatten-based大语言模型进行图学习的现状概述LLM
as
Predictor:
GNN-based大语言模型进行图学习的现状概述LLM
as
Predictor:
GNN-based大语言模型进行图学习的现状概述图任务中大语言模型的不同角色:LLM
as
Aligner大语言模型进行图学习的现状概述LLM
as
Aligner:Contrastive大语言模型进行图学习的现状概述LLM
as
Aligner:DistillationMavromatis,
Costas,
et
al.
"Train
your
own
gnn
teacher:
Graph-aware
distillation
on
textualgraphs."
Joint
European
Conference
on
Machine
Learning
and
Knowledge
Discovery
in
Databases.Cham:
Springer
Nature
Switzerland,
2023.#03大语言模型促进跨领域跨任务的统一图学习大语言模型促进跨领域跨任务的统一图学习“Cross
Domain”
before
LLMsCross
Domain
Graph
Learning
with
LLM大语言模型促进跨领域跨任务的统一图学习“Cross
Domain”
before
LLMsKDD’20:
“We
design
Graph
Contrastive
Coding(GCC)—a
self-supervised
graph
neural
network
pre-training
framework—to
capture
the
universal
networktopological
properties
across
multiple
networks.”Limitation:
the
node
features
are
notthe
same,
among
graphs
fromdifferent
domain.大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLMOne
for
all:Towards
trainingone
graphmodel
for
allclassificationtasks大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLM大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLMOFA
successfully
enableda
single
graph
model
to
be
effective
onall
graph
datasets
across
differentdomains
as
OFA-joint
performs
well
onall
datasets.Further,
we
can
see
that
OFA-jointachieves
better
results
on
most
of
thedatasetspared
toOFA-ind.
Thismay
indicate
that
by
leveraging
thetext
feature,
the
knowledge
learnedfrom
one
domain
can
be
useful
for
thelearning
of
other
domains.大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLMOverview
of
UniGraph
framework.
In
pre-training,
we
employ
a
self-supervised
approach,
leveragingTAGs
to
unify
diverse
graph
data.
This
phase
involves
a
cascaded
architecturebining
LMs
andGNNs.大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLMWe
can
observe
thatpre-training
ongraphs
from
the
same
domain
enhancesthe
performance
of
downstream
tasks.This
suggests
that
in-domain
transferremains
simpler
than
cross-domaintransfer.Experiment
results
in
few-shot
transfer.大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLM大语言模型促进跨领域跨任务的统一图学习Cross
Domain
Graph
Learning
with
LLM#04潜在研究方向What
LLMs
truly
learned
from
GraphsObservation
1:LLMs
interpret
inputs
more
as
contextual
paragraphs
than
as
graphs
with
topologicalstructures.
Neither
linearizing
nor
rewiring
ego-graph
has
significant
impact
on
theclassification
performance
of
LLMs.Linearize
ego-graph:
We
create
a
linearized
version
of
the
graph-structured
prompts
by
onlykeeping
all
neighbors’
text
attributes
in
the
prompts.What
LLMs
truly
learned
from
GraphsObservation
1:LLMs
interpret
inputs
more
as
contextual
paragraphs
than
as
graphs
with
topologicalstructures.
Neither
linearizing
nor
rewiring
ego-graph
has
significant
impact
on
theclassification
performance
of
LLMs.Rewire
ego-graph:We
randomly
rewire
theego-graph
by
differentstrategies.
Then
wepare
theperformance
of
MPNNsand
LLMs
under
eachstrategy.What
LLMs
truly
learned
from
GraphsObservation
2:LLMs
benefit
from
structural
information
only
when
the
neighborhood
is
homophilous,which
means
the
neighbors
contain
phrases
related
to
the
groundtruth
label
of
the
targetnode.What
LLMs
truly
learned
from
GraphsObservation
3:LLMs
benefit
from
structural
information
when
the
target
node
does
not
contain
enoughphrases
for
the
model
to
make
reasonable
prediction.Truly
“Generative”
Cross
Domain
LLM-based
Graph
LearningIs
there
universal
structure
features
that
benefit
for
graph
learningof
graph
from
different
domain?How
can
theseplex
topological
features,
instead
of
the
textcontext,
be
really
captured
by
LLMs?Large
Language
Models
on
Graphs:
Aprehensive
Survey言鹏韦浙江大学信息资源管理系20研究生某著名企业巴巴通义实验室实习生ReferencesZhao
W
X,
Zhou
K,
Li
J,
et
al.
A
survey
of
large
language
models[J].
arXiv
preprint
arXiv:2303.18223,
2023.Huang,
Jin,
et
al.
"Can
llms
effectively
leverage
graph
structural
information:
when
and
why."
arXiv
preprintarXiv:2309.16595
(2023).Tan,
Yanchao,
et
al.
"MuseGraph:
Graph-oriented
Instruction
Tuning
of
Large
Language
Models
for
Generic
GraphMining."
arXiv
preprint
arXiv:2403.04780(2024).He,
Yufei,
and
Bryan
Hooi.
"UniGraph:
Learning
a
Cross-Domain
Graph
Foundation
Model
From
Natural
Language."
arXivpreprint
arXiv:2402.13630
(2024).Liu,
Hao,
et
al.
"One
for
all:
Towards
training
one
graph
model
for
all
classification
tasks."
arXiv
preprintarXiv:2310.00149
(2023).Qiu,
Jiezhong,
et
al.
"Gcc:
Graph
contrastive
coding
for
graph
neural
network
pre-training."
Proceedings
of
the
26th
ACMSIGKDD
international
conference
on
knowledge
discovery
&
data
mining.
2020.Mavromatis,
Costas,
et
al.
"Train
your
own
gnn
teacher:
Graph-aware
distillation
on
textual
graphs."
Joint
European
Conferenceon
Machine
Learning
and
Knowledge
Discovery
in
Databases.
Cham:
Springer
Nature
Switzerland,
2023.Wen,
Z.,
&
Fang,
Y.
(2023).
Prompt
tuning
on
graph-augmented
low-resource
text
classification.
arXiv
preprintarXiv:2307.10230.Tang,
Jiabin,
et
al.
"Graphgpt:
Graph
instruction
tuning
for
large
language
models."
arXiv
preprint
arXiv:2310.13023
(2023).Guo,
J.,
Du,
L.,
&
Liu,
H.
(2023).
Gpt4graph:
Can
large
language
models
understand
graph
structured
data?
an
empiricalevaluation
and
benchmarking.
arXiv
preprint
arXiv:2305.15066.ReferencesXie,
Han,
et
al.
"Graph-aware
language
model
pre-training
on
a
large
graph
corpus
can
help
multiple
graphapplications."
Proceedings
of
the
29th
ACM
SIGKDD
Conference
on
Knowledge
Discovery
and
Data
Mining.
2023.Zhikai
Chen,
Haitao
Mao,
Hang
Li,
Wei
Jin,
Hongzhi
Wen,
Xiaochi
Wei,
Shuaiqiang
Wang,
Dawei
Yin,
Wenqi
Fan,
Hui
Liu,
et
al.Exploring
the
potential
of
large
language
models
(llms)
in
learning
on
graphs.
arXiv
preprint
arXiv:2307.03393,
2023Ruosong
Ye,
Caiqi
Zhang,
Runhui
Wang,
Shuyuan
Xu,
and
Yongfeng
Zhang.
2024.
Language
is
All
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 2026事业单位工勤技能-湖北-湖北政务服务办事员四级(中级工)历年参考题库含答案详解
- 2026事业单位工勤技能-湖北-湖北保健按摩师五级(初级工)历年参考题库含答案详解
- 2026事业单位工勤技能-海南-海南信号工-机车信号设备维修五级(初级工)历年参考题库含答案详解
- 2026事业单位工勤技能-河南-河南计算机信息处理员二级技师历年参考题库含答案详解
- 2026事业单位工勤技能-河南-河南垃圾清扫与处理工五级(初级工)历年参考题库含答案详解
- 2026事业单位工勤技能-河北-河北经济岗位工二级(技师)历年参考题库含答案详解
- 2026事业单位工勤技能-河北-河北堤灌维护工一级(高级技师)历年参考题库含答案详解
- 2026事业单位工勤技能-江西-江西农机驾驶维修工五级(初级工)历年参考题库含答案详解
- 2026事业单位工勤技能-江苏-江苏殡葬服务工二级(技师)历年参考题库含答案详解
- 2026事业单位工勤技能-山西-山西汽车驾驶与维修员一级(高级技师)历年参考题库含答案详解
- 2026辅警轮训考试题及答案
- 苏教版小学二年级科学上全册课程教案设计(凤凰教育)
- 2026-2030中国蓄能器行业现状调查与未来发展前景预测研究报告
- 新闻记者职业资格考试题库(1000题含答案与解析)
- (2026秋新版)北师大版四年级数学上册全册教案
- 2026年河北省中考英语试卷真题及答案详解(精校打印版)
- 神经外科护理质量信息化管理探索
- 【2026 新教材】统编版一年级上册语文识字课教学设计(天地人、金木水火土)
- en590现货销售合同
- 2026年数字电子技术课程测试卷(精练)附答案详解
- 2026年中考数学二轮复习 专题17 二次函数的综合压轴题(高频考点专练)
评论
0/150
提交评论