HiSMatch Historical Structure Matching based Temporal Knowledge Graph Reasoning Zixuan Li123 Zhongni Hou12 Saiping Guan12y Xiaolong Jin12 Weihua Peng3

2025-05-06 0 0 1.58MB 11 页 10玖币

侵权投诉

HiSMatch: Historical Structure Matching based Temporal Knowledge

Graph Reasoning

Zixuan Li1,2,3∗

, Zhongni Hou1,2, Saiping Guan1,2†

, Xiaolong Jin1,2, Weihua Peng3,

Long Bai1,2, Yajuan Lyu3, Wei Li3, Jiafeng Guo1,2, Xueqi Cheng1,2

1School of Computer Science and Technology, University of Chinese Academy of Sciences;

2CAS Key Laboratory of Network Data Science and Technology,

Institute of Computing Technology, Chinese Academy of Sciences; 3Baidu Inc.

{lizixuan, houzhongni18z, guansaiping, jinxiaolong}@ict.ac.cn

{pengweihua,lvyajuan}@baidu.com

Abstract

A Temporal Knowledge Graph (TKG) is a se-

quence of KGs with respective timestamps,

which adopts quadruples in the form of (sub-

ject,relation,object,timestamp) to describe

dynamic facts. TKG reasoning has facili-

tated many real-world applications via answer-

ing such queries as (query entity,query rela-

tion,?,future timestamp) about future. This

is actually a matching task between a query

and candidate entities based on their historical

structures, which reﬂect behavioral trends of

the entities at different timestamps. In addi-

tion, recent KGs provide background knowl-

edge of all the entities, which is also help-

ful for the matching. Thus, in this paper,

we propose the Historical Structure Matching

(HiSMatch) model. It applies two structure

encoders to capture the semantic information

contained in the historical structures of the

query and candidate entities. Besides, it adopts

another encoder to integrate the background

knowledge into the model. TKG reasoning ex-

periments on six benchmark datasets demon-

strate the signiﬁcant improvement of the pro-

posed HiSMatch model, with up to 5.6% per-

formance improvement in MRR, compared to

the state-of-the-art baselines.

1 Introduction

Knowledge Graphs (KGs), which store facts as

triples in the form of (subject,relation,object),

have been widely applied to many NLP applica-

tions, such as question answering (Lan and Jiang,

2020), dialogue generation (He et al.,2017) and

recommendation (Wang et al.,2019). However,

facts may constantly change over time. Tempo-

ral Knowledge Graphs (TKGs) is a kind of KGs

that describe such dynamic facts by extending each

triple with a timestamp as (subject,relation,object,

timestamp). Usually, a TKG is represented as a

∗

This work was done while the ﬁrst author was doing

internship at Baidu Inc.

†Corresponding author.

sequence of KG snapshots. The TKG reasoning

task is to infer new facts from known ones, which

primarily has two settings, interpolation and extrap-

olation. The former attempts to complete missing

facts in history, while the latter aims to predict fu-

ture facts with historical facts. This paper focuses

on the extrapolation setting, which is more chal-

lenging and far from being solved (Jin et al.,2020).

This task can be seen as answering the query about

the future facts (e.g., (COVID-19,Infect,?,2022-

8-1)) by selecting from all the candidate entities.

The key of answering the queries about future

facts is to understand the history thoroughly. All

the existing models conduct reasoning based on

substructures extracted from the whole history.

These substructures can be divided into two types,

i.e., query-related history (Jin et al.,2019;Zhu

et al.,2021) and candidate-related history (Li et al.,

2021b,2022;Han et al.,2021a;Deng et al.,2020).

The former contains the latest historical facts re-

lated to the subject and relation in the query, which

reﬂects the behavioral trends of the subject concern-

ing the query relation. The latter contains all the

latest historical facts of the candidates without con-

sidering the query, which indicates the behavioral

trends of all the entities. Both of these two kinds of

history are vital to TKG reasoning. Take the query

(COVID-19,Infect, ?, 2022-8-1) for example, the

query-related history contains facts like (COVID-

19,Infect, *, t), where tis before 2022-8-1. The

candidate-related history of a candidate A, includes

facts reﬂecting its own behaviors, like (A,*,*,t)

or (*,*,A,t). In the realistic situation, the occur-

rence of the fact (COVID-19,Infect,A,2022-8-1) is

caused by the interactions between these two kinds

of history. However, existing models only focus

on one kind of history and underestimate the other,

which limits their performance on TKG reasoning.

Overall, it still remains a challenge to model both

two kinds of history in a uniﬁed framework.

To reduce the computational cost caused by the

arXiv:2210.09708v1 [cs.AI] 18 Oct 2022

enormous facts in history, these two kinds of his-

tory usually contain one hop facts of the centered

entities. Thus, they cannot model the high-order

associations among the entities, which is also vital

to TKG reasoning.

Motivated by these, we consider both query-

related history and candidate-related history under

a matching framework and propose the

storical

tructure

Match

ing (

HiSMatch

) model. Speciﬁ-

cally, it applies two structure encoders to model

the semantic information in the above two kinds of

historical structures, respectively. Then, it obtains

the matching scores. Both of these two structure

encoders contain three components: (1) a structure

semantic component to model the structure depen-

dencies among concurrent facts at the same times-

tamp; (2) a time semantic component to model

the time numerical information of the historical

facts; (3) a sequential pattern component to mine

the behavioral trends from the temporal order in-

formation. Additionally, to model the high-order

associations among the entities, we consider the

most recent KGs as the background knowledge of

each query and apply a GCN-based background

knowledge encoder to obtain more informative en-

tity representations for the two structure encoders.

Our contributions are summarized as follows:

•

We ﬁrst advocate the importance of modeling

both query-related and candidate-related his-

tory for TKG reasoning and transform the task

into a matching problem between them.

•

To solve this problem, we propose HiSMatch

to comprehensively capture the information

in both historical structures via modeling

the structure dependencies among concurrent

facts, the time numerical information of his-

torical facts and the temporal order among

facts. HiSMatch complementally captures

high-order associations among entities by

modeling the recent background knowledge.

•

Extensive experiments on six commonly

used benchmarks demonstrate that HiSMatch

achieves signiﬁcantly better performance (up

to 5.6% improvement in MRR) on the TKG

reasoning task.

2 Related Work

TKG Reasoning under the interpolation setting

focuses on completing the missing facts at past

timestamps (Liao et al.,2021;Goel et al.,2020;Wu

et al.,2020;Han et al.,2020a;Jiang et al.,2016;

Dasgupta et al.,2018;Garcia-Duran et al.,2018;

Xu et al.,2021). For example, TTransE (Leblay

and Chekol,2018) extends the idea of TransE (Bor-

des et al.,2013) by adding the temporal order con-

straints among facts. Also, HyTE (Dasgupta et al.,

2018) projects the entities and relations to time-

related hyperplanes to generate time-aware repre-

sentations. TNTComplEx (Lacroix et al.,2020)

performs 4th-order tensor factorization to get the

time-aware representations of entities. However,

they cannot obtain the representations of the unseen

timestamps and are not suitable for the extrapola-

tion setting.

TKG Reasoning under the extrapolation set-

ting

aims to predict facts at future timestamps. Ac-

cording to the historical structure the models focus

on, the existing models can be categorized into two

groups: query-based and candidate-based models.

Query-based models focus on modeling the

query-related history. For example, RE-NET (Jin

et al.,2020) models the query-related subgraph

sequence. GHNN (Han et al.,2020c) introduces

the temporal point process to model the precise

time information and takes the 1-hop subgraphs of

the query entity into consideration. CyGNet (Zhu

et al.,2021) captures repetitive patterns by model-

ing repetitive facts with the same subject and rela-

tion to the query. xERTE (Han et al.,2020b) learns

a dynamic pruning procedure to ﬁnd the query-

related subgraphs. CluSTeR (Li et al.,2021a) and

TITer (Sun et al.,2021) both adopt reinforcement

learning to discover query-related paths in history.

Candidate-based models encode the latest his-

torical facts of all the candidate entities without

considering the query, and query are considered

only in the decoding phase. RE-GCN and its ex-

tension CEN (Li et al.,2021b,2022) designs an

evolutional model to get the representations of all

the candidates by modeling history at a few lat-

est timestamps. TANGO (Han et al.,2021a) uti-

lizes neural ordinary differential equations to model

the structure information for each candidate entity.

Glean (Deng et al.,2020) introduces unstructured

textual information to enrich the candidate-related

history.

Above all, none of the existing models focus

on both two kinds of history in a uniﬁed frame-

work. HiSMatch considers these two kinds of his-

tory under the matching framework and takes the

Candidate Structure EncoderQuery Structure Encoder

Sequential Pattern Component

Structure

Semantic

Component

Time Semantic

Component

Sequential Pattern Component

Matching Function

Structure

Semantic

Component

Time Semantic

Component

…

Periodic Activation

Function

FCN Layer

Background

Knowledge

Encoder

… …

……

Query-related Historical Structure

… …

<latexit sha1_base64="KavJDPMS9EyDF0lvQt+TTcogzLU=">AAAB9HicbVBNS8NAEJ3Ur1q/qh69BIvgqSQi6rHgxZNUsB/QxrDZbtulm026OymUkN/hxYMiXv0x3vw3btsctPXBwOO9GWbmBbHgGh3n2yqsrW9sbhW3Szu7e/sH5cOjpo4SRVmDRiJS7YBoJrhkDeQoWDtWjISBYK1gdDvzWxOmNI/kI05j5oVkIHmfU4JG8tJB5qfou9lTOs78csWpOnPYq8TNSQVy1P3yV7cX0SRkEqkgWndcJ0YvJQo5FSwrdRPNYkJHZMA6hkoSMu2l86Mz+8woPbsfKVMS7bn6eyIlodbTMDCdIcGhXvZm4n9eJ8H+jZdyGSfIJF0s6ifCxsieJWD3uGIUxdQQQhU3t9p0SBShaHIqmRDc5ZdXSfOi6l5V3YfLSu0+j6MIJ3AK5+DCNdTgDurQAApjeIZXeLMm1ov1bn0sWgtWPnMMf2B9/gBqTZKP</latexit>

<latexit sha1_base64="K6x9mmdKgoEIUAK0RHAfGyiiTC4=">AAAB9HicbVBNS8NAEJ3Ur1q/qh69BIvgqSQi6rHgxZNUsB/QxrDZbtulm026OymUkN/hxYMiXv0x3vw3btsctPXBwOO9GWbmBbHgGh3n2yqsrW9sbhW3Szu7e/sH5cOjpo4SRVmDRiJS7YBoJrhkDeQoWDtWjISBYK1gdDvzWxOmNI/kI05j5oVkIHmfU4JG8tJB5qfo8+wpHWd+ueJUnTnsVeLmpAI56n75q9uLaBIyiVQQrTuuE6OXEoWcCpaVuolmMaEjMmAdQyUJmfbS+dGZfWaUnt2PlCmJ9lz9PZGSUOtpGJjOkOBQL3sz8T+vk2D/xku5jBNkki4W9RNhY2TPErB7XDGKYmoIoYqbW206JIpQNDmVTAju8surpHlRda+q7sNlpXafx1GEEziFc3DhGmpwB3VoAIUxPMMrvFkT68V6tz4WrQUrnzmGP7A+fwDARZLH</latexit>

<latexit sha1_base64="dnZaEV6WUNFZU6Loos418nwoNMs=">AAAB9HicbVBNSwMxEM3Wr1q/qh69BIvgqeyKqMeCF09SwX5Auy7ZNG1Dk+w2mS2UZX+HFw+KePXHePPfmLZ70NYHA4/3ZpiZF8aCG3Ddb6ewtr6xuVXcLu3s7u0flA+PmiZKNGUNGolIt0NimOCKNYCDYO1YMyJDwVrh6HbmtyZMGx6pR5jGzJdkoHifUwJW8tNBFqQQyOwpHWdBueJW3TnwKvFyUkE56kH5q9uLaCKZAiqIMR3PjcFPiQZOBctK3cSwmNARGbCOpYpIZvx0fnSGz6zSw/1I21KA5+rviZRIY6YytJ2SwNAsezPxP6+TQP/GT7mKE2CKLhb1E4EhwrMEcI9rRkFMLSFUc3srpkOiCQWbU8mG4C2/vEqaF1Xvquo9XFZq93kcRXSCTtE58tA1qqE7VEcNRNEYPaNX9OZMnBfn3flYtBacfOYY/YHz+QPGaZLL</latexit>

Periodic Activation

Function

FCN Layer

Background KG

<latexit sha1_base64="wbU3O+mEpebFJho7nYcDhYbgciI=">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</latexit>

tqti

<latexit sha1_base64="4f1/heYS117oMZ/Nn07a291sRyk=">AAAC3HicjVHLSsNAFD2Nr1pfVRcu3ASL6EJKIqIuC25cSQX7gL5I0mkdzItkIpTQnTtx6w+41e8R/0D/wjtjBLWITkhy5tx7ztw71w5dHgvDeMlpU9Mzs3P5+cLC4tLySnF1rR4HSeSwmhO4QdS0rZi53Gc1wYXLmmHELM92WcO+OpHxxjWLYh74F2IUso5nDX0+4I4liOoVN9qeJS7tQToc91Kx0+Pjbsr23HGvWDLKhlr6JDAzUEK2qkHxGW30EcBBAg8MPgRhFxZielowYSAkroOUuIgQV3GGMQqkTSiLUYZF7BV9h7RrZaxPe+kZK7VDp7j0RqTUsU2agPIiwvI0XcUT5SzZ37xT5SlrG9Hfzrw8YgUuif1L95n5X53sRWCAY9UDp55CxcjunMwlUbciK9e/dCXIISRO4j7FI8KOUn7es640sepd3q2l4q8qU7Jy72S5Cd5klTRg8+c4J0F9v2wels3zg1LlLBt1HpvYwi7N8wgVnKKKmqr/AY940rrajXar3X2karlMs45vS7t/B3kPmV4=</latexit>

ge,l

<latexit sha1_base64="Gc7b/CNVea58Z/OLYzMUD0BVlb4=">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</latexit>

ge0,l

<latexit sha1_base64="82b4fNPOLcQa2rHDil4diQwbGk8=">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</latexit>

<latexit sha1_base64="TahbkboAySbVZvYV8T/Iq3uUNqg=">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</latexit>

<latexit sha1_base64="0lQmjQeXskv7v/B0k+rfzGS9YNU=">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</latexit>

<latexit sha1_base64="e+55lYxnSIUFhT4L+s/oJfc/syI=">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</latexit>

<latexit sha1_base64="/2qFObq8lK6BiOgAVy8Ad59Z6YA=">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</latexit>

<latexit sha1_base64="pCujRHv1fZO4BvlEybf9jRzcpck=">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</latexit>

<latexit sha1_base64="8zDXwjYRE+VtN8BwvG0cTs4qvnI=">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</latexit>

<latexit sha1_base64="pKIHNqOAOEw9XirWKrg6aCIkYJ8=">AAAC2nicjVHLSsNAFD2Nr1pfUXHlJlhEVyURUZcFN66kgn1AW0uSTtvQvEgmYgnduBO3/oBb/SDxD/QvvDOmoBbRCUnOnHvPmbn3WqHrxFzXX3PKzOzc/EJ+sbC0vLK6pq5v1OIgiWxWtQM3iBqWGTPX8VmVO9xljTBipme5rG4NT0W8fs2i2An8Sz4KWdsz+77Tc2yTE9VRt1qeyQdWL70Zd1K+1zHGVykbd9SiXtLl0qaBkYEislUJ1Be00EUAGwk8MPjghF2YiOlpwoCOkLg2UuIiQo6MM4xRIG1CWYwyTGKH9O3TrpmxPu2FZyzVNp3i0huRUsMuaQLKiwiL0zQZT6SzYH/zTqWnuNuI/lbm5RHLMSD2L90k8786UQtHDyeyBodqCiUjqrMzl0R2Rdxc+1IVJ4eQOIG7FI8I21I56bMmNbGsXfTWlPE3mSlYsbez3ATv4pY0YOPnOKdB7aBkHJWMi8Ni+TwbdR7b2ME+zfMYZZyhgip5p3jEE56VlnKr3Cn3n6lKLtNs4ttSHj4AVeeYiw==</latexit>

<latexit sha1_base64="peLc/YLjQkH25LZk+ozOUf5PCUc=">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</latexit>

<latexit sha1_base64="hPRKb+PdNi2Rxsoy7tGnMUuUCT4=">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</latexit>

<latexit sha1_base64="En2m+2cSGgdXIItWrBuzbpwcPy0=">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</latexit>

<latexit sha1_base64="+Dce+D50b1bv3QBUkfvh5WM0fug=">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</latexit>

<latexit sha1_base64="6t/G0F8v0h9HSC04hg0z1fu9hSs=">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</latexit>

<latexit sha1_base64="zBJ/EyuqftiEizu+2BxUM1pCPvY=">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</latexit>

v(tqti)

<latexit sha1_base64="XQDfW3suGBqMbPRWBKeL5dY3kEE=">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</latexit>

<latexit sha1_base64="RHfiG4pMEVZWPrBkGk/bFeiUbvk=">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</latexit>

<latexit sha1_base64="Fa+Q4e5BUYNrOVWMNLJf8KPGQ7s=">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</latexit>

TKG

Candidate-related Historical Structure

…

<latexit sha1_base64="q5B6DDlUhayRGnvob6UHgPZZAyM=">AAAB9XicbVBNS8NAEJ3Ur1q/qh69BIvoqSQi6rHgxZNUsB/QpmWz3bRLN5uwO1FKyP/w4kERr/4Xb/4bt20O2vpg4PHeDDPz/FhwjY7zbRVWVtfWN4qbpa3tnd298v5BU0eJoqxBIxGptk80E1yyBnIUrB0rRkJfsJY/vpn6rUemNI/kA05i5oVkKHnAKUEj9dJh1k/xtO9mvZRl/XLFqToz2MvEzUkFctT75a/uIKJJyCRSQbTuuE6MXkoUcipYVuommsWEjsmQdQyVJGTaS2dXZ/aJUQZ2EClTEu2Z+nsiJaHWk9A3nSHBkV70puJ/XifB4NpLuYwTZJLOFwWJsDGypxHYA64YRTExhFDFza02HRFFKJqgSiYEd/HlZdI8r7qXVff+olK7y+MowhEcwxm4cAU1uIU6NICCgmd4hTfryXqx3q2PeWvBymcO4Q+szx+755K0</latexit>

<latexit sha1_base64="fRxyTFS/JH/jLtCv2zltrc1qVG8=">AAAB9XicbVBNS8NAEN3Ur1q/qh69BIvoqSQi6rHgxZNUsB/QpmWznbRLN5uwO1FKyP/w4kERr/4Xb/4bt20O2vpg4PHeDDPz/FhwjY7zbRVWVtfWN4qbpa3tnd298v5BU0eJYtBgkYhU26caBJfQQI4C2rECGvoCWv74Zuq3HkFpHskHnMTghXQoecAZRSP10mHWT/G0L7NeClm/XHGqzgz2MnFzUiE56v3yV3cQsSQEiUxQrTuuE6OXUoWcCchK3URDTNmYDqFjqKQhaC+dXZ3ZJ0YZ2EGkTEm0Z+rviZSGWk9C33SGFEd60ZuK/3mdBINrL+UyThAkmy8KEmFjZE8jsAdcAUMxMYQyxc2tNhtRRRmaoEomBHfx5WXSPK+6l1X3/qJSu8vjKJIjckzOiEuuSI3ckjppEEYUeSav5M16sl6sd+tj3lqw8plD8gfW5w8Zm5Lx</latexit>

<latexit sha1_base64="vMQRN/ZiM0/FQgcfRUZ9HsULklA=">AAAB9XicbVBNS8NAEJ3Ur1q/qh69BIvoqSQi6rHgxZNUsB/QpmWz3bRLN5uwO1FKyP/w4kERr/4Xb/4bt20O2vpg4PHeDDPz/FhwjY7zbRVWVtfWN4qbpa3tnd298v5BU0eJoqxBIxGptk80E1yyBnIUrB0rRkJfsJY/vpn6rUemNI/kA05i5oVkKHnAKUEj9dJh1k/xtM+zXsqyfrniVJ0Z7GXi5qQCOer98ld3ENEkZBKpIFp3XCdGLyUKORUsK3UTzWJCx2TIOoZKEjLtpbOrM/vEKAM7iJQpifZM/T2RklDrSeibzpDgSC96U/E/r5NgcO2lXMYJMknni4JE2BjZ0wjsAVeMopgYQqji5labjogiFE1QJROCu/jyMmmeV93Lqnt/Uand5XEU4QiO4QxcuIIa3EIdGkBBwTO8wpv1ZL1Y79bHvLVg5TOH8AfW5w8R7pLs</latexit>

<latexit sha1_base64="50un2vwjK3axOwS9Eo7u0c3BYrc=">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</latexit>

v(tqt0

<latexit sha1_base64="52zVWba+SB4zrSJc273aClPsxQo=">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</latexit>

tqt0

<latexit sha1_base64="ZEQckflGpWESGyszPQq2g41iVW4=">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</latexit>

<latexit sha1_base64="ShVXSGTUVU8lCQRgczeUexCwvHk=">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</latexit>

<latexit sha1_base64="hEpvD7xwnjzFH5C7cI+KwiWYpZI=">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</latexit>

Figure 1: An illustrative diagram of the proposed HiSMatch model.

advantages of both kinds of models.

3 Problem Formulation

A TKG

G={G0, ..., Gt, ..., GT}

is a sequence

of KGs, each of which contains facts occurred at

timestamp

, i.e.,

Gt={E,R,Ft}

, where

the set of entities,

is the set of relations and

is the set of facts that occurred at

. Each

fact is a quadruple

(es, r, eo, t)

, where

es, eo∈ E

and

r∈ R

. For each fact in TKG, we add the

inverse quadruple

(eo, r−1, es, t)

into TKG, cor-

respondingly. The TKG reasoning task aims to

predict the missing object via answering a query

q= (eq, rq,?, tq)

with the historical KGs given.

Note that, when predicting the missing subject of a

query

q={?, rq, eq, tq}

, we can convert the query

into q={eq, r−1

q,?, tq}.

4 The HiSMatch model

HiSMatch aims to captures the semantic similar-

ity contained in the query-related and candidate-

related historical structures. For each query time,

it ﬁrst embeds the background knowledge into the

initial entity representations. With the initial rep-

resentations as input, it maps the semantic infor-

mation in these two historical structures into the

vectorized representations of structures. Based on

the structure representations, matching scores are

calculated.

Thus, as shown in Figure 1, HiSMatch consists

of four parts: the query structure encoder, the can-

didate structure encoder, the background knowl-

edge encoder, and the matching function. First,

two kinds of historical structures and a background

knowledge graph are derived from the TKG. Then,

the background knowledge encoder gets the rep-

resentations of the entities with the background

knowledge graph as input (Section 4.3). With

the learned representations as input, two structure

encoders use three components to integrate three

kinds of semantic information into the represen-

tations of query-related structure and candidate-

related structure, respectively (Section 4.1 and

4.2). Finally, the matching function calculates the

scores between the query and candidates based

on the representations of their historical structures

(Section 4.4).

4.1 Query Structure Encoder

The query-related historical structure should re-

ﬂect the behavioral trends of the query. Motivated

by this, for a query

q= (eq, rq,?, tq)

, the query-

related historical structure consists of the latest his-

torical facts with the same subject

and relation

. These facts co-occuring at the same timestamp

form a subgraph

centered on

. Then, we

obtain a subgraph sequence

{gq

t1, ..., gq

ti, ..., gq

tm}

where

t1< ... < ti< ... < tm< tq

and

is the

maximum length of the sequence.

Three kinds of information are vital in the above

historical structure, namely, the structure semantic

information of each subgraph, the time numerical

information of each subgraph, and the temporal or-

der information across subgraphs. To model these

three kinds of information, we design three compo-

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

HiSMatch:HistoricalStructureMatchingbasedTemporalKnowledgeGraphReasoningZixuanLi1,2,3,ZhongniHou1,2,SaipingGuan1,2y,XiaolongJin1,2,WeihuaPeng3,LongBai1,2,YajuanLyu3,WeiLi3,JiafengGuo1,2,XueqiCheng1,21SchoolofComputerScienceandTechnology,UniversityofChineseAcademyofSciences;2CASKeyLaboratoryofNetwor...

展开>> 收起<<

HiSMatch Historical Structure Matching based Temporal Knowledge Graph Reasoning Zixuan Li123 Zhongni Hou12 Saiping Guan12y Xiaolong Jin12 Weihua Peng3.pdf

共11页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库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，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

HiSMatch Historical Structure Matching based Temporal Knowledge Graph Reasoning Zixuan Li123 Zhongni Hou12 Saiping Guan12y Xiaolong Jin12 Weihua Peng3

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: