挖掘异构信息网络：原理与方法

5星 · 超过95%的资源需积分: 9 56 浏览量更新于2024-07-22 收藏 2.36MB PDF 举报

《挖掘异构信息网络：原理与方法》是Synthesis Lectures on Data Mining and Knowledge Discovery系列中的一本专著，由知名学者 Jiawei Han、Lise Getoor、Wei Wang、Johannes Gehrke 和 Robert Grossman共同担任系列编辑。该书主要探讨了在多源、多类型数据环境中进行知识发现和信息挖掘的关键理论与实践方法。异构信息网络(Heterogeneous Information Networks, HIN)的研究涉及到多种网络结构和数据类型，如社交网络、语义网络、生物网络等，它们在现实世界中有广泛应用，如推荐系统、知识图谱构建和复杂事件分析。异构信息网络的特点在于其节点和边具有不同的属性和类型，这增加了数据的复杂性和多样性。作者Yizhou Sun和Jiawei Han在书中详细阐述了处理这种复杂性时所面临的技术挑战，包括数据融合、跨域关联挖掘、关系抽取以及网络表示学习等核心问题。他们强调了理解不同网络结构间的相互作用、共享特征以及如何提取隐含知识的重要性。书中可能涵盖的主题包括： 1. **异构网络建模**：介绍如何构建和表示异构网络，包括节点和边的属性类型，以及如何设计适合不同网络结构的数据模型。 2. **关系抽取**：讨论从文本或网络中识别和提取实体之间的不同类型关系的技术，如实体识别、关系识别和关系分类。 3. **融合与转换**：探讨如何将来自不同源的异构数据整合到一个统一的框架中，以便进行更有效的分析。 4. **跨域分析**：针对异构网络中不同领域或类型的节点和边，研究如何进行跨领域关联挖掘和知识迁移。 5. **社区发现与聚类**：针对异构网络中的复杂结构，提出针对不同节点属性的社区检测算法，以揭示隐藏的组织模式。 6. **网络嵌入**：介绍将异构网络中的节点表示为低维向量的方法，以便于后续的机器学习任务，如推荐系统和分类。 7. **应用示例**：通过实际案例展示如何在搜索引擎优化、社交媒体分析、医疗信息检索等领域应用异构信息网络挖掘技术。《挖掘异构信息网络：原理与方法》是一本深入探讨异构数据处理技术的专业著作，为读者提供了理论基础和实践经验，对于从事数据挖掘、知识发现以及网络科学的科研人员和工程师来说，是一份宝贵的参考资料。

2 1. INTRODUCTION

are powerful and expressive representations of general real-world interactions between different kinds

of network entities in diverse domains.

In this book, we investigate the principles and methodologies for mining heterogeneous

information networks, by leveraging the semantic meaning of the types of nodes and links in a

network,and propose models and algorithms that can exploit such rich semantics and solve real-world

problems. Heterogeneous information networks often imply rather different semantic structures

from that in homogeneous networks. Links in heterogeneous networks indicate the interactions

between various types of objects in a network, and usually imply similarity or inﬂuence among these

objects, that can be difﬁcult to be expressed by traditional features. Information is propagated across

various kinds of objects in a network, via various kinds of relationships (i.e., heterogeneous links),

carrying different semantics and having different strengths in determining the “inﬂuence” across

linked objects. These principles have laid the foundation for methodologies of handling various

mining tasks in heterogeneous information networks, including ranking, clustering, classiﬁcation,

similarity search, relationship prediction, and relation strength learning. We will introduce these

mining tasks and their associated new principles and methodologies chapter by chapter.

1.1 WHAT ARE HETEROGENEOUS INFORMATION

NETWORKS?

An information network represents an abstraction of the real world, focusing on the objects and the

interactions between the objects. It turns out that this level of abstraction has great power in not

only representing and storing the essential information about the real world, but also providing a

useful tool to mining knowledge from it, by exploring the power of links. Formally, we deﬁne an

information network as follows.

Deﬁnition 1.1 (Information network) An information network is deﬁned as a directed graph

G = (

V, E) with an object type mapping function τ : V → A and a link type mapping function

φ :

E → R, where each object v ∈ V belongs to one particular object type τ(v) ∈ A, each link e ∈ E

belongs to a particular relation φ(e) ∈ R, and if two links belong to the same relation type, the two

links share the same starting object type as well as the ending object type.

Different from the traditional network deﬁnition, we explicitly distinguish object types and

relationship types in the network. Note that, if a relation exists from type A to type B, denoted as

ARB, the inverse relation R

−1

holds naturally for BR

−1

A. R and its inverse R

−1

are usually not

equal, unless the two types are the same and R is symmetric.When the types of objects |

A| > 1 or the

types of relations |

R| > 1, the network is called heterogeneous information network; otherwise,

it is a homogeneous information network.

Given a complex heterogeneous information network, it is necessary to provide its meta

level (i.e., schema-level) description for better understanding the object types and link types in the

1.1. WHAT ARE HETEROGENEOUS INFORMATION NETWORKS? 3

network. Therefore, we propose the concept of network schema to describe the meta structure of a

network.

Deﬁnition 1.2 (Network schema) The network schema, denoted as T

= (A, R), is a meta tem-

plate for a heterogeneous network G = (

V, E) with the object type mapping τ : V → A and the

link mapping φ :

E → R, which is a directed graph deﬁned over object types A, with edges as

relations from

The network schema of a heterogeneous information network has speciﬁed type constraints

on the sets of objects and relationships between the objects.These constraints make a heterogeneous

information network semi-structured, guiding the exploration of the semantics of the network.

Heterogeneous information networks can be constructed from many interconnected, large-

scale datasets, ranging from social, scientiﬁc, engineering, to business applications. Here are a few

examples of such networks.

1. Bibliographic information network. A bibliographic information network, such as the com-

puter science bibliographic information network derived from DBLP, is a typical hetero-

geneous network, containing objects in four types of entities: paper (P), venue (i.e., confer-

ence/journal) (V), author (A), and term (T). For each paper p ∈ P , it has links to a set of

authors, a venue, and a set of terms, belonging to a set of link types. It may also contain ci-

tation information for some papers, that is, these papers have links to a set of papers cited by

the paper and a set of papers citing the paper.

The network schema for a bibliographic network and an instance of such a network are shown

in Figure 1.1.

2. Twitter information network. Twitter as a social media can also be considered as an informa-

tion network, containing objects types such as user, tweet, hashtag, and term, and relation (or

link) types such as follow between users, post between users and tweets, reply between tweets,

use between tweets and terms, and contain between tweets and hashtags.

3. Flickr information network. The photo sharing website Flickr can be viewed as an informa-

tion network, containing a set of object types: image, user, tag, group, and comment, and a set

of relation types, such as upload between users and images, contain between images and tags,

belong to between images and groups, post between users and comments, and comment between

comments and images.

4. Healthcare information network. A healthcare system can be modeled as a healthcare in-

formation network, containing a set of object types, such as doctor, patient, disease, treatment,

and device, and a set of relation types, such as used-for between treatments and diseases, have

between patients and diseases, and visit between patients and doctors.

1.2. WHY IS MINING HETEROGENEOUS NETWORKS A NEW GAME? 5

in general captures much richer information than its homogeneous network counterpart. A homo-

geneous information network is usually obtained by projection from a heterogeneous information

network, but with signiﬁcant information loss. For example, a co-author network can be obtained

by projection on co-author information from a more complete heterogeneous bibliographic net-

work. However, such projection will lose valuable information on what subjects and which papers

the authors were collaborating on. Moreover, with rich heterogeneous information preserved in an

original heterogeneous information network, many powerful and novel data mining functions can

be developed to explore the rich information hidden in the heterogeneous links across entities.

Why is mining heterogeneous networks a new game? Clearly, information propagation across

heterogeneous node and links can be very different from that across homogeneous nodes and links.

Based on our research into mining heterogeneous information networks, especially our studies on

ranking-based clustering [66; 69], ranking-based classiﬁcation [30; 31], meta-path-based similar-

ity search [65], relationship prediction [62; 63], relation strength learning [61; 67], and network

evolution [68], we believe there are a set of new principles that may guide systematic analysis of

heterogeneous information networks. We summarize these principles as follows.

1. Information propagation across heterogeneous links. Similar to most of the network ana-

lytic studies, links should be used for information propagation. However, the new game is how

to propagate information across heterogeneous types of nodes and links , in particular, how to compute

ranking scores, similarity scores, and clusters, and how to make good use of class labels, across

heterogeneous nodes and links. No matter how we work out new, delicate measures, deﬁni-

tions, and methodologies, a golden principle is that objects in the networks are interdependent

and knowledge c an only be mined using the holistic information in a network.

2. Search and mining by exploring network meta structures. Different from homogeneous

information networks where objects and links are being treated either as of the same type or

as of un-typed nodes or links, heterogeneous information networks in our model are semi-

structured and typed, that is, associated with nodes and links structured by a set of types,

forming a network schema. The network schema provides a meta structure of the information

network. It provides guidance of search and mining of the network and helps analyze and

understand the semantic meaning of the objects and relations in the network. Meta-path-

based similarity search and mining introduced in this book demonstrate the usefulness and

the power of exploring network meta structures.

3. User-guided exploration of infor mation networks. In a heterogeneous information network,

there often exist numerous semantic relationships across multiple types of objects, carrying

subtly different semantic meanings. A certain weighted combination of relations or meta-

paths may best ﬁt a speciﬁc application for a particular user. Therefore, it is often desirable to

automatically select the right relation (or meta-path) combinations with appropriate weights

for a particular search or mining task based on user’s guidance or feedback. User-guided or

feedback-based network exploration is a useful strategy.

6 1. INTRODUCTION

1.3 ORGANIZATION OF THE BOOK

The ﬁrst chapter introduces the problem of mining heterogeneous information networks. After

that, the book is organized into three parts, each containing two chapters that present principles and

methodologies for mining heterogeneous information networks, organized according to different

mining tasks. Finally, Chapter 8 outlines a few open research themes in this domain. The major

contents of Chapters 2–7 are summarized as follows.

In Par t I: Ranking-Based Clustering and Classiﬁcation, we introduce several studies on ba-

sic mining tasks such as clustering and classiﬁcation in heterogeneous information networks, by

distinguishing the information propagation across different types of links.

• Chapter 2: Ranking-based clustering. For link-based clustering of heterogeneous informa-

tion networks, we need to explore links across heterogeneous types of data. Recent studies

develop a ranking-based clustering approach (e.g., RankClus [66] and NetClus [69]) that

generates both clustering and ranking results efﬁciently. This approach is based on the ob-

servation that ranking and clustering can mutually enhance each other because objects highly

ranked in each cluster may contribute more towards unambiguous clustering, and objects more

dedicated to a cluster will be more likely to be highly ranked in the same cluster.

• Chapter 3: Classiﬁcation of heterogeneous information networks. Classiﬁcation can also

take advantage of links in heterogeneous information networks. Knowledge can be effectively

propagated across a heterogeneous network because the nodes of the same type that are linked

similarly via the same typed links are likely to be similar. Moreover, following the idea of

ranking-based clustering, one can explore ranking-based classiﬁcation since objects highly

ranked in a class are likely to play a more important role in classiﬁcation. These ideas lead to

effective algorithms, such as GNetMine [31] and RankClass [30].

In Part II: Meta-Path-Based Similarity Search and Mining, we introduce a systematic approach

for dealing with general heterogeneous information networks with a speciﬁed network schema,

using a meta-path-based methodology. Under this framework, similarity search and other mining

tasks such as relationship prediction can be addressed by systematic exploration of the network meta

structure.

• Chapter 4: Meta-path-based similarity search. Similarity search plays an important role in

the analysis of networks. By considering different linkage paths (i.e., meta-path) in a network,

one can derive various semantics on similarity in a heterogeneous information network. A

meta-path-based similarity measure, PathSim, is introduced in [65], which aims at ﬁnding

peer objects in the network. PathSim turns out to be more meaningful in many scenarios

compared with random-walk-based similarity measures.

• Chapter 5: Meta-path-based relationship prediction. Heterogeneous information network

brings interactions among multiple types of objects and hence the possibility of predicting re-

lationships across heterogeneous typed objects. By systematically designing meta-path-based

剩余160页未读，继续阅读

Quantum_bit

粉丝: 2
资源: 39

挖掘异构信息网络：原理与方法

Recommendation in Heterogeneous Information Networks

Ranking-Based Classification of Heterogeneous Information Networks

Minimized-Cost Cube Query in Heterogeneous Information Networks

Meta Structure- Computing Relevance in Large Heterogeneous Information Networks

Aspect-Level Deep Collaborative Filtering via Heterogeneous Information Networks

Ranking-based Clustering on General Heterogeneous Information Networks by Network Projection

MetaStructure: Heterogeneous Information Networks的关联度计算新方法

Anonymous Network Information Acquirement Protocol for Mobile Users in Heterogeneous Wireless Networks

Advances in Mining Heterogeneous Healthcare Data.pdf

Learning influence from heterogeneous social networks

最新资源