数据文件 - 麻省理工学院人工智能研究专业

数据文件

Microsoft Concept Graph: Mining Semantic
Concepts for Short Text Understanding

Lei Ji1,2†, Yujing Wang1, Botian Shi3, Dawei Zhang4, Zhongyuan Wang5 & Jun Yan6

1Microsoft Research Asia, Haidian District, 北京 100080, 中国

2Institute of Computing Technology, Chinese Academy of Sciences, Haidian District, 北京 100049, 中国

3Beijing Institute of Technology, Haidian District, 北京 100081, 中国

4MIX Labs, Haidian District, 北京 100080, 中国

5Meituan NLP Center, Chaoyang District, 北京 100020, 中国

6AI Lab of Yiducloud Inc., Huayuan North Road, Haidian District, 北京 100089, 中国

关键词: Knowledge extraction; 概念化; Text understanding

引文: L. 吉, 是. 王, 乙. Shi, D. 张, Z. 王 & J. 严. Microsoft concept graph: Mining semantic concepts for short text

理解. 数据智能 1(2019). 土井: 10.1162/dint_a_00013

已收到: 十一月 20, 2018; 修改: 二月 12, 2019; 公认: 二月 19, 2019

抽象的

Knowlege is important for text-related applications. 在本文中, we introduce Microsoft Concept Graph,
a knowledge graph engine that provides concept tagging APIs to facilitate the understanding of human
语言. Microsoft Concept Graph is built upon Probase, a universal probabilistic taxonomy consisting of
instances and concepts mined from the Web. We start by introducing the construction of the knowledge
graph through iterative semantic extraction and taxonomy construction procedures, which extract 2.7 百万
concepts from 1.68 billion Web pages. We then use conceptualization models to represent text in the
concept space to empower text-related applications, such as topic search, query recommendation, Web
table understanding and Ads relevance. Since the release in 2016, Microsoft Concept Graph has received
多于 100,000 pageviews, 2 million API calls and 3,000 registered downloads from 50,000 visitors over
64 国家.

† 通讯作者: Lei Ji (电子邮件: leiji@microsoft.com; ORCID: 0000-0002-7569-3265).

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

1. 介绍

Concepts are indispensable for humans and machines to understand the semantic meanings underlined
in the raw text. Humans understand an instance, especially an unfamiliar instance, by its basic concept in
an appropriate level, which is defined as Basic-level Categorization (BLC) by psychologists and linguists.
例如, people may not understand “Gor Mahia”, but with the concept “football club” described in
维基百科, people can capture the semantic meaning easily. Psychologist Gregory Murphy began his highly
acclaimed book [1] with the statement “Concepts are the glue that holds our mental world together”. 自然
magazine book review [2] calls it an understatement, because “Without concepts, there would be no
mental world in the first place”. To enable machines to understand the concept of an instance like human
beings, one needs a knowledge graph consisted of instances, 概念, as well as their relations. 然而,
we observe two major limitations in existing knowledge graphs, which motivate us to build a brand-new
knowledge taxonomy, Probase [3], to tackle general purpose understanding of human language.

1). Previous taxonomies have limited concept space. Many existing taxonomies are constructed by
human experts and are difficult to be scaled up. 例如, Cyc project [4] contains about 120,000
concepts after 25 years of evolution. Some other projects, like Freebase [5], resort to crowd sourcing
efforts to increase the concept space, which still lacks general coverage of many other domains and
thus holds a barrier for general purpose text understanding. There are also automatically constructed
knowledge graphs, such as YAGO [6] and NELL [7]. 尽管如此, the coverages of these concept
spaces are still limited. The number of concepts of Probase and some other popular open-domain
taxonomies are shown in Table 1, which demonstrates that Probase has a much larger concept space.

桌子 1. Concept space comparison of existing taxonomies.

Existing taxonomies

Number of concepts

Freebase [5]
WordNet [8]
WikiTaxonomy [9]
YAGO [6]
DBPedia [10]
Cyc [4]
NELL [7]
Probase [3]

1,450
25,229
111,654
352,297
259
≈120,000
123
≈5,400,000

2). Previous taxonomies treat knowledge as black and white. Traditional knowledge base aims at
providing standard, well-defined and consistent reusable knowledge, and treats knowledge as black
and white. 然而, treating knowledge as black and white obviously has restrictions, 尤其
when the concept space is extremely large, because different knowledge has different confidence
intervals or probabilities, and the best threshold of probabilities depends on the specific application.
Different from previous taxonomies, our philosophy is to annotate knowledge facts with probabilities
and let the application itself decide the best way of using it. We believe this design is more flexible
and can be utilized to benefit a broader range of text-based applications.

239

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

Based on the Probase knowledge taxonomy, we propose a novel conceptualization model to learn text
representation in the Probase concept space. The conceptualization model (also known as the Concept
Tagging Model) aims to map text into semantic concept categories with some probabilities. It provides
computers the commonsense computing capability and makes machines “aware” of the mental world of
human beings, through which machines can better understand human communication in text. 基于
the conceptualization model, the Probase taxonomy can be applied to facilitating various text-based
applications, including topic search, query recommendation, Ads relevance, Web table understanding, ETC.

An overview of the entire framework is illustrated in Figure 1, which mainly consists of three layers:

(1) knowledge construction layer, (2) conceptualization layer, 和 (3) application layer.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 1. The framework overviews.

1). Knowledge construction layer. In the knowledge construction layer, we address the construction
of Probase knowledge network. 第一的, the isA facts are mined automatically from the Web via a
semantic iteration extraction procedure. 第二, the taxonomy is constructed following a rule-based
taxonomy construction algorithm. 最后, we calculate the probability score for each related
instance/concept in the taxonomy.

数据智能

240

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

2). Conceptualization layer. Based on the constructed sematic knowledge network, we design a
conceptualization model to represent raw text in the Probase concept space. The model can be
divided into three major components, 即 (1) single instance conceptualization, (2) 语境-
aware single instance conceptualization, 和 (3) short text conceptualization.

3). Application layer. The conceptualization model enables us to generate “Bag-of-Concepts”
representation of raw text, which can be utilized to enhance various categories of applications,
including but not limited to topic search, query recommendation, Ads relevance calculation and
Web table understanding. We also build a Probase browser and a tagging model demo in the
application layer, which provide users a quick insight into a specific query.

The rest of this paper is organized as follows. 部分 2 讨论相关工作, and Section 3 presents
the construction of the Probase semantic network. 部分 4 introduces the conceptualization model built
upon the sematic network, and Section 5 shows some example applications. 此外, 部分 6 lists the
data sets and statistics. 最后, we make a conclusion in Section 7.

2. RELATED WORKS

Knowledge graph has attracted great interests in many research fields. There are many existing knowledge

graphs built either manually or automatically.

1). WordNet [8] Different from traditional thesauruses, WordNet groups words together based on their
meanings instead of morphology. Terms are grouped into synsets, where each term represents a
distinct concept. Synsets are interlinked by conceptual semantics and lexical relations. WordNet
has more than 117,000 synsets for 146,000 条款. WordNet is widely used for term disambiguation
as it defines various senses for a term.

2). DBpedia [10] is a project of extracting Wikipedia Infobox into knowledge facts which can be
semantically queried using SPARQL. The knowledge in DBpedia is extracted from Wikipedia and
collaboratively edited by the community. DBpedia has released 670 million triples in RDF format.
DBpedia provides an ontology including 280 类, 3.5 million entities and 9,000 属性.
3). YAGO [6], abbreviation for Yet Another Great Ontology, is an open sourced huge semantic knowledge
根据, which has fused knowledge extracted from WordNet, GeoNames and Wikipedia. YAGO
combines the taxonomy of WordNet with the Wikipedia category to assign entities to more than
200,000 类. YAGO is comprised by more than 120 million facts about 3 million entities. Based
on manual evaluation, the accuracy of YAGO is about 95%. YAGO also attaches temporal and
special information to the entities and relations by some declarative extraction rules.

 http://www.geonames.org/
 https://www.wikipedia.org/

241

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

4). Freebase [5] is a large knowledge base containing a lot of human labeled data contributed by
community members. Freebase extracts data from many sources including Wikipedia, NNDB,
Fashion Model Directory and MusicBrainz. Freebase has more than 125 million facts, 4,000 类型
和 7,000 特性. 后 2016, all data in Freebase have been transferred to Wikidata.

5). ConceptNet [11] is a semantic network aiming to build a large commonsense knowledge base in
a crowd sourcing manner, which focuses on the commonsense relations including isA, partOf,
usedFor and capableOf. ConceptNet contains over 21 million edges and over 8 million nodes.
6). NELL [7] is a continuously running system which extracts facts from text in hundreds of millions of
Web pages in an iterative way, while patterns are boosted during the process. NELL has accumulated
more than 50 million candidate beliefs and has extracted 2,810,379 asserted instances of 1,186
different categories and relations.

7). WikiTaxonomy [9] presents a taxonomy automatically generated from the categories in Wikipedia,
which generates 105,418 isA links from a network of 127,325 categories and 267,707 links. 这
system achieves 87.9 balanced F-measure according to a manual evaluation on the taxonomy.
8). KnowItAll [12] aims to automate the process of extracting large collections of facts from the
Web in an unsupervised, domain-independent and scalable manner. The system has three major
成分: Pattern Learning (PL), Subclass Extraction (SE) and List Extraction (LE), achieving great
improvements on the recall while maintaining precision and enhancing the extraction rate.

The existing knowledge graphs suffer from low coverage and lack of well-organized taxonomies.
而且, none of them focuses on extracting the super-concepts and sub-concepts of an instance. 到
automatically generate the taxonomy, we propose Probase which automatically constructs the semantic
network of isA facts.

3. SEMANTIC NETWORK CONSTRUCTION

The entire data construction procedure of the Probase semantic network will be introduced in detail in
the following subsections. 第一的, in Section 3.1, we describe the iterative data extraction process; 然后, 这
taxonomy construction step is introduced in Section 3.2; 部分 3.3 introduces the algorithm of probability
calculation.

3.1 Iterative Extraction

The Probase semantic network [3] is built upon isA facts extracted from the Web. The isA facts can be
formulated as pairs consisting of a super-concept and a sub-concept. 例如, “cat isA animal” forms
a pair, where “cat” is the sub-concept and “animal” is the super-concept. 在这项工作中, we propose a method
based on semantic iteration to mine the isA pairs from Web.

 http://www.nndb.com/
 https://www.fashionmodeldirectory.com/
 https://musicbrainz.org
 https://www.wikidata.org

数据智能

242

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

3.1.1 Syntactic vs. Semantic Iteration

Before our work, the state-of-the-art information extraction methods [7, 8, 12] rely on a syntactic
integrative (bootstrapping) 方法. It starts with a set of seed patterns to discover example pairs with high
confidence. 然后, we can derive new patterns from the extracted pairs and use the new patterns to extract
more pairs. The iterative process is continued until a certain stop criterion is reached. 然而, we observe
in practice that the syntactic iteration methods have indispensable barriers in deep knowledge acquisition
because high quality syntactic patterns are rare. 所以, we propose a semantic interactive approach by
which the new pairs can be extracted with high confidence based on knowledge accumulated from existing
对.

3.1.2 The Semantic Iteration Algorithm

第一的, we extract a set of candidate pairs by Hearst patterns [13] (桌子 2). 例如, if we have
sentence “… animals such as cat …”, we can extract a pair which means cat isA animal.
有时, there is ambiguity in the pattern matching process. 例如, in the sentence “… animals
other than dogs such as cats …”, we can extract two candidate pairs , 和 .
The algorithm must have the ability to decide which one is the correct super-concept among “animals” and
“dogs”. Another example is “… companies such as IBM, Nokia, Proctor and Gamble …”. 在这种情况下, 我们
have multiple choices in the sub-concept, 即, 和 < companies, Proctor and Gamble>. 再次, the algorithm should automatically determine their correctness.

桌子 2. Hearst patterns examples.

图案

NP such as {NP,}*{(或者 | 和)} NP
Such NP as {NP,}*{(或者 | 和)} NP
NP{,} 包括 {NP,}* {(或者 | 和)} NP
NP{,NP}*{,} and other NP
NP{,NP}*{,} or other NP
NP{,} 尤其 {NP,}*{(或者 | 和)} NP

1
2
3
4
5
6

We denote the candidate set of super-concepts for sentence s as Xs, and the candidate set of sub-concepts
for sentence s as Ys. Γ is the set of isA pairs that have already been extracted as ground truth. The remaining
task for the algorithm is to detect the correct super-concepts and sub-concepts from Xs and Ys, 分别,
based on the knowledge accumulated i n Γ. For each sentence s, if we have multiple choices for the super-
概念, we must choose one as the correct super-concept. It is based on the observation that when
ambiguity exists in super-concept matching, there is only one correct answer. After determining the correct
super-concept, the goal is to filter out the correct sub-concepts from candidates in Ys. Unlike the super-
concept case, we may expect more than one valid sub-concept, and the result sub-concepts should be a
subset of Ys.

243

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

3.1.3 Super-Concept Detection

Let

x x
,
{
1

……
,

米

we compute likelihood p(xk|Ys) for each candidate xk. We assume

y y
2,
1

are independent with each other given any super-concept xk, and then we have

… ∈
,

y
, n

是
s

p x Y
k
s

(

∝

)

p x p Y x
)
(

(

= ∏
p x
(
)

)

我

p y x
我

(

(1)

where p(xk) is the percentage of pairs in Γ that contain xk as the super-concept, 和 p(做|xk) is the percentage
of pairs in Γ that have yi as the sub-concept when the super-concept is xk. There are pairs that do not exist
in Γ, 因此, we let p(做|xk) = ε if no existence can be found for that pair; where ε is set to be a small
positive number. Without losing generality, we assume x1 and x2 are the top two candidates with the highest
probability scores, 和 p(x1|Ys) > p(x2|Ys). 然后, we can compute the ratio of two probabilities as follows:
∏
p x
)
(
1
∏
)

p y x
)
我
1

r x x
(
,
1

p x
(

(2)

=
我
n

(

)

p y x
我

(

我

x1 will be chosen as the correct super-concept if r(x1,x2) is larger than a pre-defined threshold. 如果不, 这
sentence is skipped in the current iteration and may be recovered in a later round when Γ is more informative.
直观地, the likelihood p (动物|cats) should be much larger than p (dogs|cats); 所以, we can correctly
select “animals” as the result super-concept.

3.1.4 Sub-Concept Detection

In our implementation, sub-concept detection is based on the features extracted from the original

句子, which is motivated by two basic observations.

Observation 1: The closer a candidate sub-concept is to the pattern keyword, the more likely it is a

valid sub-concept.

Observation 2: If we are certain that a candidate sub-concept at the k-th position (calculated by the
distance from the pattern keyword) is valid, the candidate sub-concepts from position 1 to position k−1 are
also likely to be correct.

例如, consider the following sentence: “… representatives in North America, 欧洲, the Middle
East, 澳大利亚, 墨西哥, 巴西, 日本, 中国, and other countries …”. Here “and other” is the pattern
keyword; China is the candidate sub-concept closest to the pattern keyword. 明显地, China is a correct
sub-concept. 此外, if we know that Australia is a legal sub-concept of “countries”, we can be more
confident that Mexico, Brazil and Japan are all correct sub-concepts of the same category; while North
美国, Europe and the Middle East are less likely to be the instances of the same category.

具体来说, we find the largest k such that the likelihood p(yk|X) is above a pre-defined threshold. 然后,
y1, y2,…, yk are all treated as valid sub-concepts of the super-concept x. Note that there are sometimes
ambiguity in the sub-concept yi. 例如, in the sentence “… companies such as IBM, Nokia, Proctor

数据智能

244

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

and Gamble…” at position 3, the noun phrase can be “Proctor” or “Proctor and Gamble”. Suppose the
candidate at the position k to be c1, c2,…, we calculate the conditional probability of each candidate given
the super-concept x as well as all sub-concepts before position k:

p y
(

c x y y

| ,

,
1

我

……
,

)
1

(
p c x
|
我

)

∏

p y c x
|
j

(

我

(3)

As before, y1, y2,…, yk–1 are assumed to be independent given the super-concept x; p(词|X) is the percentage
of existing pairs in Γ where ci is a sub-concept when x is the super-concept; p(yj|词, X) is the likelihood that
yj is a valid sub-concept given x is the super-concept and ci is another valid sub-concept in the same
句子. Suppose c1 and c2 are the top 2 ranked concepts, and we pick c1 as the final concept if r(c1, c2)
exceeds a certain ratio. The value of r(c1, c2) can be calculated as:

r c c
,
(
1

)

p c x
(
| )
1

p c
(

X
| )

∏
∏

=
-

1
1

j
k

p y c x
|
j

,
1

(

)

p y c x
|

(

)

(4)

直观地, the probability p(Proctor and Gamble|公司) should be much larger than p(Proctor|
公司) after Γ accumulates enough knowledge, and thus the algorithm is able to select the correct
candidate automatically.

3.2 Taxonomy Construction

The taxonomy construction procedure mainly consists of three steps, (1) local taxonomy construction,
(2) horizontal merge and (3) vertical merge. 第一的, we build the local taxonomy for each sentence. 然后,
we perform horizontal merge and vertical merge in sequence on the local taxonomy collection to construct
a global taxonomy.

1). Local Taxonomy Construction. 数字 2 illustrates the process to build a local taxonomy from
each sentence. 例如, in the sentence “A such as B, C, D”, we can get three pairs ,
, 和 where A is the super-concept and B, C, D are the sub-concepts. We can build
a local taxonomy with A as the root node and B, C, D as child nodes.

数字 2. Local taxonomy construction.

245

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

2). Horizontal Merge. After the local taxonomy for each sentence is built, the next step is horizontal
merge. Suppose we have two local taxonomies T 1
A whose root is A1 and A2, 分别, 在哪里
A1 and A2 correspond to the same surface form. If we are sure that A1 and A2 express the same sense,
we can merge the two trees horizontally, 如图 3. We design a similarity function to
calculate the probability that A1 and A2 are semantically equal. 直观地, if the children of A1 and
A2 are more overlapped, we can be more confident that they are of the same sense. 所以, 我们
; 和一个
adopt absolute overlap for the similarity calculation function, IE。,
constant threshold d is used to determine whether two local taxonomies can be horizontally merged.

(
1
f A A
,

A and T 2

1
A

∩

)

数字 3. Horizontal merge.

3). Vertical Merge. Given two local taxonomies rooted by Ai and B1, where B is a child of Ai. If we
are confident that B is of the same sense as B1, we can merge the two taxonomies vertically. 作为
shown in Figure 4, after merging, Ai is the root; the original subtree B is merged with the taxonomy
rooted by B1; and the other subtrees C and D remain at the same position. To determine if B and B1
are of the same sense, we calculate the absolute overlap of R1’s children and B’s siblings (emphasized
by colored nodes in Figure 4).

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

数字 4. Vertical merge: single sense alignment.

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

There is another case which is more complicated (数字 5). Both T 1

A, as the child nodes of R1 and R2 have considerable overlap with B’s siblings in T i

B can be vertically merged
with T i
A. 然而, 这
child nodes of R1 and R2 do not have enough overlap, indicating that R1 and R2 may express different senses.
在这种情况下, we split two senses in the merged taxonomy, IE。, 时间 1
B are merged as two distinct sub-
trees in taxonomy T i
A.

B and T 2

数据智能

246

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

数字 5. Vertical merge: multiple sense alignment.

3.3 Probability Calculation

Probability is an important feature of our taxonomy design. Different from previous approaches which
treat knowledge as black and white, our solution is to live with noisy data with probability scores and let
the application make the best use of it. We provide two kinds of probability scores, namely plausibility and
typicality.

Plausibility is the joint probability of an isA pair. For each isA claim E, we use p(乙) to denote its
probability to be true. Here we adopt a simple noisy-or model to calculate the probability. Assume E is
derived from a set of sentences {s1, s2, ……, sn}, a claim is false if every piece of evidence from {s1, s2, ……, sn}
is false. Assuming every piece of evidence is independent, 我们有:

( )
乙

= −
1

( )
乙

= −
1

−∏
(1

我

p s
(
我

)),

(5)

where p(si) is the probability of evidence si to be true. We characterize each si by a set of features Fi
包括: (1) the PageRank score of the Web page from which sentence si is extracted; (2) the Hearst pattern
used to extract evidence pair from si; (3) the number of sentences with x as the super-concept; (4)
the number of sentences with y as the sub-concept; (5) the number of sub-concepts extracted from sentence
si; (6) position of y in the sub-concepts list from sentence si. Supposing the features are independent, 我们
apply Naïve Bayes [14] to derive p(si) based on the corresponding feature set Fi. We exploit WordNet to
build a training set used for learning the Naïve Bayes Model. Given a pair that appears in the
WordNet, if there is a path between x and y in the WordNet taxonomy, (IE。, x is an ancestor of y), 我们
regard the pair as a positive example; 否则, we treat the pair as negative.

Typicality is the conditional probability between a super-concept and its instance (sub-concept).
直观地, robin is more typical of the concept bird than is ostrich, while Microsoft is more typical of the
concept company than is Xyz inc. 所以, we need a probability score to stand for the typicality of an
实例 (sub-concept) to its corresponding super-concept. The typicality measure of an instance i to a
super-concept x is formulated as:

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

(

i x
|

)

∑

⋅
n x i p x i
( , )
( , )
′
n x i p x i
)
( ,
( ,

∈′
我

我

′

)

(6)

247

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

where x is a super-concept, i is an instance of x, n(X, 我) is the number of appearance that supports i as an
instance of x; p(X, 我) is the plausibility of pair , and Ix is a set consisting of all instances of super-concept
X. 例如, suppose x = Company, i = Microsoft, j = Xyz inc. It can be imagined that n(X, 我) 应该
be much larger than n(X, j), so Microsoft should obtain a much bigger typicality score than Xyz Inc with
respect to Company. 相似地, we can also define the typicality score denoting the probability of a concept
x to instance i.

(

x i
|

)

∑

⋅
n x i p x i
( , )
′
′
(

( , )
(

n x i p x i
, )
, )

∈′
x I

我

(7)

4. CONCEPTUALIZATION

In this section, we introduce the conceptualization model which leverages the Probase semantic
knowledge network to facilitate text understanding. Conceptualization model (also known as the Concept
Tagging model) aims to map text into semantic concept categories with some probabilities. It provides
computers the common sense of semantics and makes machines “aware” of the mental world of human
beings, through which the machines can better understand human communication in text.

We consider three sub-tasks for building a conceptualization model. (1) Single Instance Conceptualization,
which returns Basic-Level Categorization (BLC) for a single instance. 举个例子, “Microsoft” could be
automatically mapped to Software Company and Fortune 500 company, ETC。, with some prior probabilities.
(2) Context-Aware Single Instance Conceptualization, which produces the most appropriate concepts based
on different contexts. 举个例子, “Apple” could be mapped to Fruit or Company without context, 但
with context word “pie”, “Apple” should be mapped to Fruit with higher probability. (3) Short Text
概念化, which returns the types and concepts related to a short sequence of text. 例如,
in the sentence “He is playing game on Apple iPhone and eating an apple”, the first “Apple” is Company
while the second “Apple” is Fruit.

4.1 Single Instance Conceptualization

Assume e is an instance, and c is a super-concept; we can obtain the Basic-Level Categorization (BLC)
results based on typicality scores P(e|C) 和P(C|e), where P(e|C) denotes the typicality of an instance e to
concept c, 和P(C|e) denotes the typicality of a concept c to instance e. We propose several metrics as
representative measures used for BLC [15].

MI is the mutual information between e and c, 定义为:

PMI denotes the pointwise mutual information, which is a widely used measure of the association

(
MI e c

)

)
P e c log
,

(

P e c
)
( ,
( )
P e P c
(

)

(8)

between two discrete variables.

PMI e c

( ,

)

日志

P e c
)
( ,
( )
P e P c
(

)

日志

数据智能

)

P e c P c
(
|
(
( )
P e P c
)

(

)

(
log P e c
|

)

logP e

( ).

(9)

248

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

NPMI is a normalized version of PMI, which is proposed to make PMI less sensitive to occurrence

frequency and is more interpretable.

NPMI e c

( ,

)

PMI e c
( ,
)
-
logP e c
( ,

)

)
(
log P e c
|
-
logP e c

( ,

logP e
( )
)

(10)

Both PMI and NPMI suffer from the trade-off between general and specific concepts. 一方面,
general concepts may be the correct answer, but they do not have the capability to distinguish instances
of different sub-categories; 另一方面, specific concepts may be more representative, 但是
coverage is low. 所以, we further propose PMIk and Graph Traversal measures to tackle this problem.

PMIk makes a compromise to avoid producing either too general or too specific concepts.

(
Rep e c

)

)
P c e P e c

(

If we take the logarithm of scoring function Rep(e,C), we get:

日志

Rep e c

( ,

)

日志

P e c
)
( ,
( )
P e P c
(

)

(
PMI e c

)

日志 ( ,

P e c

This in fact corresponds to PMI2, which is a normalized form in the PMIk Family [16].

(11)

(12)

Graph Traversal is a common way to calculate the relatedness of two nodes in a large network. 这

scores calculated by general graph transversal can be formulated as:

(
Time e c

)

∑

∞

(

)
k
2 *

(
P e c
k

)

时间

∑
(
时间

(

1
+

)
(
P e c
k
,
2 *
k
(
)
∑
1 *
1

时间

)

= +

∞

k T

∑
(
P e c
k

)

(

)
k
2 *

(
P e c
k

)

≥

∑

(

)
k
2 *

(
P e c
k

)

时间

(13)

where Pk(e,C) is the probability of staring from e to c and back to e in 2k steps. When k = 2 which represents
a 4-step random walk, 我们有:

′
Time e c

( ,

=
) 2 *

)
P c e P e c
|

(

)

4 * (1

P c e P e c
| ))

(

= −

)) 4 2 * ( | )

P c e P e c
|

(

)

= −

4 2 *

Rep e c

( ,

(14)

因此, it is verified that the simple, easy-to-compute scoring method of Rep(e,C) is equivalent to a graph

traversal approach under the constraint of 4-step random walk.

4.2 Context-Aware Single Instance Conceptualization

One instance may be mapped to distinct concepts according to different contexts. 例如, 为了
“apple ipad”, we want to annotate “apple” with company or brand, and “ipad” with device or product.
The major challenge is how to distinguish and detect the correct concepts in different contexts. We propose
a framework of context-aware single instance conceptualization [17] which consists of two parts: (1) 离线
knowledge graph construction and (2) online concept inference.

249

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

4.2.1 Offline Knowledge Graph Construction

There are millions of concepts in the Probase semantic network. We first perform clustering on all the
concepts to construct the concept clusters. Concretely, we adopt a K-Medoids clustering algorithm [18] 到
group the concepts into 5,000 clusters. We adopt concept clusters instead of concepts themselves in the
graph for noise reduction and computation efficiency. In the rest of this section, we use concept to denote
concept cluster. We build a large knowledge graph offline, which is comprised of three kinds of sub-graphs,
包括 (1) instance to concept graph, (2) concept to concept graph and (3) non-instance term to concept
图形. 数字 6 shows a piece of the graph around the term watch.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 6. A subgraph of heterogeneous semantic network around watch.

Instance to concept graph. We directly use the Probase semantic network as the instance to concept
图形. 磷(C|e) is served as the typicality probability of concept (cluster) c to instance e, which can be
computed by

(
P c e

)

= ∑

(
*
P c e
|
C

∈
*

)

(15)

where c* represents a concept belonging to cluster c, P*(c*|e) is the typicality of the concept c* to instance e.

Concept to concept graph. We assign a transition probability P(c2|c1) to the edge between two concepts
c1 and c2. The probability is derived by aggregating the co-occurrences between all (明确的) instances
of the two concepts.

(
P c c
2
1

)

∑
∑ ∑

∈
c C

∈
e c e
我

n e e
我

(

)

∈
C

∈

e c e
我

n e e
我
∈
C

(

)

(16)

where c denotes a set containing all concepts (clusters), and the denominator is applied for normalization.
在实践中, we only take the top 25 related concepts for each concept for edge construction.

Non-instance term to concept graph. There are terms that cannot be matched to any instances or
概念, IE。, verbs or adjectives. For better understanding of the short text, we also mine lexical relationships

数据智能

250

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

between non-instance terms and concepts. 具体来说, we use the Stanford Parser [19] to obtain POS
tagging and dependency relationships between tokens in the text, and the POS tagging results reveal the
类型 (例如, 形容词, 动词, ETC。) of each token. Our goal is to obtain the following two distributions.

磷(z|t) stands for the prior probability that a term t is of a particular type z. 例如, the word watch
在哪里

is a verb with probability P(动词|手表) = 0.8374. We compute the probability as
n(t, z) is the frequency term t annotated as type z in the corpus, and n(t) is the total frequency of term t.

n t z
( ,
n t
( )

(
P z t
|

)

磷(C|t, z) denotes the probability of a concept c, given the term t of a specific type z. 例如,
磷(movie|手表, 动词) depicts how likely the verb watch is associated with the concept movie lexically.
Speciﬁcally, we detect co-occurrence relationships between instances, verbs and adjectives in Web
documents parsed by the Stanford Parser. To obtain meaningful co-occurrence relationships, we require
that the co-occurrence be embodied by dependency, rather than merely appearing in the same sentence.
We first obtain P(e|t, z), which denotes how likely a term t of type z co-occurs with instance e:

(

P e t z

| ,

)

n e t
( , )
z
*
n e t
(
, )

∑

(17)

where nz(e, t) is the frequency of term t and instance e having a dependency relationship when t is of type
z. 然后, taking instances as the bridge, we calculate the relatedness between non-instance terms (adjectives/
动词) and concepts.

(

P c t z

| ,

)

∑

(

P c e t z

, | ,

)

∑

P c e
| )

(

P e t z
| ,

(

(18)

In Equation (18), e ∈ c means that e is an instance of concept c, and we make an assumption that a

concept c is conditionally independent with term t and type z when the instance e is given.

∈
e c

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

4.2.2 Online Concept Inference

We adopt the heterogeneous semantic graph built offline to annotate the concepts for a query. 第一的, 我们
segment the query into a set of candidate terms which use Probase as lexicon and identify all occurrences
of terms in the query. 第二, we retrieve the subgraph out of the heterogeneous semantic graph by
concentrating on the query terms. 最后, we perform multiple random walks on the sub-graph to find the
concepts with the highest weights after convergence.

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

4.3 Short Text Conceptualization

Short text is hard to understand in three aspects: (1) Compared with long sentences, short sentences lack
syntactic features and cannot directly apply POS tagging; (2) Short texts do not have sufficient statistical
signals; (3) Short text is usually ambiguous due to the lack of context terms. Many research works focus on
statistical approaches like topic models [20], which extract latent topics as implicit semantics. 然而,
we argue that semantic knowledge is needed to get a better understanding of short texts. 因此, we aim to

251

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

utilize the semantic knowledge network to enhance short text understanding [21]. Different from the
knowledge graph built for context-aware single instance conceptualization (部分 4.2), we add a novel
sub-graph, typed-term co-occurrence graph, which is an undirected graph where nodes are typed-terms
and edge weights represent the lexical co-occurrence between two typed-terms. 尽管如此, 号码
of typed-terms is extremely large, which will increase storage cost and affect the efficiency of calculation
on the network. 所以, we compress the original typed-term co-occurrence network by retrieving
Probase concepts for each instance. 然后, the typed-terms can be replaced by the related concept clusters
and the edge weights are aggregated accordingly. 数字 7 shows an example of the compression procedure.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 7. The compression procedure of typed-term co-occurrence network.

Given a piece of short text, each term is associated with the type and corresponding concepts. We define
the types as instance, 动词, adjective and concept. For each term with type instance, we also learn the
corresponding concepts for the term. Given a piece of short text, the online inference procedure contains
three steps, including text segmentation, type detection and concept labeling. An example is illustrated in
数字 8. Given the query “b ook disneyland hotel california”, the algorithm first segments it as “book
disneyland hotel california”; 然后, it detects the type for each segment; at last, the concepts are labeled for
each instance. As shown in the figure, the output is “book[v] disneyland[e](park) hotel[C] california[e](状态)”, 哪个
means that book is a verb, disneyland is an instance of the concept park, hotel is a concept, and california
is an instance of the concept state.

数字 8. An example of short text understanding.

数据智能

252

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

4.3.1 Text Segmentation

The goal of text segmentation is to divide a short text into a sequence of meaningful components. 到
determine a valid segmentation, we define two heuristic rules: (1) except for stop words, each word belongs
to one and only one term; (2) terms are coherent (IE。, terms mutually reinforce each other). 直观地,
{april in paris lyrics} is a better segmentation of “april in paris lyrics” than {april in paris lyrics}, since “lyrics”
is more semantically related to songs than months or cities. 相似地, {vacation april in paris} is a better
segmentation of “vacation april in paris” than {vacation april in paris}, because “vacation”, “april” and
“paris” are highly coherent with each other, while “vacation” and “april in paris” are less coherent.

The text segmentation algorithm can be conducted in the following steps. 第一的, we construct a term graph
(TG) which consists of candidate terms and their relationships. 下一个, we add an edge between two candidate
terms when they are not mutually exclusive and set the edge weight to reﬂect the strength of mutual
reinforcement. 最后, the problem of ﬁnding the best segmentation is transformed into the task of ﬁnding
a clique in the original TG, 和 100% word coverage while maximizing the average edge weights.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

4.3.2 Type Detection

The type detection procedure annotates the type for each term as verb, 形容词, instance or concept.
例如, term “watch” appears in the instance-list, concept-list, as well as verb-list of our vocabulary,
and thus the possible typed-terms of “watch” are watch[C], 手表[e] and watch[v]. For each term, the type
detection algorithm determines the best typed-term from the set of possible candidates. In the case of
“watch free movie”, the best typed-terms for “watch”, “free”, and “movie” are watch[v], 自由的[adj] and movie[C],
分别.

Traditional approaches resort to POS tagging algorithms which consider lexical features only, 例如, Markov
模型 [22]. 然而, such surface features are insufficient to determine types of terms especially in the
case of short text. In our solution, we calculate the probability by considering both traditional lexical
features and semantic coherence features. We formulate the problem of type detection into a graph model
and propose two models, namely Chain model and Pairwise model.

Chain model borrows the idea of ﬁrst order bilexical grammar and considers topical coherence between
adjacent typed-terms. 尤其, we build a chain-like graph where nodes are typed-terms retrieved from
the original short text. Edges are added between each pair of typed-terms mapped from adjacent terms in
the original text sequence, and the edge weights between typed-terms are calculated by affinity scores (看
the example in Figure 9(A)). Chain model only considers semantic relations between consecutive terms
which will lead to mistakes.

Pairwise model adds edges between typed-terms mapped from each pair of terms rather than adjacent
terms only. As an example, in Figure 9 (乙), there are edges between non-adjacent terms “watch” and
“movie”. The goal of the Pairwise Model is to ﬁnd the best sequence of typed-terms which guarantees that
the maximum spanning tree (MST) constructed by the selected sequence has the largest total weight. 作为

253

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

shown in Figure 9 (乙), as long as the total weight of edges between (手表[v], movie[C]) 和 (自由的[adj], movie[C])
is the largest, {手表[v], 自由的[adj], movie[C]} can be successfully recognized as the best sequence of type
detection for the query “watch free movie”. We employ the Pairwise model in our system as it achieves
higher accuracy experimentally compared with the Chain model.

(A) Type detection result of “watch free movie
using the Chain model is {手表[e], 自由的[ad j],
movie[C]}”.

(乙) Type detection result of “watch free movie
using the Pairwise model is {手表[v], 自由的[ad j],
movie[C]}”.

数字 9. Example of Chain model and Pairwise model.

4.3.3 Concept Labeling

The goal of concept labeling is to re-rank the candidate concepts according to the context for each
实例. The most challenging task in concept labeling is to deal with ambiguous instances. Our intuition
is that a concept is appropriate for an instance only if it is a common semantic concept of that instance
and is supported by the surrounding context at the same time. Take “hotel California eagles” as an example,
where both animal and music band are popular semantic concepts to describe “eagles”. If we find a concept
song in the context, we can be more confident that music band should be the correct concept for “eagles”.

After type detection, we have obtained a set of instances for concept labeling. For each instance, 我们
collect a set of concept candidates and perform instance disambiguation based on a Weighted-Vote
方法, which is a combination of Self-Vote and Context-Vote. Self-Vote denotes the original affinity
重量 (calculated by normalized co-occurrence) of a concept cluster c associated to the target instance;
while Context-Vote leverages the affinity weights between the target instance and other concepts found in
the context. In the case of “hotel california eagles”, the original concept vector of the instance eagles is
(, , , , ……) and the concept vector for
context “hotel california” is (, , , ,
……). After disambiguation by Weighted-Vote, the final conceptualization result of eagles is (,
, , ……).

数据智能

254

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

5. APPLICATION

Probase Browser shows the backbone of the Probase taxonomy and provides an interface to search for
super-concepts, sub-concepts, as well as similar concepts corresponding to the given query. 数字 10 是一个
snapshot of the Probase browser.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 10. A snapshot of the Probase browser.

Tagging Service provides the capability of tagging a piece of text with a concept vector, based on which
the semantic similarity can be calculated, and various text processing applications can be affiliated. 数字
11 shows a snapshot of the instance conceptualization demo when querying a single instance “python”.
数字 12 shows the snapshot given “apple” in the context “pie” and “ipad”, 分别. As shown in the
figure, our tagging service can map the term “apple” into correct concepts under different contexts. 一个
example of short text conceptualization is illustrated in Figure 13 by querying the tagging service with
“apple engineer is eating the apple”. The result shows the capability of our tagging algorithm to distinguish
different senses of “apple” in the short text scenario.

255

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

数字 11. Snapshot of single instance conceptualization.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

数字 12. Snapshot of context-aware single instance conceptualization.

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数据智能

256

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 13. An example of short text conceptualization.

Topic Search [23] aims at understanding the topic underlined in each query for better search relevance.
数字 14 shows a snapshot when a user queries “database conference in Asian cities”. As shown in the
figure, the search results correctly rank VLDB 2002, 2006, 和 2010 at the top, which are held in Hong
孔, Seoul and Singapore, 分别. Traditional Web search takes queries as sequences of keywords
instead of understanding the semantic meanings, so it is hard to generate the correct answers for the
example query. To achieve this goal, we present a framework that improves Web search experiences using
Probase knowledge and the conceptualization models. 第一的, it classifies Web queries into different patterns
according to the concepts and entities in addition to keywords contained in the queries. 然后, it produces
answers by interpreting the queries with the help of Probase semantic concepts. Our preliminary results
showed that the framework was able to understand various types of topic-like queries and achieved much
higher user satisfaction.

257

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 14. The framework of topic search.

Understanding Web Tables [24]. We use Probase to help interpret and understand Web tables, 哪个
unlocks the wealth of information hidden in the Web pages. To tackle this problem, we build a pipeline
for detecting Web tables, understanding their contents, and applying the derived knowledge to support
semantic search. 从 300 million Web documents, we extract 1.95 billion raw HTML tables. Many of
them do not contain useful or relational information (例如, used for page layout purpose); others have
structures that are too complicated for machines to understand. We use a rule-based filtering method to
获得 65.5 million tables (3.4% of all the raw tables) that contain potentially useful information. We adopt
Probase taxonomy to facilitate the understanding of table content, by associating the table content with
one or more semantic concepts in Probase. Based on the knowledge mined from Web tables, we build a
semantic search engine over tables to demonstrate how structured data can empower information retrieval
on the Web. A snapshot of our semantic search engine is shown in Figure 15. As illustrated in the figure,
when a user queries “American politicians birthday”, the search engine returns with aggregated Web tables
consisting of birthday and other related knowledge of various American politicians.

数据智能

258

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 15. Snapshot of the Web tables.

Channel-based Query Recommendation [25] aims to anticipate user search needs when browsing
different channels, by recommending the hottest and highly related queries for a given channel. As shown
图中 16, there are three channels, 消息, Sports and Entertainment, and several queries are recommended
under each channel to enable the users to explore the hottest topics related to the target channel. 之一
the main challenges is how to represent queries and channels which are short pieces of text. Traditional

259

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

representation methodology treats text as bag of words, which suffers from mismatch of surface forms
of words, especially when text is short. 所以, we leverage the Probase taxonomy to obtain a “Bag of
Concepts” representation for each query and channel, in order to avoid surface mismatching and handle
the problem of synonym and polysemy. By leveraging the large taxonomy knowledge base of Probase, 我们
learn a concept model for each category, and conceptualize a short text to a set of relevant concepts.
而且, a concept-based similarity mechanism is presented to classify the given short text to the most
similar category. Experiments showed that our framework could map queries to channels with a high degree
of precision (average precision = 90.3%).

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

数字 16. Query recommendation snapshot.

Ads Relevance [3]. In sponsored search, the search engine maps each query to the related ad bidding
keywords. Since both the query and bidding keywords are short, the traditional bag-of-words approaches
do not work well in this scenario. 所以, we can leverage Probase concept taxonomy to enhance Ads
relevance calculation. For each short text, we first identify instances from it, and map it to basic-level
concepts with score Rep(e, C); 然后, we merge the concepts to generate a concept vector representing the
semantics of the target text. 最后, the relevance score can be calculated through the cosine similarity
measure between the concept vectors of the query and bidding keywords. We conduct our experiments
on real Ads click logs collected from Microsoft Bing search engine. We calculate the relevance score of
each candidate pair of query and bidding keywords, divide the pairs into 10 buckets based on the relevance
分数, and report the average clickthrough rate (CTR) within each bucket. The result is demonstrated in
数字 17, which shows that the CTR numbers have a strong linear correlation with the relevance scores
calculated by our model.

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数据智能

260

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 17. The correlation of C TR with Ads relevance score.

5. DATA AND ANALYSIS

5.1 数据

The Microsoft Concept Graph can be downloaded at https://concept.research.microsoft.com/, 这是
a sub-graph of the semantic network we introduce in this paper. The core taxonomy of Microsoft Concept
Graph contains above 5.4 million concepts, whose distribution is shown in Figure 18, where Y axis is the
number of instances each concept contains (logarithmic scale), and on the X axis are the 5.4 百万
concepts ordered by their size. Our concept space is much larger than other existing knowledge bases
(Freebase contains no more than 2,000 概念, and Cyc has about 120,000 概念).

261

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

数字 18. The distribution of concepts in Microsoft Concept Graph.

5.2 Concept Coverage

Based on the query logs of Microsoft Bing search engine, we estimate the coverage of concepts mined
by our methodology. If at least one concept can be found in a query, the query is considered to be covered.
We thus calculate the percentage of queries that can be covered by Probase and compare the metrics
against the other four taxonomies. We utilize the top 50 million queries in Bing’s query log to compute the
coverage metrics, and the results are shown in Figure 19. We can see clearly from the figure that Probase
has the largest coverage, YAGO ranks the second, and Freebase has a rather low coverage although its
instance space is large. It demonstrates that Freebase’s instance distribution is very skew (most instances
are distributed in several top categories and lack the general coverage of other concepts).

数据智能

262

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

数字 19. Concept coverage of different taxonomies.

5.3 Precision of isA pairs

To estimate the correctness of extracted isA pairs in Probase, we create a benchmark data set of 40
concepts in various domains. For each concept, we randomly pick up 50 instances (sub-concepts) and ask
human reviewers to evaluate their correctness. 数字 20 depicts the precision of isA pairs for all the 40
concepts we manually evaluate. The overall precision is 92.8% by averaging over all the concepts.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

数字 20. Precision of extracted isA pairs on 40 概念.

We further draw the curve of average precision after each iteration (shown in Figure 21). 而且, 这
number of concepts and isA pairs discovered after each iteration are drawn in Figure 22. It is obvious that
the precision degrades after each round while the number of discovered concepts and isA pairs increases.
所以, the best iteration number must be chosen as a trade-off between precision and facts number (set as 11
in our implementation).

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

263

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

数字 21. Precision of isA pairs after each iteration.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数字 22. The number of discovered concepts and isA pairs after each iteration.

5.4 Conceptualization Experiment

在这个部分, we mainly present the experimental results of conceptualization for both single instance

and short text.

1). Single Instance Conceptualization

Data set preparation. We asked human labelers to manually annotate the correctness of the concepts.
The label is defined in four categories as shown in Table 3. 每个 (实例, 概念) is assigned to three
labelers to annotate. The final label is defined by the majority label. We will ask the fourth annotator to
make the final vote for records without majority label. In all, 有 5,049 labeled records.

桌子 3. Labeling guideline for conceptualization.

Label

意义

Examples

Excellent

Well matched concepts at the basic level

(bluetooth, wireless communication protocol)

好的

Fair

坏的

A little general or speciﬁc

(bluetooth, accessory)

Too general or speciﬁc

Non-sense concepts

(bluetooth, feature)

(bluetooth, 问题)

数据智能

264

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

Measurement. We employ Precision@k and nDCG to evaluate the effectiveness. For Precision@k, 我们
treat Good/Excellent as score 1, and Bad/Fair as 0. We calculate the precision of top-k concepts as follows:

Precision@

∑

相对

我

=
1
k

(19)

where reli is the score we deﬁne above. For nDCG, we treat Bad as 0, Fair as 1, Good as 2, and Excellent
作为 3. Then we calculate nDCG as follows:

nDCG

相对

我

∑

我

ideal

相对

我

∑

我

相对
我
logi
ideal

_
logi

相对

我

(20)

where reli is the relevance score of the result at rank i, and ideal reli is the relevance score at rank i of an
ideal list, obtained by sorting all relevant concepts in decreasing order of the relevance score.

Experimental result. 数字 23 shows the evaluation of the top 20 results using both Precision and
nDCG with and without smoothing. We compare our proposed scoring functions with various baselines:
MI, NPMI, and PMI3. Where PMI3 is defined as:

From the result, we can see that our proposed scoring function outperforms baseline on both precision

3
采购经理人指数

(

e c
,

)

日志

p e c
)
( ,
( )
p e p c
(

)

(21)

and nDCG.

2). Short Text Conceptualization

Data set preparation. To validate our generalizability, we build two data sets to evaluate our algorithm
including user search query and tweets. To build a query data set, we first manually picked 11 ambiguous
terms including “apple”, “fox” with instance ambiguity, “watch”, “book”, “pink”, “blue”, “population”,
“birthday” with type ambiguity, and “April in Paris”, “hotel California” with segmentation ambiguity. 然后
we randomly selected 1,100 queries containing 11 ambiguous terms. 而且, we randomly sampled
其他 400 queries without any restriction. In all, there are 1,500 queries. To build tweets data set, 我们
also randomly sampled 1,500 tweets using Twitter’s API. To clean the tweets, we removed some tweet-
specific features, such as @username, hashtags, urls, ETC. We asked human labelers to manually annotate
the correctness. For each record, we assign at lease three labelers and pick up the majority vote as final
标签.

Experimental result. To evaluate the effectiveness of the algorithm, we compared our methods with
several baseline methods. [27] conducts instance disambiguation in queries based on similar instances,
[28] conducts instance disambiguation in queries based on related instances, 和 [26] conducts instance
disambiguation in tweets based on both similar and related instances. 桌子 4 presents the results which
show that our conceptualization is not only effective but also robust.

265

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

数字 23. Precision and nDCG comparison.

桌子 4. Precision of short text understanding.

Query
Tweet

[27]

0.694
–

[28]

0.701
–

[26]

–
0.841

Ours

0.943
0.922

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

数据智能

266

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

6. 结论

在本文中, we present the end-to-end framework of building and utilizing the Microsoft Concept
图形, which consists of three major layers, namely semantic network construction, concept conceptualization
and applications. In the semantic network construction layer, we focus on the knowledge extraction and
taxonomy construction procedures to build an open-domain and probabilistic taxonomy known as Probase.
Like human mental process, we then represent text by concept space using conceptualization models, 和
empower many applications including topic search, query recommendation, Ads relevance calculation as
well as Web table understanding. The system has received wide public attention ever since released in
2016 (more than 100,000 pageviews, 2 million API calls and 3,000 registered downloads from 50,000
visitors over 64 国家).

作者贡献

All of the authors contributed equally to the work. J. 严 (jun.yan@yiducloud.cn), Z. 王 (wzhy@
outlook.com) 和D. 张 (zhangdawei@outlook.com) used to be the leaders of the Microsoft Concept
Graph system, and now L. 吉 (leiji@microsoft.com, corresponding author) is the leader and is developing
new component for V2.0. 是. 王 (Yujing.Wang@microsoft.com) 和L. Ji summarized the methodology
part of this paper. 是. Wang and B. Shi (botianshi@bit.edu.cn) summarized the applications and experiment.
All authors have made meaningful and valuable contributions in revising and proofreading the manuscript.

参考

[1] G. 墨菲. The big book of concepts. 剑桥, 嘛: 麻省理工学院出版社, 2004. 国际标准书号: 9780262632997.
[2] 磷. Bloom. Glue for the mental world. 自然 421 (2003), 212–213. 土井: 10.1038/421212A.
[3] 瓦. 吴, H. 李, H. 王, & K. 朱. Probase: A probabilistic taxonomy for text understanding. 在: 会议记录
的 2012 ACM SIGMOD International Conference on Management of Data, ACM, 2012, PP. 217–228.
土井: 10.1145/2213836.2213891.

[4] D.B. 莱纳特, & R.V. Guha. Building large knowledge-based systems: Representation and inference in the Cyc

项目. Reading, 嘛: Addison-Wesley, 1990. 国际标准书号: 9780201517521.

[6]

[5] K. Bollacker, C. 埃文斯, 磷. Paritosh, 时间. Sturge, & J. 泰勒. Freebase: A collaboratively created graph database
for structuring human knowledge. 在: 诉讼程序 2008 ACM SIGMOD International Conference on
Management of Data, ACM, 2008, PP. 1247–1250. 土井: 10.1145/1376616.1376746.
F.M. Suchanek, G. Kasneci, & G. Weikum. YAGO: A core of semantic knowledge. 在: Proceedings of the 16th
International Conference on World Wide Web, ACM, 2007, PP. 697–706. 土井: 10.1145/1242572.1242667.
[7] A. 卡尔森, J. Betteridge, 乙. Kisiel, 乙. Settles, E.R.H. 小。, & T.M. 米切尔. Toward an architecture for never-
ending language learning. 在: Proceedings of the 24th Conference on Artificial Intelligence, AAAI, 2010,
PP. 1306–1313. 土井: 10.1145/1807406.1807449.

[8] G.A. 磨坊主. WordNet: A lexical database for English. ACM 通讯 38(11) (1995), 39–41.

[9]

土井: 10.1145/219717.219748.
S.P. Ponzetto, & 中号. Strube. Deriving a large-scale taxonomy from Wikipedia. 在: Proceedings of the 22nd
National Conference on Artificial Intelligence, AAAI, 2007, PP. 1440–1445. 可用于: http://www.aaai.
org/Papers/AAAI/2007/AAAI07-228.pdf.

[10] S. Auer, C. Bizer, G. Kobilarov, J. Lehmann, 右. Cyganiak, & Z. Ives. DBpedia: A nucleus for a web of open
数据. 在: Proceedings of the 6th International Semantic Web Conference/Asian Semantic Web Conference,
施普林格, 2007, PP. 722–735. 土井: 10.1007/978-3-540-76298-0_52.

267

数据智能

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

[11] 右. Speer, & C. Havasi. Representing general relational knowledge in ConceptNet 5. 在: 诉讼程序
8th International Conference on Language Resources and Evaluation, 2012, PP. 3679–3686. 可用于:
http://www.lrec-conf.org/proceedings/lrec2012/pdf/1072_Paper.pdf.

[12] 氧. Etzioni, 中号. Cafarella, D. Downey, A.P. Shaked, S. Soderland, D.S. Weld, & A. Yates. Web-scale
information extraction in knowitall: Preliminary results. 在: Proceedings of the 13th International Conference
on World Wide Web, ACM, 2004, PP. 100–110. 土井: 10.1145/988672.988687.

[13] 嘛. Hearst. Automatic acquisition of hyponyms from large text corpora. 在: Proceedings of the 14th confer-

ence on Computational linguistics, 前交叉韧带, 1992, PP. 539–545. 土井: 10.3115/992133.992154.

[14] A. 麦卡勒姆, & K. Nigam. A comparison of event models for naive bayes text classification. 在: 会议记录

of AAAI-98 Workshop on Learning for Text Categorization, AAAI, 1998, PP. 41–48.

[15] Z. 王, H.X. 王, J. Wen, & 是. Xiao. An inference approach to basic level of categorization. 在: Proceed-
ings of the 24th ACM International Conference on Information and Knowledge Management (CIKM), ACM,
2015, PP. 653–662. 土井: 10.1145/2806416.2806533.

[16] 乙. Daille. Approche mixte pour l’extraction de terminologie: Statistique lexicale et filtres linguistiques. 博士

dissertation, Université Paris VII, 1994.

[17] Z. 王, K. 赵, H. 王, X. 猛, & J. Wen. Query understanding through knowledge-based conceptu-
alization. 在: Proceedings of the 24th International Joint Conference on Artificial Intelligence, AAAI, 2015,
PP. 3264–3270.

[18] 磷. 李, H. 王, K. 朱, Z. 王, & X. 吴. Computing term similarity by large probabilistic isA knowledge.
在: Proceedings of the 22nd ACM International Conference on Conference on Information & 知识
管理, ACM, 2013, PP. 1401–1410. 土井: 10.1145/2505515.2505567.

[19] D. Marneffe, 乙. MacCartney, & C.D. 曼宁. Generating typed dependency parses from phrase structure
parses. 在: Proceedings of the 5th International Conference on Language Resources and Evaluations, 2006,
PP. 449–454.

[20] D. Blei, A.Y. 的, & M.I. 约旦. Latent dirichlet allocation. Journal of Machine Learning Research 3(Jan)

(2003), 993–1022. 可用于: http://jmlr.csail.mit.edu/papers/v3/blei03a.html.

[21] 瓦. Hua, Z. 王, H. 王, K. 郑, & X. 周. Short text understanding through lexical-semantic
分析. 在: IEEE 31st International Conference on Data Engineering, IEEE, PP. 495–506. 土井: 10.1109/
ICDE.2015.7113309.

[22] J. Kupiec. Robust part-of-speech tagging using a hidden Markov model. Computer Speech & 语言

6(3) (1992), 225–242. 土井: 10.1016/0885-2308(92)90019-Z.

[23] 是. 王, H. 李, H. 王, & K.Q. 朱. Toward topic search on the web. Technical report, Microsoft Research,
2010. 可用于: https://www.microsoft.com/en-us/research/publication/toward-topic-search-on-the-web/
?from=http%3A%2F%2Fresearch.microsoft.com%2Fapps%2Fpubs%2Fdefault.aspx%3Fid%3D166386.
[24] J. 王, H. 王, Z. 王, & K.Q. 朱. Understanding tables on the web. 在: Proceedings of the 31st
International Conference on Conceptual Modeling, 施普林格, 2012, PP. 1–14. 土井: 10.1007/978-3-642-
34002-4_11.

[25] F. 王, Z. 王, Z. 李, & J. Wen. Concept-based short text classification and ranking. 在: 会议记录
the 23rd ACM International Conference on Conference on Information and Knowledge Management, ACM,
2014. 土井: 10.1145/2661829.2662067.

[26] 磷. Ferragina, & U. Scaiella. TAGME: On-the-ﬂy annotation of short text fragments (by wikipedia entities). 在:
Proceedings of the 19th ACM International Conference on Information and Knowledge Management, ACM,
2010, PP. 1625–1628. 土井: 10.1145/1871437.1871689.

[27] 是. 歌曲, H. 王, Z. 王, H. 李, & 瓦. 陈. Short text conceptualization using a probabilistic knowledge-
根据. 在: Proceedings of the 22nd International Joint Conference on Artiﬁcial Intelligence, AAAI, 2011,
PP. 2330–2336. 土井: 10.5591/978-1-57735-516-8/IJCAI11-388.

[28] D. Kim, H. 王, & A. 哦. Context-dependent conceptualization. 在: Proceedings of the 23rd International

Joint Conference on Artiﬁcial Intelligence, AAAI, 2013, PP. 2654–2661.

数据智能

268

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

作者简介

Lei Ji received her Bachelor and Master degrees from Beijing Institute of
Technology and is currently a PhD student in computer science from Institute
of Computing Technology, Chinese Academy of Sciences. She is currently
a researcher at Microsoft Research Asia. Her research interests include
knowledge graph, cross-modality visual and language learning.

Yujing Wang received her Bachelor and Master degrees from Peking
大学. She is currently a Researcher at Microsoft Research Asia. 她
research interests include AutoML, text understanding and knowledge graph.

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我

t
/

我

A
r
t
我
C
e
–
p
d

F
/

1
3
2
3
8
6
8
3
8
0
2
d
n
_
A
_
0
0
0
1
3
p
d

我

Botian Shi is currently a PhD student in computer science from Beijing
Institute of Technology, who conducted this work during internship at MSRA.
His research interests include natural language processing and multi-modal
knowledge based image/video understanding.

乙
y
G
你
e
s
t

哦
n
0
7
S
e
p
e
米
乙
e
r
2
0
2
3

269

数据智能

Microsoft Concept Graph: Mining Semantic Concepts for Short Text Understanding

Dawei Zhang received his Master degree from Peking University. He is
currently the Founder of MIX Labs. His research interests include machine
学习, natural language processing and knowledge graph.

Zhongyuan Wang received his Bachelor, Master, and PhD degrees in
computer science at Renmin University of China in 2007, 2010, 和 2015,
分别. He is currently a Senior Researcher and Senior Director of
Meituan-Dianping. His research interests include knowledge base, Web data
矿业, online advertising, machine learning and natural language processing.

Jun Yan received his PhD degree in school of mathematical science at Peking
大学. He is currently the Chief AI Scientist of Yiducloud. His research
interests include medical AI research, knowledge graph mining and online
advertising.

数据智能

270

我

D
哦
w
n
哦
A
d
e
d

F
r
哦
米
H

t
t

:
/
/

d
我
r
e
C
t
.

米

我
t
.

e
d
你
d
n

我