Set Expander: A Knowledge-based System for Entity Set Expansion
Weronika T. Adrian
a
and Paweł Caryk
AGH University of Science and Technology, al. A. Mickiewicza 30, 30-059 Krakow, Poland
Keywords:
Entity Set Expansion, Knowledge Graphs, Semantics, Knowledge-based Systems, Entity Recommendation.
Abstract:
Entity Set Expansion (ESE) is a problem that underlies several important tasks, such as entity recommendation,
query expansion, synonymy finding etc. Traditional strategies relied on corpus-based methods to recognize
the intended category of the input words. But with the growing importance and visibility of knowledge graphs,
new methods based on explicit knowledge representation have been put forward to solve the ESE problem. In
this paper, we review the existing knowledge-based methods for entity set expansion and introduce a new on-
line tool called Set Expander that uses semantic technologies and knowledge bases to solve the ESE problem
efficiently. We present the algorithms and implemented techniques that ensure responsiveness and effective-
ness of the tool. We analyze the strengths and weaknesses of the proposed solution and envision the future
research directions.
1 INTRODUCTION
Entity Set Expansion (ESE) problem consists in find-
ing entities, words or terms “similar” to the ones given
as input. The problem implicitly assumes that one has
to first understand what the input entities have in com-
mon, what they are, to what category they belong or
what features they share; and then to retrieve, given
some knowledge sources, more things “of the same
kind” (Sarmento et al., 2007). The input terms are
called the seeds, as we want to “grow” a bigger class,
a set of entities from them. There are multiple do-
mains of application of the ESE problem: from en-
tity suggestion in recommendation engines, to user-
guided dictionary creation (Kohita et al., 2020).
“Traditional” approaches to entity set expansion
relied on big textual corpus, within which the algo-
rithms identified the patterns in which the seeds ap-
peared, and matched those patterns with the corpus
to find new entities. This kind of approach is called
bootstrapping, because we start from a small set of
seeds and iteratively learn new patterns and words that
by generating new patterns guide the algorithms
further. These algorithm were prone to so-called
“concept drift” where mistakes at some point accu-
mulated over time and several ways to mitigate these
deviations were proposed. Apart from bootstrapping,
several methods of learning features from text (e.g.,
news) corpus (Zhang and Liu, 2011), Web (Wang and
a
https://orcid.org/0000-0002-1860-6989
Cohen, 2008; Hu and Jia, 2015) or social media me-
dia (Zhao et al., 2017) have been proposed.
With the renewed interest and availability of large
structured knowledge bases, also new knowledge-
based methods for ESE have been proposed. Instead
of using textual patterns or co-occurrence statistics of
terms in textual corpus, these new methods use struc-
tured knowledge bases to inform the category recog-
nition and entity retrieval. In this paper, we review
the newest approaches to Entity Set Expansion and
present a new tool called Set Expander that commu-
nicates online with integrated knowledge graph Ba-
belNet (Navigli and Ponzetto, 2012) and reason over
the fetched knowledge.
2 A REVIEW OF KNOWLEDGE
BASED APPROACHES TO
ENTITY SET EXPANSION
2.1 Knowledge-based Methods for ESE
Knowledge graphs (Yan et al., 2018) are ontolog-
ical resources that describe the universe of inter-
est in terms of instances, classes and relationships
among them. With the increasing interest in knowl-
edge graphs, also new methods for ESE have been
put forward. Zhang et al. (Zhang et al., 2017) pro-
posed to use so-called semantic features which re-
494
Adrian, W. and Car yk, P.
Set Expander : A Knowledge-based System for Entity Set Expansion.
DOI: 10.5220/0010715100003058
In Proceedings of the 17th International Conference on Web Information Systems and Technologies (WEBIST 2021), pages 494-500
ISBN: 978-989-758-536-4; ISSN: 2184-3252
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reser ved
flects the relations among the nodes in a graph. Zheng
et al. (Zheng et al., 2017) calculate the relatedness of
terms relying on the or meta-paths that reflect both
categories and relations in which the terms appear.
Semantic features of knowledge graphs are also the
core of the method proposed by Chen et al. (Chen
et al., 2018) A method that mitigates the noise in
knowledge graphs has been proposed by Rastogi et
al in (Rastogi et al., 2019).
2.2 Multifaceted Set Expansion
The input words may have multiple meanings and
therefore several ways of expansion should be con-
sidered for a set of seeds. Rong et al. (Rong et al.,
2016) addressed the problem of multifaceted set ex-
pansion, i.e., set expansion that takes into consider-
ation different possible categories of expansion. The
authors used skip-grams to learn “sibling” entities of
the input terms and create so-called ego-networks. In
these networks, clustering of sibling entities points
to possible expansion directions. The algorithm fuse
the networks with user-created ontologies, such as
Wikipedia and word embeddings i.e., numerical vec-
tor representations of words.
Using online semantic resources was also pro-
posed by Adrian et al. in (Adrian and Manna, 2018).
The authors used logic programming approach to find
the optimal common ancestors of the input terms,
based on network structure that contains information
integrated from several knowledge bases. They used
the semantic query language SPARQL (P
´
erez et al.,
2009) to fetch the needed information, modelled the
obtained knowledge as a graph and give it to the rea-
soner as an input of an optimization problem. A tool
written in declarative ASP (Adrian et al., 2018) iden-
tifies, in the resulting network, clusters of meanings of
the input words that serve as starting points for multi-
faceted set expansion.
Recently, the multi-faceted set expansion has also
been addressed in (Zhu et al., 2019), where the au-
thors propose a new framework called FUSE. The
framework works in three phases, first identifying
clusters that correspond to different facets, then op-
timizing the clusters by joining shared conceptualiza-
tions, and finally expanding the facets using a pre-
trained models.
2.3 Recent Hybrid Proposals
The problem seems to enjoy a renewed interest in
the recent years with multiply hybrid proposals. A
joint model for learning Entity Set Expansion and at-
tribute extraction has been proposed in (Zhang et al.,
2016), A system based on lexical features and dis-
tributed representations has been proposed in (Yu
et al., 2019). Several solutions using language prob-
ing (Zhang et al., 2020), pattern rank (Xiao et al.,
2020) or enriching a corpus-based approach with so-
called auxiliary sets and co-expansion (Huang et al.,
2020) have been put forward. Finally, a bootstrapping
system that relies on a neural network has been pre-
sented in (Yan et al., 2020).
2.4 Available Tools
Several years ago, the ESE functionality was available
in the Google spreadsheet software. Unfortunately,
this feature has been discontinued and the technol-
ogy has ever since been protected by a patent. Thus,
several attempts to put online an ESE tool have been
undertaken. WordGrabBag.com
1
is an online tool
that let users enter a set of words and return a list
of words “of the same kind”. It is based on the
Word2Vec (Mikolov et al., 2013) method and neural
language models to represent data from Wikipedia.
The tool uses an optimized search algorithm to ex-
tract the new words relevant to the input query (more
details can be found here: https://code.google.com/
archive/p/word2vec/). SetExpan (Shen et al., 2017)
is a corpus based tool that served as a base for a
system for interactive dictionary generation (Kohita
et al., 2020).
3 SET EXPANDER SYSTEM
DESCRIPTION
3.1 General Assumptions
We assume an ontological knowledge base i.e., one
that contains information about instances, classes and
relationships. For clarity, we denote abstract concepts
as “classes”, real world objects as “instances” and we
will refer to any of them as “entities” (C I = E,
where C is a set of concepts, I is a set of instances
and E is a set of entities).
As for the ESE problem, we assume we get n
words as input; each word may have m meanings
(senses); each sense points to an instance in a knowl-
edge base; each instance may (and usually do) be-
long to k
0
(immediate) categories, that in turn are sub-
classes of other k
1
categories that are sub-classes of
further k
2
categories etc.
We divide the problem into two sub-problems:
1
See http://wordgrabbag.com/.
Set Expander: A Knowledge-based System for Entity Set Expansion
495
1. to recognize the common category of the entities
represented by the input words, and
2. to retrieve more relevant entities based on the
identified category.
3.2 Technologies and Knowledge
Resources
We have decided to use BabelNet (Navigli and
Ponzetto, 2012) as a primary knowledge source. The
BabelNet is a multilingual semantic encyclopedia,
semi-automatically created from a number of es-
tablished information resources such as Wikipedia,
WordNet, Wikidata, OmegaWiki or GeoNames. Con-
sequently, it integrates a paramount amount of knowl-
edge about abstract concepts and real-world objects
(so-called Named Entities), collecting their attributes
and relationships with other entities.
BabelNet provided several APIs. After carefully
considering all of them, we have decided to use the
SPARQL interface to fetch the needed data. The lan-
guage allows to formulate complex queries that we
generate within the program.
3.3 Main Algorithms for ESE
To solve the first problem (see Sect. 3.1), we gather
structured knowledge from our knowledge base and
return a graph-based representation. We start with
querying BabelNet with a SPARQL query built for
the input words. For each word the query:
1. asks for all the possible senses (pointing to par-
ticular instances in BabelNet) for the given word,
and
2. asks for the classes and their super-classes up to a
required depth (from 1 to 5).
Using a single query we can fetch both the senses of
the input words and the hierarchy of categories of the
senses. On Listing 1, one can see an example query
generated for a word “Java” and hierarchy depth 4:
SELECT DISTINCT ?A ?B WHERE {
? e n t r i e s a lemon : L e x i c a l E n t r y .
? e n t r i e s lemon : s e n s e ? s e n s e .
? s e n s e lemon : r e f e r e n c e ? s y n s e t .
? s y n s e t a s k o s : Co nc e pt .
? e n t r i e s r d f s : l a b e l ? l a b e l .
? s y n s e t bnlemon : s y n s e t T y p e ? sy ns et T y p e .
? s y n s e t s k o s : b r o a d e r ?X1 .
?X1 s k o s : b r o a d e r ?X2 .
?X2 s k o s : b r o a d e r ?X3 .
?X3 s k o s : b r o a d e r ?X4 .
FILTER (
? l a b e l = J av a ”@en | | ? l a b e l =” j a v a ”@en
| | ? l a b e l =JAVA”@en
) .
FILTER (
? s y n s e t T y pe =NE
) .
{ ?A r d f s : l a b e l ? l a b e l .
? s y n s e t bnlemon : s y n s e t I D ?B }
UNION
{ ? s y n s e t bnlemon : s y n s e t I D ?A .
?X1 bnlemon : s y n s e t I D ?B }
UNION
{ ?X1 bnlemon : s y n s e t I D ?A .
?X2 bnlemon : s y n s e t I D ?B }
UNION
{ ?X2 bnlemon : s y n s e t I D ?A .
?X3 bnlemon : s y n s e t I D ?B }
UNION
{ ?X3 bnlemon : s y n s e t I D ?A .
?X4 bnlemon : s y n s e t I D ?B }
}
Listing 1: Getting senses and categories from BabelNet.
The results of the query are organized into graphs
(one for a sense of the input term) in which the nodes
are the entities and the edges denote the semantic re-
lations among them.
Then, we identify the most specific common cate-
gory, taken into consideration all the combinations of
the graphs created for each sense for all the words. As
an example, consider two words w1 and w2. If we de-
note snsXY as a graph of categories for sense Y from
word X, then from the following set of graphs:
[[sns11, sns12, sns13], [sns21, sns22, sns23]]
we get the following combinations to analyze:
[sns11, sns21], [sns11, sns22], [sns11, sns23],
[sns12, sns21], [sns12, sns22], [sns12, sns23],
[sns13, sns21], [sns13, sns22], [sns13, sns23]
In graph theory, the problem is referred to as finding
the lowest common ancestor and in Description Log-
ics (Baader et al., 2004): finding the least common
subsumer(s) for most specific concepts of the given in-
stances. Our algorithm finds the common nodes in the
combinations and return the one closest to the senses
(see Fig. 1).
It may happen that there are more than one cate-
gory equally distant from the senses. In this case, we
keep all of these categories as possible users’ intents
and expand all of them in the next step.
Then to retrieve more relevant entities, we per-
form the following query:
SELECT DISTINCT ? e x p and ? e n t r y
c o u n t ( ? r e l a t e d ) a s ? co u n t
WEBIST 2021 - 17th International Conference on Web Information Systems and Technologies
496
WHERE {
? e xpan d s k os : b r o a d e r
< h t t p : / / b a b e l n e t . o rg / r d f / { C at eg o ry I D }> .
? e xpan d s k os : e xa c t Ma tc h ? e n t r y .
? e xpan d bnlemon : sy n s e tT y p e NE” .
? e xpan d s k os : r e l a t e d ? r e l a t e d
FILTER (
s t r s t a r t s ( s t r ( ? e n t r y ) ,
” h t t p : / / d b p ed i a . or g / r e s o u r c e / ” )
)
] }
GROUP BY ? ex p and ? e n t r y
ORDER BY DESC( ? c o un t )
LIMIT 10
Listing 2: Getting relevant entities from BabelNet.
where:
?expand in query denotes ID of searching entities
?entry is link reference to entry in DB resources,
from which name is extracted
?related denotes related ids to founded entity
The candidate entities must fulfill some constraints:
they need to have the required category in “broader”’
relations and only “NE” (named entities) are desir-
able. To rank the candidate entities, for each ex-
panded category only the 10 most related entities are
returned.
...
s
1
s
2
s
n
Lowest
Common
Category
1
Lowest
Common
Category
2
(a)
(b)
(c)
Figure 1: Identifying the lowest common categories within
combinations of graphs for (a) input words, that have (b)
different senses that refer to instances in BabelNet and that
belong to (c) classes that form a hypernymy hierarchy.
4 RESULTS AND DISCUSSION
4.1 Implementation Principles
We have implemented the SetExpander tool in
Python, using Django framework. With respect to the
first (unpublished) version of the tool, some improve-
ments have been implemented:
Calls to Babelnet API to retrieve the entities of a
given categories are executed in a parallel loop,
using Pool from multiprocessing library.
In the prototype version of SetExpander querying
of word synsets was divided into two parts: first,
all senses of the given words were found, then for
each sense, we searched for categories where it
belongs. Now one query is performed for senses
and their categories. Additionally, for each sense
a graph of categories is built, with a depth from 1
to 5.
The query to find the entities of the identifieds cat-
egories searches for ids and names. To improve
the quality of the returned data and limit the re-
sponse size a constraint to the query was added,
the result are ordered by how many related rela-
tionships they have and then limited to 10.
The system code repository is available online.
2
The user interface is simple and consists of a search
input field in which the user should enter the input
words (see Fig. 2). Once the input words are entered
Figure 2: Start screen of the Set Expander tool.
the tool calculates their common category or cate-
gories and return the results grouped by the category
names (see Fig. 3).
4.2 Experiments and Conclusion
We have performed tests on the tool to check its
behaviour, strengths and limits. Exemplary results
for sets of various input words are presented in Ta-
ble 1. In general, the response time of the applica-
tion is satisfactory; for two words with four levels of
graph depth, execution of find common categories
takes approximately 10 milliseconds, for three words
it takes approx. 100 ms and for four words
2
See https://anonymous.4open.science/r/
e547777c-7d30-46c6-b988-36c82b3c05cc/.
Set Expander: A Knowledge-based System for Entity Set Expansion
497
Table 1: Expansion of example input set of words within identified categories.
Seeds Category Expansion
Rome,
Budapest,
Vienna
national capital Prague, Belgrade, Pretoria, Sao paulo, Paris, Madrid, Rome, Berlin, Warsaw, Kiev
City New york city, Frankfurt, Pretoria, Paris, Naples, Madrid, Berlin, Saint petersburg,
work Alice’s adventures in wonderland, Don quixote, Asterix, Family guy, Alicia alonso,
Iliad, Tarzan, A christmas carol, The count of monte cristo, Dracula
album U-vox, Brilliant (album), 20 years of jethro tull, Systems of romance,
Rage in eden, Visage (visage album), Metamatic, Blow up your video,
audio recording Extended play, Hiroshima – rising from the abyss
film Back to the future, The good, the bad and the ugly, Warner bros, Casino royale,
New hollywood, E.t. the extra-terrestrial, Titanic (1997 film), Kevin bacon, Jaws
Einstein,
Newton,
Galileo
human John Paul II, Christ, Leonardo da vinci, Sigmund Freud, Gottfried Wilhelm Leibniz,
Auguste Comte, American, Karl Marx, Albert Einstein, Mason
taxon Epsilon 15
name Lexus, Sara, Canute, American forces network, Eirik, Peter (given name),
Robert the bruce, Gregory (given name), Adam (given name)
Neptune,
Pluto,
Saturn
planet Pluto, Mars, Jupiter, Mercury, Venus, Earth, Saturn, Uranus, Neptune,
superior planet Pluto, Mars, Jupiter, Uranus, Neptune
deity Artemis, Dionysus, Aphrodite, Kore, Apollo, Hades, Devil, Poseidon, God, Athena
spiritual being Devil
musical group One direction, The byrds, Coldplay, S club 7, Ramones, Jls, Bullet for my valentine,
Opeth, Nine inch nails, Underoath
music Confederate states of america, Film score, Balkan music, Timeline of musical events,
Wolfgang amadeus mozart, West side story, Emas, Karlheinz stockhausen, Last.fm
musical work Nocturnes, op. 9 (chopin), Elijah (oratorio), In the hall of the mountain king,
Appalachian spring, Symphonic dances (rachmaninoff), Masquerade (khachaturian),
The art of fugue, Capriccio espagnol, Suite espa
˜
nola no. 1, op. 47
composition Symphony no. 1 (mahler), Das lied von der erde, Symphony no. 4 (mahler),
The tales of hoffmann, Symphony no. 2 (mahler), Symphony no. 3 (mahler),
Book of lamentations, Paolo conte, Music of star wars, Symphony no. 8 (mahler)
album Biophilia (album), Grammy award for album of the year, Post (bj
¨
ork album), Vespertine,
The velvet rope, Debut (bj
¨
ork album), Extended play, Octavarium (album), Ray of light
individual Krypto, God the son, Zapatista army of national liberation, Lapsed catholic
band The byrds, Coldplay, Ramones, Underoath, Bullet for my valentine, Opeth,
Nine inch nails, Billy talent, White lies
audio recording Hiroshima – rising from the abyss, Extended play
around 1.1s. Implementing parallelism significantly
improved the response time that mainly depends on
communication with BabelNet.
We discovered that increasing the depth of cate-
gory hierarchy querying does not necessarily improve
the quality of results. In fact, the quality depends
strongly on the domain of the query and while for
categories such as science (e.g., names of the plan-
ets) or geography (e.g., names of cities) it is sufficient
to reach to depth 2, when it comes to people, even fa-
mous, we had to reach to depth 3 to get meaningful
answers.
We could also observe that the richness of Babel-
Net comes with some noisiness and so for future work
we plan to extend our research in two directions:
1. to analyze more relations, not only hypernymy
(subclass-superclass relations) and to include
them into the common category definition pro-
cess, e.g., as in (Yang and Powers, 2005), where
meronymy i.e., “part-of relations are included
into the definition of semantic similarity, and
WEBIST 2021 - 17th International Conference on Web Information Systems and Technologies
498
Figure 3: Results of the Set Expander tool for input: “Ein-
stein, Newton, Bach”. As one could expect, the common
category of this input is a general “last name”, as these peo-
ple did not share a profession that would point to a more
specific category. Surprisingly, there are also “craters” that
are named after the entered names, and so more caters have
been returned as a possible expansion.
2. to post-process the results of the expansion step
to filter out some entities that are less common,
have less “evidence” of their quality (currently,
we have taken a step in this direction by ordering
and selecting k first candidate entities by the num-
ber of “related” relationships between the candi-
dates and the seeds).
5 SUMMARY
In this paper, we reviewed the existing approaches to
the ESE problem, with a special focus on the most
recent ones. We have put forward a proposal of a
new knowledge-based system called Set Expander
that builds on semantic technologies and knowledge
graph resources. The system is available online and
upon configuration works in an interactive mode. The
tool described in this paper is flexible and extendable
and we plan to continue working on its improvements.
REFERENCES
Adrian, W. T., Alviano, M., Calimeri, F., Cuteri, B., Do-
daro, C., Faber, W., Fusc
`
a, D., Leone, N., Manna,
M., Perri, S., et al. (2018). The asp system dlv: ad-
vancements and applications. KI-K
¨
unstliche Intelli-
genz, 32(2):177–179.
Adrian, W. T. and Manna, M. (2018). Navigating online
semantic resources for entity set expansion. In Proc.
of PADL’18, pages 170–185.
Baader, F., Horrocks, I., and Sattler, U. (2004). Descrip-
tion logics. In Handbook on ontologies, pages 3–28.
Springer.
Chen, J., Chen, Y., Zhang, X., Du, X., Wang, K., and Wen,
J.-R. (2018). Entity set expansion with semantic fea-
tures of knowledge graphs. Journal of Web Semantics,
52:33–44.
Hu, W. and Jia, C. (2015). A bootstrapping approach to
entity linkage on the semantic web. Journal of Web
Semantics, 34:1–12.
Huang, J., Xie, Y., Meng, Y., Shen, J., Zhang, Y., and Han,
J. (2020). Guiding corpus-based set expansion by aux-
iliary sets generation and co-expansion. In Proceed-
ings of The Web Conference 2020, pages 2188–2198.
Kohita, R., Yoshida, I., Kanayama, H., and Nasukawa, T.
(2020). Interactive construction of user-centric dic-
tionary for text analytics. In Proceedings of the 58th
Annual Meeting of the Association for Computational
Linguistics, pages 789–799.
Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., and
Dean, J. (2013). Distributed representations of words
and phrases and their compositionality. In Advances in
neural information processing systems, pages 3111–
3119.
Navigli, R. and Ponzetto, S. P. (2012). Babelnet: The au-
tomatic construction, evaluation and application of a
wide-coverage multilingual semantic network. Artifi-
cial Intelligence, 193:217–250.
P
´
erez, J., Arenas, M., and Gutierrez, C. (2009). Seman-
tics and complexity of sparql. ACM Transactions on
Database Systems (TODS), 34(3):1–45.
Rastogi, P., Poliak, A., Lyzinski, V., and Van Durme, B.
(2019). Neural variational entity set expansion for au-
tomatically populated knowledge graphs. Information
Retrieval Journal, 22(3-4):232–255.
Rong, X., Chen, Z., Mei, Q., and Adar, E. (2016). Egoset:
Exploiting word ego-networks and user-generated on-
tology for multifaceted set expansion. In Proceedings
of the Ninth ACM international conference on Web
search and data mining, pages 645–654.
Sarmento, L., Jijkuon, V., De Rijke, M., and Oliveira, E.
(2007). more like these” growing entity classes from
seeds. In Proceedings of the sixteenth ACM confer-
ence on Conference on information and knowledge
management, pages 959–962.
Shen, J., Wu, Z., Lei, D., Shang, J., Ren, X., and Han,
J. (2017). Setexpan: Corpus-based set expansion
via context feature selection and rank ensemble. In
Joint European Conference on Machine Learning and
Knowledge Discovery in Databases, pages 288–304.
Springer.
Wang, R. C. and Cohen, W. W. (2008). Iterative set expan-
sion of named entities using the web. In 2008 eighth
IEEE international conference on data mining, pages
1091–1096. IEEE.
Xiao, Z., Li, C., and Chen, H. (2020). Patternrank+ nn: A
ranking framework bringing user behaviors into entity
set expansion from web search queries. ACM Trans-
actions on the Web (TWEB), 14(3):1–15.
Yan, J., Wang, C., Cheng, W., Gao, M., and Zhou, A.
(2018). A retrospective of knowledge graphs. Fron-
tiers of Computer Science, 12(1):55–74.
Set Expander: A Knowledge-based System for Entity Set Expansion
499
Yan, L., Han, X., He, B., and Sun, L. (2020). End-to-end
bootstrapping neural network for entity set expansion.
In Proceedings of the AAAI Conference on Artificial
Intelligence, volume 34, pages 9402–9409.
Yang, D. and Powers, D. M. (2005). Measuring seman-
tic similarity in the taxonomy of WordNet. Australian
Computer Society.
Yu, P., Huang, Z., Rahimi, R., and Allan, J. (2019).
Corpus-based set expansion with lexical features and
distributed representations. In Proceedings of the
42nd International ACM SIGIR Conference on Re-
search and Development in Information Retrieval,
pages 1153–1156.
Zhang, L. and Liu, B. (2011). Entity set expansion in opin-
ion documents. In Proceedings of the 22nd ACM con-
ference on Hypertext and hypermedia, pages 281–290.
Zhang, X., Chen, Y., Chen, J., Du, X., Wang, K., and
Wen, J.-R. (2017). Entity set expansion via knowledge
graphs. In Proceedings of the 40th International ACM
SIGIR Conference on Research and Development in
Information Retrieval, pages 1101–1104.
Zhang, Y., Shen, J., Shang, J., and Han, J. (2020). Em-
power entity set expansion via language model prob-
ing. arXiv preprint arXiv:2004.13897.
Zhang, Z., Sun, L., and Han, X. (2016). A joint model
for entity set expansion and attribute extraction from
web search queries. In Thirtieth AAAI Conference on
Artificial Intelligence.
Zhao, H., Feng, C., Luo, Z., and Pei, Y. (2017). Entity
set expansion on social media: A study for newly-
presented entity classes. In Chinese National Con-
ference on Social Media Processing, pages 116–128.
Springer.
Zheng, Y., Shi, C., Cao, X., Li, X., and Wu, B. (2017).
Entity set expansion with meta path in knowledge
graph. In Pacific-Asia conference on knowledge dis-
covery and data mining, pages 317–329. Springer.
Zhu, W., Gong, H., Shen, J., Zhang, C., Shang, J., Bhat, S.,
and Han, J. (2019). Fuse: Multi-faceted set expansion
by coherent clustering of skip-grams. arXiv preprint
arXiv:1910.04345.
WEBIST 2021 - 17th International Conference on Web Information Systems and Technologies
500