The paper is devoted to the problem of the bibliometric study of publications on the topic “Cross-lingual Semantic Similarity”, available in the Dimensions database. Visualization of scientific networks showed fragmentation of research, limited interaction of organizations. Leading countries, leading organizations and authors are highlighted. Overlay visualization allowed us to assess the trends in citing authors. The expansion of the geography of research is shown. For international cooperation, the uniformity of semantic approaches to describing the concepts of critical infrastructure, incidents, resources and services related to their maintenance and protection is important. The stated approaches can be applied for visualization and modeling of technological development in the modern digital world. Semantic similarity is a longstanding problem in natural language processing (NLP). The semantic similarity between two words represents the semantic proximity (or semantic distance) between two words or concepts. This is an important problem in natural language processing, as it plays an important role in finding information, extracting information, text mining, web mining and many other applications.
Linguistic similarities were studied by researchers from different fields using numerous statistical, linguistic and neuroscientific approaches.
The semantic properties of languages are usually evaluated using the embedding of words, which projects a linguistic dictionary onto the vector space of a given number of dimensions, in which the semantic relations of words are stored.
In artificial intelligence and cognitive science,
semantic similarities were used for various scientific
assessments and measurements, as well as for decoding
complex interfaces of conceptualizing feelings [
Theoretically, semantic similarity refers to the idea of commonality in the characteristics between words or concepts in a language. Although this is a property of the relationship between concepts or feelings, it can also be defined as a measurement of the conceptual similarity between two words, sentences, paragraphs, documents, or even two parts of a text.
Recently, there has been a growing interest in finding
semantically similar words in different languages based
on comparable data easily accessible from the Internet
(for example, Wikipedia, news) [
According to Hotho et al. [
Text analysis in big data analytics is becoming a
powerful tool for processing unstructured text data,
analyze it to extract new knowledge and identify
meaningful models and correlations hidden in the data.
Text mining refers to the extraction of information and
implicit patterns previously unknown in automatic or
semi-automatic mode from a huge unstructured text data
such as natural language texts [
Tech Mining refers to the application of text mining
methods to technical documentation. For the purposes of
patent analysis, this is called “patent mining”. Tech
Mining (TM) [
Whereas many applications have employed certain
similarity functions to compute the semantic similarity
between terms, most of the traditional approaches solving
the problem by using dictionaries such as WordNet. The
main problem is that a lot of terms (e.g. abbreviations,
acronyms, brand names etc.) that are not covered by these
kinds of dictionaries [
Tech Mining is the application of text mining tools to scientific and technical information resources. The evergrowing volume of scientific results represents a boom in technological innovation, but also complicates efforts to obtain useful and concise information for solving problems. This problem extends to technological mining, where the development of methods compatible with big data is an urgent problem.
In the current patent analysis, numerous patent documents use different words to describe the same event, leading to semantic inconsistency and polysemy due to the many meanings that may exist for a single word. To solve this problem, document analysis often requires combining synonyms into the same semantic dimension. On the other hand, different words can be used to describe the same events.
The methods for measuring the semantic similarity of
texts are necessary for the development of areas of
information retrieval, data mining and text analysis. Such
methods will help to avoid patent infringement in the
development of technological capabilities to achieve
future competitive advantages [
The growing popularity of data science is also affecting high-tech industries. However, since they usually have different core competencies - the creation of cyberphysical systems, and not, for example, machine learning algorithms or data mining - to delve into the science of data by specialists in the field, such as system engineers or architects, can be more cumbersome than expected.
In recent years, in order to help subject matter experts
use data science, scientists have been developing semantic
search engines. So, for example, Semantic Snake Charmer
(SSC) [
An accurate assessment of the actual similarity
between documents is fundamental for many automatic
text analysis applications, such as thesaurus generation
[
Semantic space is an attempt to model the
characteristics of human semantic memory, which is
guided by the principle that words with similar meanings
are found in a similar language environment. Semantic
space is a vector space that captures the value
quantitatively from the point of view of coincidence
statistics, where words (or concepts) are represented as
vectors in a high-dimensional space [
Latent semantic analysis (LSA) is based on the fact
that words that have similar meanings tend to occur in
similar texts [
Knowledge-based methods suffer from a limited number of common vocabulary words that are commonly used in general English literature and often not suitable for specific domains.
The vector space model is classically used to evaluate
the semantic similarity between two documents. Terms
are represented in this semantic space as vectors called
word embeddings. The possibilities of determining textual
similarity based on vector representations of terms in a
semantic space in which the proximity of vectors can be
interpreted as semantic similarity [
The LSA method has an advantage over most modern
information retrieval methods because it has the ability to
measure the similarity of two texts that use completely
different words. However, there are morphological
problems of the correct identification of terms, as well as
more fundamental problems with homonymy / polysemy
and synonymy. Techniques that depend on large
enclosures tend to overestimate relatively unrelated
sentences or relatively related sentences (e.g., LSAs).
LSAs overestimate the similarity score of compared pairs
of sentences [
Semantic spaces can be constructed either using the additive model or the multiplicative model. Both additive and multiplicative approaches to constructing semantic space do not take into account the word order among the components (i.e., words or phrases). Traditional clustering algorithms usually rely on the BOW (Bag of Words) approach, and the obvious drawback of BOW is that it ignores the semantic relationship between words.
Researchers expanded DSM to include the
compositional structure of the language, and called these
models compositional-DSM (CDSM). CDSM models
suggest that the meaning of a word can be interpreted by
its context, and the meaning of a sentence can be obtained
from its compositions [
Assessing semantic similarities between concepts is a
key tool to improve understanding of texts. The structured
knowledge provided by ontologies is widely used to
evaluate similarities. However, in many areas several
ontologies modeling the same concepts in different ways
are available. The paper describes the criteria for choosing
ontologies for assessing semantic similarity [
A measure of calculating the similarity between
sentences or between documents using an ontology is
proposed. The similarity is evaluated using the concept
vector of the document (proposal), formed by finding the
links between the ontology terms and the content of the
document (proposal) [
The vector space model is used to identify potentially
useful services and evaluate web services [
Similarity measures are used to select a contextsensitive application that matches the current context of the user. Personalization of services is directly related to the user's preferences, displaying his contextual information from the user environment.
A semantic similarity measure is a tool for assessing
the similarity between instances of the context, which
allows to select services in accordance with their
relevance for a given request, profile and user preferences.
With this approach, the context is considered as a set of
information representing spatio-temporal information
about the user, as well as his preferences and interests,
which is used as a factor in classifying services by
relevance [
The data sets of common STS problems were widely
used to study similarities at the sentence level and
semantic representations [
The CL-WES method [
The similarity score of the interlanguage pairs in
English and Spanish was calculated as the average of the
corresponding language ratings in the monolingual data
sets [
The skip-gram model has become one of the most
popular for the study of word representations in NLP [
The cross-language definition of semantic textual similarity is an important step for the detection and evaluation of interlanguage plagiarism; research in this area is rare.
A comparable corpus consists of documents in two or more languages or varieties that are not translations of each other and deal with similar topics. Comparable bodies are, by definition, multilingual and interlanguage collections of text. The Internet can be used as a huge resource of multilingual texts.
To search for publications, the Dimensions database (https://app.dimensions.ai/) was used, which provides open access to more than 95 million publication records and related metrics for individual users. The search keywords used were “cross-lingual semantic similarity”. 2050 articles were discovered.
VOSviewer (https://www.vosviewer.com/) was used to visualize scientific networks. VOSviewer uses a remote approach to visualizing bibliometric networks. In a bibliometric network, there are often large differences between nodes in the number of edges they have to other nodes.
Popular sites, for example, representing highly cited
publications or highly prolific researchers, may have
several orders of magnitude more connections than their
less popular counterparts. When analyzing bibliometric
networks, normalization of these differences between
nodes is usually performed. VOSviewer by default applies
the normalization of communication strength [
2050 articles of 2825 authors from 64 countries were discovered. The dynamics of publications is shown in Fig. 1. The trend line is clearly exponential, the determination coefficient (R2), which is also called the approximation confidence value, is 0.6648. Initial publications date back to the 80s of the 20th century, but research has been growing since the beginning of the 21st century.
We reviewed a collaborative network of organizations (Fig. 3). For 684 organizations, the minimum number of articles of the organization was taken to be five; such organizations were allocated 64 in 6 clusters. Fig. 3 shows that only a small number of universities interact. The largest cluster included 11 European universities and organizations: Dublin City University; Fondazione Bruno Kessler; German Research Center for Artificial Intelligence; National University of Distance Education; Trinity College Dublin; University of Alicante; University of Edinburgh; University of Sheffield; University of The Basque Country; University of Trento; University of Wolverhampton.
We examined a co-authorship network by country, the minimum number of articles by the author was taken to be five. Of 2825 authors of 64 countries, 35 are associated in five clusters (Fig. 4). The two largest clusters included 9 countries. The first cluster included countries: Austria, Canada, China, India, Iran, Japan, the Netherlands, Taiwan, and the USA. The second cluster included countries: Belgium, Finland, France, Greece, Slovenia, Spain, Switzerland, Tunisia, Great Britain.
The citation index in recent years is the main measure of the value of both a scientist and an institution, so we examined citation networks.
We examined the citation network for documents, the
minimum number of publications by the author was taken
equal to ten. 298 authors from 2050 were identified in 14
clusters (Fig. 5).
The most cited author is Navigli, Roberto (759
citations) [
VOSviewer also supports overlay renderings. In overlay rendering, the color of a node indicates a specific property of the node, for example, the year of publication. We presented the authors citation network in an overlay visualization option to assess citation trends (Fig. 6). The figure clearly shows that R. Navigli is the founder in the area.
A citation network by authors was built. The minimum number of publications by the author was taken to be five. 16 authors were identified from 2050 in 2 clusters (Fig. 7). Mittal Namitali (2017), Rettinger Achim, Gipp Bela, Li Juanzi and Zhan Lei (2016) are the most recent of the most cited authors. The geographical aspects of citation were considered. A citation network for countries was built with a minimum of five publications. 34 countries were identified (Fig. 8). It is seen that the geography of research is expanding. So, in the last goals, Brazil, Czech Republic, Iran, Egypt, Tunisia have joined the research.
A bibliometric study of publications on the topic “Cross-lingual Semantic Similarity”, available in the Dimensions database, was carried out. In recent years, there has been a significant increase in research. included 11 European universities and organizations from Ireland, Italy, Germany, Spain, Scotland, and Great Britain.
Visualization of the co-authorship network by country showed that 35 countries interact in research, countries are connected in five clusters. The two largest clusters included 9 countries. In the largest clusters, including 9 countries, the leading ones were the USA and China, Great Britain and Spain.
The visualization of the citation network revealed 298
of the most cited authors out of 2050. The most cited
author is Navigli, Roberto (759 citations). More than 200
citations from Rosso, Paolo (239) and Moens,
MarieFrancine (216) [
Overlay visualization made it possible to evaluate the
citation trends of the authors; it turned out that the most
cited author, Navigli, Roberto, is also the founder of
research in this field [
The most recent cited authors are Mittal Namitali (2017 citation), Rettinger Achim, Gipp Bela, Li Juanzi and Zhan Lei (2016 citation).
Consideration of the geographical aspects of citation showed an expansion of the geography of research. So, in the last goals, Brazil, Czech Republic, Iran, Egypt, Tunisia have joined the research.
The reported study was funded by RFBR according to the research projects № 18-07-00225, 18-07-00909, 1807-01111 and 20-04-60185.