Text Type Differentiation Based on the Structural Properties of Language Networks

Martinčić-Ipšić, Sanda; Miličić, Tanja; Meštrović, Ana

doi:10.1007/978-3-319-46254-7_43

Cited by 6 publications

(3 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note however, that this comes at the cost of the high computation complexity of the procedures for calculating the properties. The GOW model in its diverse variants has been applied to many natural language processing tasks, including text summarization (Antiqueira et al, 2009), keyword extraction (Beliga et al, 2015(Beliga et al, , 2016, text genre detection (Grabska-Gradzińska et al, 2012;Martinčić-Ipšić et al, 2016b) and document classification (Blanco & Lioma, 2012;Hassan et al, 2007;Malliaros & Skianis, 2015;Nguyen et al, 2016;Papadakis et al, 2016;Rossi et al, 2012;Rousseau et al, 2015).…”

Section: Network Based Modelsmentioning

confidence: 99%

The Influence of Feature Representation of Text on the Performance of Document Classification

2019

Self Cite

View full text Add to dashboard Cite

In this paper we perform a comparative analysis of three models for feature representation of text documents in the context of document classification. In particular, we consider the most often used family of models bag-of-words, recently proposed continuous space models word2vec and doc2vec, and the model based on the representation of text documents as language networks. While the bag-of-word models have been extensively used for the document classification task, the performance of the other two models for the same task have not been well understood. This is especially true for the network-based model that have been rarely considered for representation of text documents for classification. In this study, we measure the performance of the document classifiers trained using the method of random forests for features generated the three models and their variants. The results of the empirical comparison show that the commonly used bag-of-words model has performance comparable to the one obtained by the emerging continuous-space model of doc2vec. In particular, the low-dimensional variants of doc2vec generating up to 75 features are among the top-performing document representation models. The results finally point out that doc2vec shows a superior performance in the tasks of classifying large

show abstract

Section: Network Based Modelsmentioning

confidence: 99%

The Influence of Feature Representation of Text on the Performance of Document Classification

2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…The use of multi-layer sentence networks (distinguishing links between sentences from the same document and those connecting sentences from different documents) has facilitated NLP undertakings such as multi-document summarization [ 22 ]. Moreover, increasing interest has been devoted to the rather microscopic features of linguistic networks, which are examined with measures including those concerning node centrality [ 22 – 24 ].…”

Section: Introductionmentioning

confidence: 99%

Linguistic emergence from a networks approach: The case of modern Chinese two-character words

Jin

Liu

2021

PLoS ONE

View full text Add to dashboard Cite

The models of linguistic networks and their analytical tools constitute a potential methodology for investigating the formation of structural patterns in actual language use. Research with this methodology has just started, which can hopefully shed light on the emergent nature of linguistic structure. This study attempts to employ linguistic networks to investigate the formation of modern Chinese two-character words (as structural units based on the chunking of their component characters) in the actual use of modern Chinese, which manifests itself as continuous streams of Chinese characters. Network models were constructed based on authentic Chinese language data, with Chinese characters as nodes, their co-occurrence relations as directed links, and the co-occurrence frequencies as link weights. Quantitative analysis of the network models has shown that a Chinese two-character word can highlight itself as a two-node island, i.e., a cohesive sub-network with its two component characters co-occurring more frequently than they co-occur with the other characters. This highlighting mechanism may play a vital role in the formation and acquisition of two-character words in actual language use. Moreover, this mechanism may also throw some light on the emergence of other structural phenomena (with the chunking of specific linguistic units as their basis).

show abstract

“…In addition to using five known methods, namely the weighted common neighbors (CN), the weighted Jaccard coefficient (JC), the weighted preferential attachment (PA), the weighted Adamic-Adar (AA) and the weighted resource allocation index (RA) [156,157,199], we also propose selectivity (SE) [11] and inverse selectivity (IS) as two effective weighted similarity measures. Selectivity is defined as the average weight distributed on the links incident to the single node, and has proven efficient for different language network tasks, ranging from the differentiation between original and shuffled text [21] to the differentiation of text genres [200] and for keyword extraction [201,202]. We also note that link prediction on Twitter has been studied before in [203], where CN, AA, JC and RA measures were combined with the information about corresponding communities as determined with a variant of the label propagation algorithm in unweighted and directed networks.…”

Section: Introductionmentioning

confidence: 99%

Language networks

Physics Subject Headings (PhySH)

View full text Add to dashboard Cite

The language networks book provides insights into the principles of modeling and analyzing structural properties of language-manly in its written form, hence text. Book guidelines the basic principles of text preprocessing, covering the very initial steps needed for any natural language processing task. Further, the book examines the possibilities of representing text in a complex networks framework. The second part overviews the application of language networks as one of a data science disciplines. It covers important data science topics for the processing of the big textual data from extracting the most salient structural parts of documents, across differentiation between text genres to predicting the speeding of the information through social media. Finally, the last part of the book is tasked with formal modeling of the linguistic subsystems in a multilayer complex networks formalism, which allows systematic study of language across all of its subsystems. The first part of the book studies the general principles of language networks construction and analysis. It covers language network construction types. Specifically, it analyzes the effects of constructing directed vs. undirected, weighted vs. unweighted network from lemmatized (stemmed) or non-lemmatized texts with stopwords included or excluded. The effects of text randomization are studied enabling better insights into characteristics of language networks compared to their shuffled counterparts. Some preliminary experiments reveal the possibilities of the differentiation of the structural properties of networks constructed from different text types and in different languages like Croatian, English, and Italian. Next, some initial insights into the characterization of syllabic networks are presented. The analysis of motifs of the linguistic networks reveals the typical building blocks of the structure of networks of the literature in the Croatian language. Finally, the first part of the book concludes with the LaNCoA a Python Toolkit for the construction and analysis of language networks implementing the majority of the findings presented in this part of the book. The second part of the book is dedicated to the applications of language networks. The language networks enable the extraction of the most salient words in texts-keywords and extraction of the domain knowledge-context studied on the content of Wikipedia entries. The applicative part of language networks includes the differentiation between different text types and polarization of tweets, as well. Finally, the possibilities of predicting the future content of tweets solely from the structural properties of the complex language networks are presented. Contents I Language Networks Construction

show abstract

Text Type Differentiation Based on the Structural Properties of Language Networks

Cited by 6 publications

References 23 publications

The Influence of Feature Representation of Text on the Performance of Document Classification

The Influence of Feature Representation of Text on the Performance of Document Classification

Linguistic emergence from a networks approach: The case of modern Chinese two-character words

Language networks

Contact Info

Product

Resources

About