Word Adjacency Networks Research Articles

The local topology of dynamical network models for biology

Abstract The search for motifs—recurrent patterns in network topology—has led to the identification of universal classes of complex systems across diverse fields and has served as a quantitative tool to reveal common properties in both evolved and designed networks. In this study, we investigate the presence and significance of network superfamilies—previously identified through the census of triadic motifs—in the largest data set of dynamic, biological network models. We present triad significance profiles of 71 existing biological network models, all experimentally inspired. The generated data are treated in an unbiased manner and consistently clustered into two classes using several unsupervised techniques. The more prevalent class does exhibit a strong correlation with the superfamily of sensory transmission networks, which are characterized by the feedforward loop motif commonly found in signal-processing systems. Surprisingly, the remaining class shows a better correlation with the superfamily of word-adjacency networks. To better understand this, the results are analysed for varying network size thresholds, and their connection to the effect of model building activity is examined. It is highlighted that the more the model focuses on smaller portions of the regulatory network, as a result of the coarse-graining of the boundary dynamics and the peripheral regions of the network, the more its topology starts resembling that of ‘sentences’ of word-adjacency networks.

False statements in Egan et al.’s defence of their word adjacency network method

False statements in Egan et al.’s defence of their word adjacency network method

“I would I had that corporal soundness”: Pervez Rizvi's Analysis of the Word Adjacency Network Method of Authorship Attribution

Abstract In his two-part article ‘An Analysis of the Word Adjacency Network Method—Part 1—The evidence of its unsoundness’ and ‘Part 2—A true understanding of the method’ Digital Scholarship in the Humanities, 38: 347-78 (2022), Pervez Rizvi attempts to replicate the Word Adjacency Network (WAN) method for authorship attribution and show that it does not produce the new knowledge that we, its inventors, claim for it. In the present essay, we will show that Rizvi misrepresents fundamental aspects of the WAN method, that his attempted replication fails not because the method is flawed but because he erred in replicating it, and that Rizvi misunderstands key aspects of the mathematics of Information Theory that the method uses.

An analysis of the Word Adjacency Network method—Part 2—A true understanding of the method

AbstractThis second of a two-part examination of the Word Adjacency Network (WAN) method considers the idea behind the method, which is that authors place function words in proximity to each other in such distinctive ways that the distinctions can be used for authorship attribution. By doing control experiments, it shows that the attributions made by the method are not due to distinctive word adjacencies at all, contrary to the claim of its inventors. It shows that the method is in fact another word-counting method, the counting hidden by superfluously complicated mathematics. It concludes that the claim that the placements of function words in a text can be used for attributing its authorship remains unsubstantiated. It also demonstrates that the claimed theoretical basis for the method, that WAN are Markov chains, is false.

An analysis of the Word Adjacency Network method—Part 1—The evidence of its unsoundness

AbstractA new mathematical method for authorship attribution, known as the Word Adjacency Network method, has emerged and its inventors have claimed a high success rate with it. They have announced that their method has shown that Christopher Marlowe wrote some scenes in the Henry VI trilogy of plays, an attribution which has made newspaper headlines. This first part of a two-part paper replicates the method’s success in attributing complete plays but demonstrates that it cannot be relied on to attribute individual scenes correctly, particularly where Marlowe is a candidate author. It shows by examining the method’s calculations that its unreliability is due to its mass exclusion of relevant evidence, that exclusion being necessitated by its use of a mathematical formula which can now be seen to be unsuited to the texts of early modern plays.

How the Word Adjacency Network (WAN) works

Abstract The Word Adjacency Network (WAN) method is a newly invented method for attributing the authorship of texts based on internal evidence about the proximities of high-frequency words in those texts. The method has been the subject of mathematically rigorous explanations in scientific journals and here is offered instead a relatively non-technical account for humanist readers unfamiliar with advanced mathematical nomenclature. The description will refer to the operation of the algorithm as it is implemented in a newly completed open-source, open-access version that the authors have made available for free unlimited download on the WorldWide Web.

A Response to Rosalind Barber’s Critique of the Word Adjacency Method for Authorship Attribution

The Word Adjacency Network (WAN) method of authorship attribution was introduced in a series of papers by the present authors between 2015 and 2018. ANQ has published two critiques of the method (B...

Linguistic data mining with complex networks: A stylometric-oriented approach

By representing a text by a set of words and their co-occurrences, one obtains a word-adjacency network being a reduced representation of a given language sample. In this paper, the possibility of using network representation to extract information about individual language styles of literary texts is studied. By determining selected quantitative characteristics of the networks and applying machine learning algorithms, it is possible to distinguish between texts of different authors. Within the studied set of texts, English and Polish, a properly rescaled weighted clustering coefficients and weighted degrees of only a few nodes in the word-adjacency networks are sufficient to obtain the authorship attribution accuracy over 90%. A correspondence between the text authorship and the word-adjacency network structure can therefore be found. The network representation allows to distinguish individual language styles by comparing the way the authors use particular words and punctuation marks. The presented approach can be viewed as a generalization of the authorship attribution methods based on simple lexical features.Additionally, other network parameters are studied, both local and global ones, for both the unweighted and weighted networks. Their potential to capture the writing style diversity is discussed; some differences between languages are observed.

In narrative texts punctuation marks obey the same statistics as words

From a grammar point of view, the role of punctuation marks in a sentence is formally defined and well understood. In semantic analysis punctuation plays also a crucial role as a method of avoiding ambiguity of the meaning. A different situation can be observed in the statistical analyses of language samples, where the decision on whether the punctuation marks should be considered or should be neglected is seen rather as arbitrary and at present it belongs to a researcher’s preference. An objective of this work is to shed some light onto this problem by providing us with an answer to the question whether the punctuation marks may be treated as ordinary words and whether they should be included in any analysis of the word co-occurrences. We already know from our previous study (S. Drożdż et al., Inf. Sci. 331 (2016) 32-44) that full stops that determine the length of sentences are the main carrier of long-range correlations. Now we extend that study and analyse statistical properties of the most common punctuation marks in a few Indo-European languages, investigate their frequencies, and locate them accordingly in the Zipf rank-frequency plots as well as study their role in the word-adjacency networks. We show that, from a statistical viewpoint, the punctuation marks reveal properties that are qualitatively similar to the properties of the most frequent words like articles, conjunctions, pronouns, and prepositions. This refers to both the Zipfian analysis and the network analysis. By adding the punctuation marks to the Zipf plots, we also show that these plots that are normally described by the Zipf–Mandelbrot distribution largely restore the power-law Zipfian behaviour for the most frequent items.Our results indicate that the punctuation marks can fruitfully be considered in the linguistic studies as their inclusion effectively extends dimensionality of an analysis and, therefore, it opens more space for possible manifestation of some previously unobserved effects.

Network analysis of named entity co-occurrences in written texts

The use of methods borrowed from statistics and physics to analyze written texts has allowed the discovery of unprecedent patterns of human behavior and cognition by establishing links between models features and language structure. While current models have been useful to unveil patterns via analysis of syntactical and semantical networks, only a few works have probed the relevance of investigating the structure arising from the relationship between relevant entities such as characters, locations and organizations. In this study, we represent entities appearing in the same context as a co-occurrence network, where links are established according to a null model based on random, shuffled texts. Computational simulations performed in novels revealed that the proposed model displays interesting topological features, such as the small world feature, characterized by high values of clustering coefficient. The effectiveness of our model was verified in a practical pattern recognition task in real networks. When compared with traditional word adjacency networks, our model displayed optimized results in identifying unknown references in texts. Because the proposed representation plays a complementary role in characterizing unstructured documents via topological analysis of named entities, we believe that it could be useful to improve the characterization of written texts (and related systems), specially if combined with traditional approaches based on statistical and deeper paradigms.

Concentric network symmetry grasps authors' styles in word adjacency networks

Several characteristics of written texts have been inferred from statistical analysis derived from networked models. Even though many network measurements have been adapted to study textual properties at several levels of complexity, some textual aspects have been disregarded. In this paper, we study the symmetry of word adjacency networks, a well-known representation of text as a graph. A statistical analysis of the symmetry distribution performed in several novels showed that most of the words do not display symmetric patterns of connectivity. More specifically, the merged symmetry displayed a distribution similar to the ubiquitous power-law distribution. Our experiments also revealed that the studied metrics do not correlate with other traditional network measurements, such as the degree or the betweenness centrality. The discriminability power of the symmetry measurements was verified in the authorship attribution task. Interestingly, we found that specific authors prefer particular types of symmetric motifs. As a consequence, the authorship of books could be accurately identified in 82.5% of the cases, in a dataset comprising books written by 8 authors. Because the proposed measurements for text analysis are complementary to the traditional approach, they can be used to improve the characterization of text networks, which might be useful for applications based on stylistic classification.

Modeling the average shortest-path length in growth of word-adjacency networks.

We investigate properties of evolving linguistic networks defined by the word-adjacency relation. Such networks belong to the category of networks with accelerated growth but their shortest-path length appears to reveal the network size dependence of different functional form than the ones known so far. We thus compare the networks created from literary texts with their artificial substitutes based on different variants of the Dorogovtsev-Mendes model and observe that none of them is able to properly simulate the novel asymptotics of the shortest-path length. Then, we identify the local chainlike linear growth induced by grammar and style as a missing element in this model and extend it by incorporating such effects. It is in this way that a satisfactory agreement with the empirical result is obtained.

Authorship Attribution Through Function Word Adjacency Networks

A method for authorship attribution based on function word adjacency networks (WANs) is introduced. Function words are parts of speech that express grammatical relationships between other words but do not carry lexical meaning on their own. In the WANs in this paper, nodes are function words and directed edges stand in for the likelihood of finding the sink word in the ordered vicinity of the source word. WANs of different authors can be interpreted as transition probabilities of a Markov chain and are therefore compared in terms of their relative entropies. Optimal selection of WAN parameters is studied and attribution accuracy is benchmarked across a diverse pool of authors and varying text lengths. This analysis shows that, since function words are independent of content, their use tends to be specific to an author and that the relational data captured by function WANs is a good summary of stylometric fingerprints. Attribution accuracy is observed to exceed the one achieved by methods that rely on word frequencies alone. Further combining WANs with methods that rely on word frequencies alone, results in larger attribution accuracy, indicating that both sources of information encode different aspects of authorial styles.

COMPLEX NETWORK ANALYSIS OF LITERARY AND SCIENTIFIC TEXTS

We present results from our quantitative study of statistical and network properties of literary and scientific texts written in two languages: English and Polish. We show that Polish texts are described by the Zipf law with the scaling exponent smaller than the one for the English language. We also show that the scientific texts are typically characterized by the rank-frequency plots with relatively short range of power-law behavior as compared to the literary texts. We then transform the texts into their word-adjacency network representations and find another difference between the languages. For the majority of the literary texts in both languages, the corresponding networks revealed the scale-free structure, while this was not always the case for the scientific texts. However, all the network representations of texts were hierarchical. We do not observe any qualitative and quantitative difference between the languages. However, if we look at other network statistics like the clustering coefficient and the average shortest path length, the English texts occur to possess more clustered structure than do the Polish ones. This result was attributed to differences in grammar of both languages, which was also indicated in the Zipf plots. All the texts, however, show network structure that differs from any of the Watts–Strögatz, the Barabási–Albert, and the Erdös–Rényi architectures.

Edge direction and the structure of networks

Directed networks are ubiquitous and are necessary to represent complex systems with asymmetric interactions--from food webs to the World Wide Web. Despite the importance of edge direction for detecting local and community structure, it has been disregarded in studying a basic type of global diversity in networks: the tendency of nodes with similar numbers of edges to connect. This tendency, called assortativity, affects crucial structural and dynamic properties of real-world networks, such as error tolerance or epidemic spreading. Here we demonstrate that edge direction has profound effects on assortativity. We define a set of four directed assortativity measures and assign statistical significance by comparison to randomized networks. We apply these measures to three network classes--online/social networks, food webs, and word-adjacency networks. Our measures (i) reveal patterns common to each class, (ii) separate networks that have been previously classified together, and (iii) expose limitations of several existing theoretical models. We reject the standard classification of directed networks as purely assortative or disassortative. Many display a class-specific mixture, likely reflecting functional or historical constraints, contingencies, and forces guiding the system's evolution.

PROSE AND POETRY CLASSIFICATION AND BOUNDARY DETECTION USING WORD ADJACENCY NETWORK ANALYSIS

Word adjacency networks constructed from written works reflect differences in the structure of prose and poetry. We present a method to disambiguate prose and poetry by analyzing network parameters of word adjacency networks, such as the clustering coefficient, average path length and average degree. We determine the relevant parameters for disambiguation using linear discriminant analysis (LDA) and the effect size criterion. The accuracy of the method is 74.9 ± 2.9% for the training set and 73.7 ± 6.4% for the test set which are greater than the acceptable classifier requirement of 67.3%. This approach is also useful in locating text boundaries within a single article which falls within a window size where the significant change in clustering coefficient is observed. Results indicate that an optimal window size of 75 words can detect the text boundaries.

Clustering signatures classify directed networks

We use a clustering signature, based on a recently introduced generalization of the clustering coefficient to directed networks, to analyze 16 directed real-world networks of five different types: social networks, genetic transcription networks, word adjacency networks, food webs, and electric circuits. We show that these five classes of networks are cleanly separated in the space of clustering signatures due to the statistical properties of their local neighborhoods, demonstrating the usefulness of clustering signatures as a classifier of directed networks.

Superfamilies of evolved and designed networks.

Complex biological, technological, and sociological networks can be of very different sizes and connectivities, making it difficult to compare their structures. Here we present an approach to systematically study similarity in the local structure of networks, based on the significance profile (SP) of small subgraphs in the network compared to randomized networks. We find several superfamilies of previously unrelated networks with very similar SPs. One superfamily, including transcription networks of microorganisms, represents "rate-limited" information-processing networks strongly constrained by the response time of their components. A distinct superfamily includes protein signaling, developmental genetic networks, and neuronal wiring. Additional superfamilies include power grids, protein-structure networks and geometric networks, World Wide Web links and social networks, and word-adjacency networks from different languages.