Antonio Neme

Stylistics analysis and authorship attribution algorithms based on self-organizing maps

Abstract The style followed by authors can be thought of as a collection of attributes that defines the stylistics space. Texts from the same author tend to be similar in that space. However, the identification of stylistics spaces has proven to be challenging. Associated with the stylistics space is the authorship attribution task. On it, a text of unknown authorship is presented to a system, and the system is expected to identify the author of the text. Two modules define an authorship attribution algorithm: the stylistics space and a classifier. We present a methodology that includes both, a module that allows the identification of novel stylistics spaces, and a classifier to confront the authorship attribution task from the features that define space. The methodology imbricates feature selection, anomaly detection, classification, and visualization algorithms. We applied the capabilities of self-organizing maps not only for visualization but also for anomaly detection, which defines the basis of the classifier. We compared our authorship attribution algorithm with two existing ones. Our methodology achieved similar or better results under bag-of-words-related stylistics spaces, and it presented the lowest error under a novel stylistics space based on the rate of introduction of new words.

Authorship attribution as a case of anomaly detection: A neural network model

Writings by the same author usually share specific traits, the so-called stylome, which is defined as an abstraction of the constraints and specific sequences of words and phrases used in the texts. Although identifying a stylome has been elusive, some advancements in this area have been made. Here, we present a system trained with texts from a given author that then unveiled some of its features and, in turn, detected texts not written by that author, or written within a different style. The system is based on time series processing capabilities of an unsupervised neural network model known as the self-organizing map. The core idea is that a system trained with texts by one author should detect an anomaly when presented with texts from other authors. We present results of authorship identification in several contexts including known benchmarks as well as some examples from literature, journalism, and popular science.

An electoral preferences model based on self-organizing maps

Abstract A model of spatial pattern formation of electoral preferences is presented. In some voting districts, patterns of electoral preferences emerge, such that in nearby areas citizens tend to vote for the same candidate whereas in geographically distant areas the most voted candidate is that whose political position is distant to the latter. Those patterns resemble the spatial structure achieved by a non-supervised neural network model, named self-organizing map. This model is able to achieve spatial order from disorder and at the same time, to form a topographic map of the external random field, identified with advertising from the media. In this model, individuals are represented in two spaces: a static geographical location, and a dynamic political position. The modification of the latter leads to a pattern in which both spaces are correlated. Numerically, we explore conditions not previously studied that may lead to these kind of spatial patterns.

Self-Organizing Map Formation with a Selectively Refractory Neighborhood

Decreasing neighborhood with distance has been identified as one of a few conditions to achieve final states in the self-organizing map (SOM) that resemble the distribution of high-dimensional input data. In the classic SOM model, best matching units (BMU) decrease their influence area as a function of distance. We introduce a modification to the SOM algorithm in which neighborhood is contemplated from the point of view of affected units, not from the view of BMUs. In our proposal, neighborhood for BMUs is not reduced, instead the rest of the units exclude some BMUs from affecting them. Each neuron identifies, from the set of BMUs that influenced it in previous epochs, those to whom it becomes refractory to for the rest of the process. Despite that the condition of decreasing neighborhood over distance is not maintained, self-organization still persists, as shown by several experiments. The maps achieved by the proposed modification have, in many cases, a lower error measure than the maps formed by SOM. Also, the model is able to remove discontinuities (kinks) from the map in a very small number of epochs, which contrasts with the original SOM model.

Self-Organizing Maps with Non-cooperative Strategies (SOM-NC)

The training scheme in self-organizing maps consists of two phases: i) competition, in which all units intend to become the best matching unit (BMU), and ii) cooperation, in which the BMU allows its neighbor units to adapt their weight vector. In order to study the relevance of cooperation, we present a model in which units do not necessarily cooperate with their neighbors, but follow some strategy. The strategy concept is inherited from game theory, and it establishes whether the BMU will allow or not their neighbors to learn the input stimulus. Different strategies are studied, including unconditional cooperation as in the original model, unconditional defection, and several history-based schemes. Each unit is allowed to change its strategy in accordance with some heuristics. We give evidence of the relevance of non-permanent cooperators units in order to achieve good maps, and we show that self-organization is possible when cooperation is not a constraint.

Algorithms Inspired in Social Phenomena

Natural computing finds its source of inspiration in diverse biological phenomena and social behaviors from mainly insects and birds. In this chapter, we instead propose human social phenomena. The presented algorithms have been applied in optimization endeavours with success or are promising tools in the design of optimization techniques.

Self organizing maps as models of social processes: the case of electoral preferences

We propose the use of self-organizing maps as models of social processes, in particular, of electoral preferences. In some voting districts patterns of electoral preferences emerge, such that in nearby areas citizens tend to vote for the same candidate whereas in geographically distant areas the most voted candidate is that whose political position is distant to the latter. Those patterns are similar to the spatial structure achieved by self-organizing maps. This model is able to achieve spatial order from disorder by forming a topographic map of the external field, identified with advertising from the media. Here individuals are represented in two spaces: a static geographical location, and a dynamic political position. The modification of the later leads to a pattern in which both spaces are correlated.

Visualizing patterns in the air quality in mexico city with self-organizing maps

Air pollution in big cities is a major health problem. Pollutants in the air may have severe consequences in humans, creating conditions for several illness and also affect tissues and organs, and also affect other animals and crop productivity. From several years now, the air quality has been monitored by stations distributed over major cities, and the concentration of several pollutants is measured. From these data sets, and applying the data visualization capabilities of the self-organized map, we analyzed the air quality in Mexico City. We were able to detect some hidden patterns regarding the pollutant concentration, as well as to study the evolution of air quality from 2003 to 2010.

Detection of different authorship of text sequences through self-organizing maps and mutual information function

Writers tend to express their ideas with different styles, defined with the so called firm or stylome, which is an abstraction of the general constraints and specific combinations of words within their language they decide to follow. Although capturing this style has proven to be very difficult, some advances have been achieved. Here, we present a novel system that is trained with texts from the same author, and is able to unveil some of its features, and to apply them to detect texts not written by the same author, or, at least, not written with the previously learned features. The system is an hybrid model based in self-organizing maps and in information-theoretic aspects. In the model, mutual information function of unknown texts are compared to the mutual information function of texts from a known author. If the distance between these two distributions exceeds a certain threshold, then the unknown text is from a different author, otherwise the authorship is the same. The decision threshold is obtained by the self-organizing map trained with the texts from the same author. We present results in authorship identification in several contexts including classic literature, journalism (political, economical, sports), and scientific divulgation.

The Self-Organized Chaos Game Representation for Genomic Signatures Analysis

Genomic signatures are important as a source of comparison and classification of genomes. In particular, the Chaos Game Representation, an iterative mapping method, generates a frequency distribution of nucleotides of length k and represents it on a lattice of size 2x2. However, it lacks continuity in the sense that very different sequences are represented on contiguous cells of the lattice. Here, we propose an alternative method that organizes cells in such a way that continuity is higher than in the Chaos Game Rrepresentations. The cell organization is the outcome of a Self-Organizing Map and the obtained frequency distribution is named Self-Organized Chaos Game Representation. Experiments show that this visualization method is, at least, as good as the Chaos Game Representation, but it gives it a more intuitive sense when interpreting the images.