A neural network-based analysis of the expression of eco-ethical awareness of the trichotomy of nature, industry and society in Victorian English literature

Character embedding and character relationship embedding refer to the character and character relationship being characterized by vectors, which, like the idea of word embedding, is an attempt to characterize an abstract semantic information by numerical vectors. Character representation in Victorian English literature is a comprehensive summary of the character’s personality traits, social attributes, and external conditions, while character relationship representation is a comprehensive summary of the emotional and social attributes between characters.

In this paper, the core data for character embedding and character relationship embedding are the words spoken by the characters and the dialogues between the characters, and the definition of characters and the definition of character relationships are also based on these two parts. In the data preparation of this paper, a generalized self-supervised learning method is used to construct the data, and the core idea is to determine whether the two input segments are semantically unified or disjoint. For character embedding, the two inputs to the network are either the same person or two different people’s N-sentence conversation contents, along with the truth value of the classification. If these two pieces of content are the dialogue content of the same person, the final classification result will be positive. If the two conversations are not the same person’s conversations, the final classification result is a negative example. The data source for this task is the dialog content with annotations that takes over the dialog subject extracted in the above paper, which contains information such as the subject person and allows for character embedding and character relationship embedding. In addition, to avoid interference with the prediction of the model by the names of people appearing in the dialog, this paper masks the names of people appearing in the dialog.

In this paper, a classification model is proposed for solving the embedding problem of characters and their relationships. The model mainly employs a neural network to perform the binary classification task, with the embedding results of two sub-networks. The model used is the SBERT model. SBERT is an improvement made based on the Bert model in order to better utilize Bert’s textual comprehension capabilities in the field of Natural Language Processing [24].

The neural network utilizes two networks with shared weights to model the inputs, where different sequences are passed into each of the two networks during the training phase, and the networks are given modeled representations of these two sequences. Then, the similarity between these two sequence vector representations is calculated to measure the similarity between the two different incoming sequences and binary classification is performed based on the similarity. Eventually, it is possible to obtain networks that have good modeling representations of the sequences.

The architecture of the SBERT character embedding model is shown in Fig. 1. SBERT adds a pooling operation to the output of BERT to obtain a fixed size sentence embedding. In SBERT, three pooling strategies are used, using the output of the CLS word character, computing the average of all output vectors, and computing the maximum timeout value of the output vector, with the default configuration being the average.

Figure 1.

SBERT people embedded model architecture

In the experiments of this paper, the two main types of outputs, utilizing CLS word symbols and computing the average of all output vectors, were used. For the loss function, in this paper, we connect the sentence embeddings u and v with the element level difference |u – v| and multiply it with the trainable weight W_t ∈ R^3n×k. The final objective function is: 1o=softmax(Wt(u,v,| u−v |))

Where n is the dimension of the sentence hierarchy, k is the number of labels, and the cross-entropy loss function is used in this paper. In specific experiments, this paper uses Bert, which has been pre-trained, to initialize the neural network and bind its parameters, i.e., the two Bert sub-modules share the parameters during training.

In syntactic analysis, this paper focuses on dependency syntactic analysis, which is one of the most commonly used techniques in practice. Dependency analysis is an important task in natural language processing, aiming to obtain syntactic information in sentences in the form of dependencies between words. It is used in semantic analysis, machine translation, relation extraction and many other text analysis tasks, and can provide fine-grained semantic information for “denotation disambiguation”.

The advantage of dependent syntactic analyses is that they can deal with morphologically rich languages with relatively free word order. Syntax requires separate rules for each position in the syntax tree where such gerund phrases may occur. Dependency-based approaches have only one connection type to represent this particular grounded relationship. Thus, dependency-intensive grammar approaches abstract from word order information and represent only the information needed for parsing. In recent years, machine learning has developed rapidly, and the task of constructing models using neural networks to perform dependency syntactic analysis has achieved very good results while consuming less computational resources.

To find out the relationship between characters in literature, this paper uses the K-means clustering method to cluster the characters and analyze the experimental results. [26]. Clustering algorithm is an unsupervised learning algorithm, in unsupervised learning, the target attributes do not exist, it divides the data into different categories based on the features present in the data, making the data within the categories more similar.

Clustering is essentially a division method used for data analysis that treats observations of data as objects based on the location and distance between different input data points. The division of objects into mutually exclusive clusters is accomplished in such a way that the objects in each cluster are kept as close as possible to each other, but as far away as possible from the objects in other clusters.

k-means clustering is a commonly used method to automatically divide a dataset into k groups. It does this by selecting k initial cluster centers and then iteratively refining them.

First, each instance is assigned to its nearest cluster center.

Then, each cluster center is updated to the average of its constituent instances. The algorithm is convergent when the instances are assigned clusters without further changes. If expressed in terms of a data expression, assuming that the clusters are divided into (B₁, B₂, …, B_k), the aim is to minimize the squared error E as: 3E=∑i=1k∑x∈Bi‖ x−μi ‖22 where μ_i is the mean vector of cluster B_i, which is the center of mass, and the expression is: 4μi=1| Bi |∑x∈Bix

In the traditional k-means algorithm, the input data is the sample set A = {A₁, A₂, …, A_m}, the cluster tree k for clustering, the maximum number of iterations S, and the output is the division of clusters B = {B₁, B₂, …, B_k}.

k samples are randomly selected from the data set A as the initial k center-of-mass vectors {μ₁, μ₂, …, μ_k}, for s = 1,2,…,S, the cluster divisions are initialized as B_s = ϕ, s = 1,2,…,k, and for i = 1,2,…,m, the distances between sample x_i and the individual center-of-mass vectors μ_j (1,2,…,k) are computed as: 5dij=‖ xi−μj ‖2

The cluster is updated by labeling the smallest x_i as the category μ_i corresponding to d_ij: 6Bλi=Bλi∪{ xi }

For j = 1,2,…,k, recalculate the new center of mass for all sample points in B_j: 7μj=1| Bj |∑x∈Bjx

If all k center-of-mass vectors remain unchanged, output the cluster division B = {B₁, B₂, …, B_k}.

Character portraits can vividly depict individual characters and help readers understand them. Characters are an important component of literary works, and a more comprehensive understanding of a character’s image can be achieved through the description of a literary work’s character’s appearance, action description, language description and other aspects. In this paper, we choose to describe the characters of literary works in terms of character and gender, and analyze the characters of literary works through automatic classification.

The process of automatic character classification is shown in Fig. 3. The automatic classification of characters includes the automatic classification of personality and gender for the characters in the collected literature. Two methods are used to represent character vectors.

Figure 3.

Classification process of character and gender

Method I, using the character vector representation mentioned in Fig. c_pdnan_vec.

Method II, use the average vector of feature words as the vector of characters of literary works, denoted as c_ave_vec, using the training feature word vector GloVe.

In automatic classification of data, neural network-based classifiers are used. Some of the common neural networks are Perceptual Machine, BP, RBF networks, and so on. A perceptron is the simplest feed forward network, which inputs the feature vector of the instances and outputs the category of the instances. The perceptron corresponds to a separating hyperplane in the input space that divides the instances into two classes. The perceptual machine aims to find this hyperplane. A loss function based on misclassification is imported to find the hyperplane, and the gradient descent method is used to optimize the loss function. In this paper, the perceptual machine is chosen to classify the character vectors. The nodes of the input layer, hidden layer, and output layer of the perceptual machine are 200, 100, and 5 respectively. The number of iterations is 200, and the training process is as follows:

Given a dataset T = {(x₁, y₁),(x₂, y₂),…,(x_N, y_N)}, x ∈ X = Rⁿ, y_i = {+1, –1}, i = 1,2,…,N, the input x_i denotes the feature vector of the instance, which corresponds to a point in the input space, and the output y_i denotes the category of the instance, the expression from the input space to the output space is given by: 8y=sign(a⋅x+b)

This function is called the perceptron, where w, b are called the parameters of the model, a are called the weights, b is called the bias, and a·x is denoted as the inner product of a and x, where: 9sign(x)={ +1,x>0−1,x<0

A dataset T is considered linearly differentiable if there exists some hyperplane S : a · x + b = 0 that correctly classifies all sample points in the dataset. By correctly classified, it means, if a · x_i + b > 0 in sample point (x_i, y_i), then y_i = 1. If EE in sample point a · x + b < 0, then y_i = –1. otherwise, the data set T is linearly indistinguishable.

If there are some points that fail to be classified, the distance from the hyperplane of all misclassified points is used as a measure of the degree of classification failure. The distance r₁ from a point x_i to the hyperplane a · x + b = 0 in the n -dimensional space Rⁿ is: 10r=| a⋅xi+b |‖ a ‖

The distance of a misclassified point is always greater than zero, and the distance r₂ from the misclassified point to the hyperplane should be expressed as: 11r2=−yi(a⋅xi+b)‖ a ‖

||a|| is a constant, and assuming that all misclassified points are in the set M, the loss function obtained is: 12L(a,b)=−∑xi∈Myi(a⋅xi+b)

The problem of finding (a,b) is transformed into minimizing the loss function, i.e., into an optimization problem. The smaller the loss function, the fewer the misclassified sample points and the lower the degree of classification failure.

A literary character, whether real or virtual, has spatio-temporal characteristics. For real characters, space-time refers to the context of time in which they live. For a virtual character, it refers to the context of the art work (including novels, comics, plays, movies, etc.) in which he or she lives. Therefore, in this section, by analyzing and estimating the time-space in which the characters in the inter-temporal social network are located, a spatio-temporal coupling network is derived from it, which can support the study of simultaneous spatio-temporal character association.

Taking the Victorian era as an example, Catherine (1695-1761), the protagonist in Emily’s Wuthering Heights, and Poundbuy (1785-1848) in Dickens’ Hard Times. The chronological distributions generated by extracting the year observations in the lexical entries through the automatic analysis technique for literary works are shown in Figures 4 and 5. It can be seen that both literary characters show a single-peak curve pattern, with the highest frequency of co-occurrence in all Victorian British literature being in 1823 and 1856, respectively. Most of the character entries have a high year frequency distribution in the real survival interval of the characters, with an overall bell curve pattern. However, there are also some characters whose year distributions have a multi-peak pattern, which occurs mostly due to the fact that historical character entries refer to research, film and television productions, or commemorations of the character by future generations, resulting in smaller peaks even after the character’s death.

Figure 4.

Cooccurrence distribution of Emily’s Wuthering Heights (1785-1848)

Figure 5.

Cooccurrence distribution of Bounderby in Dickens’s Hard Times (1695-1761)

Based on the above characteristics, this study introduces the concept of active interval, which identifies the window of time with the highest year frequency in the chronological distribution of a person as the active interval of the person, and serves to analyze the approximate interval of the person’s actual survival period from the multi-peak distribution. Based on the general life span of human beings, the window size of the active interval is set to 100 years, and the program slides the window in the form of a sliding window from the smallest year observation on the axes to the largest observation year by year, in order to match the active interval with the highest year frequency. The year observations falling within the active intervals of a person’s entry are considered to be a set of samples related to the years of the person’s real social activities.

This study hypothesizes that the degree of dispersion of the year distribution within the active interval correlates with the length of a character’s lifespan, i.e., characters with shorter lifespans have a more concentrated year distribution. In order to test the above hypothesis, 10,000 characters of this paper on Victorian English literature with complete birth and death years in the infoboxes were randomly selected in this paper. These characters were divided into training and testing sets by 4:1, and the standard deviation s of the years in the active interval of each character as well as the true life span length y (based on the year of death - year of birth in the infobox) were calculated. In the training set, the correlation between the year standard deviation s and the character’s lifespan y was examined using the bivariate correlation analysis module of SPSS, and the results showed a Pearson correlation coefficient of 0.681 (p<0.01), indicating a significant positive correlation.

Next, using the linear regression analysis module of SPSS, scatter plots of the two variables, year standard deviation s and character longevity y, were obtained and linearly fitted as shown in Figure 6. The fitted regression model is y = 24.7 + 1.71s. The model’s goodness of fit R² = 0.505, the closer the R² is to 1 indicates that the model’s explanatory power is stronger, it is generally believed that in social sciences R² ≥ 0.25 that has a good degree of goodness of fit, so the model can be applied to the survival of the characters in this paper on the spatio-temporal analysis of the survival of the characters of the Victorian era of British literature.

Figure 6.

The relationship between standard deviation of active year and life span

R-type clustering is a clustering method based on variables, while Q-type clustering is a clustering method based on similarity or distance measures. R-type clustering does not take into account the category of the data points, but clusters them only based on the similarity between the variables. Q-type clustering is commonly used to solve classification or grouping problems. In R-type clustering, different statistical methods are usually used to measure the similarity between variables, such as correlation coefficient, chi-square test, etc. R-type clustering is usually applied in natural sciences and engineering fields, such as medicine, biology, and industrial process control. In this paper, R-type clustering is used, and the resulting co-occurrence matrix is imported into SPSS software. The systematic clustering method is used, and intergroup linkage is selected. The squared Euclidean distance is selected, which can initially determine the degree of association between each keyword.

Also taking Dickens’ Hard Times as an example, the automatic analysis of literature based on neural networks obtained the clustering results of the literary work’s assistant hero, Pond’s Bay, as shown in Figure 7. The clustering purity of this work reaches up to 0.719, and the story line is centered around “Contradiction”. Coking Coal Town is the environment where Pontébé is located, and it is also a heavily polluted industrial town, which is covered by smog all day long and “even the sunlight can’t penetrate”. But Pontembe, the rich man who controls the destiny of the town, continues to ignore this and continues to profit at the expense of destroying the environment. Pontebello’s fervent pursuit of money and fame has caused him to lose all sense of propriety and shame, and he believes that everything in the world, except for money and fame, has no value at all, including cloth, workers, feelings, and even life can be “manufactured” like a machine that produces commodities. He treats his workers harshly, treating them as objects equal to machines, whose only value is to produce products for profit.

Figure 7.

Dickens’s “hard times” clustering results

Victorian Britain was one of the first countries in the world to suffer from the encroachment of industrial civilization and science and technology, which caused serious environmental pollution. Industrialization and the development of science and technology has brought about the negative impact gradually exposed in front of the people, factory emissions of waste smoke, waste water, waste made the city was seriously polluted, while a large number of laborers from the rural areas to the industrial cities, due to the assets of the blind expansion and overdevelopment, as well as the rapid expansion of mankind’s materialistic desires, the deprivation of nature is also more and more obvious. Writers of this period were also aware of the impact of industrial civilization on nature and the original order of human life, and through meticulous observation and unique strokes, they conducted a series of criticisms and reflections on the development of advanced science and technology and its products in their works.

The Mill on the Floss depicts traditional Victorian country life and natural beauty in conflict with an industrialized society. The text describes the hometown of the main character, Maggie: the vast plains, the wide Floss River, and the beautiful town of St. Ogle, the rich pasture land, and the cheerful Lippo River. In order to maximize his profits and accumulate wealth, Mr. Pivot attempts to use water from the Lippo River to cultivate his own land upstream from the river, which results in Maggie’s father and Dulliver, a miller on the Floss River, being sued by Wickham. In this novel, the process of industrialization increases the selfishness and greed of human beings, which leads to uncontrolled demands and plundering of nature, and the natural ecological balance of harmony between human beings and nature is broken, and an ecological crisis ensues, and the flood that erupts at the end of the novel at Dolcott’s Mill contrasts sharply with the natural scenery on both sides of the River Floss at the beginning of the novel.

In Charlotte Bronte’s Jane Eyre, the story takes place at a time when the industrialization of England has spread from the cities to the countryside, and many villages have opened small factories for the mining and processing of ores, and the plundering and exploitation of natural resources by these factories have destroyed large areas of vegetation, and the land and the natural environment have been damaged.

George Eliot portrays a stark portrayal of the ecological crisis in industrialized Britain in Adam Bede. Snowfield, where Dinah has lived and worked for a long time, is an industrial area. It has well-developed cotton mills, linen mills, and lead mining, while the common people live in poverty.

In the Victorian writers, the blind expansion and over-exploitation of mankind has made many cities and villages devastated, completely isolated from nature, losing the beauty of natural freshness, which can be said to be a combination of air pollution, water pollution, solid waste pollution and noise pollution.

Victorian writers wrote a graphic description of the rapid development of the Industrial Revolution, people only cared about the immediate interests and self-interest, wantonly exploiting and utilizing nature, destroying the ecological and natural environment and other behaviors. At the same time, these behaviors were accused and attacked, but also revealed that the writer’s harmony between man and nature was destroyed after the loss and sadness of emotions.

In large industrial cities such as London and Birmingham in Victorian England, factories were dotted everywhere, the machines in the factories kept emitting deafening roars, and the tall smoke emitting thick smoke could be seen almost everywhere, and people working under the slavery of the machines struggled hard in the noise and smoke, and their will and patience were gradually exhausted by them. Industrialization, urbanization and the promotion and popularization of science and technology have greatly improved the material living conditions of human beings, but a series of negative impacts derived from them have not only led to the rapid deterioration of the ecological environment, but also brought about a great spiritual crisis to human beings. Human desires have also expanded dramatically with the development of science and technology and the emergence of ecological crises. From the beginning of plundering and conquering nature, mankind has moved towards plundering and oppressing mankind itself, distorting harmonious human relations and causing the alienation of civilization. This new law of life strongly binds human nature and disregards human emotions, resulting in chaotic order of life and cruel competition for survival.

Charles Dickens depicted the manifestation of Victorian anthropocentrism in his works. Dong Bei Fathers and Sons is a satire of Dong Bei’s anthropocentric values by Dickens. The bankruptcy of Dong Bei and Sons also highlights the inevitable fate that anthropocentric values will lead to tragedy. Through the depiction of Victorian characters, Dickens reveals that man is only a part of nature, not the ruler of nature, and the consequence of violating the laws of nature is inevitably to be punished by nature.

In Emily’s Wuthering Heights, the alienation of humans by industrial civilization has already begun. The main character, Catherine, has a passionate and natural personality and is in love with Hickory, who represents natural wildness. However, starting from the mistake of entering the Painted Fields, Catherine gradually deviates from her nature due to the temptation of the civilized world and decides to marry Edgar, who is the symbol of the civilized world. After entering Edgar’s civilized world, Kathryn falls into a state of spiritual commitment, which is confusion after she loses her nature. It is not until three years later that Hickory’s return awakens her nature, which has been lost to money and status, and Catherine is once again faced with the choice between nature and civilization. Realizing that she made the wrong choice, that her blind pursuit of civilization had severed her roots in nature, Katharine eventually returns to her true self at the end of her life and is buried in a corner of the church cemetery on the moor.

George Eliot provides a detailed description of the alienated civilization and relationships of the Victorian era in The Mill on the Floss. The Dodson family in the text is a rural middle class with a high social status and economic base. But all they care about is business and profit, and inter-family relationships are sustained by social status and economic stability. The bankrupt Doliver family did not get the help and care given by the Dodson family, and it was even regarded as a great shame for the Dodson family, and the family affection was completely alienated, and the imbalance of the relationship between people was the evil consequence of the imbalanced social ecology and alienated civilization in the Victorian era.

The mid-Victorian period was the middle stage of industrial civilization, capitalism was still in the ascendant stage, and the shortcomings of the capitalist society and the damages brought by the industrial civilization had not yet been fully exposed. Therefore, the “civilized people” in the works of writers of this period were not yet modern people in the modern sense; most of them were educated to a certain extent and hoped to gain social recognition through their own struggles, but most of them ended in failure. In the conflict between the self and society, they deeply experience both the warmth and coldness of the human world. Communication between people becomes very difficult, trust is lost, conflicts and contradictions continue, and alienation becomes the ultimate end, even among the closest people. The progress of material civilization has not corresponded with the corresponding progress of spiritual and social civilization. On the contrary, in an era of rapidly expanding materialistic desires, spiritual society has been gradually disregarded and distorted.

Eliot portrays the tragedy in the life of Maggie Dulliver, the heroine, in The Mill on the River Foss, through her journey of mental and ecological imbalance after a series of setbacks. Maggie grows up to be gifted and studious, but because she is a girl, she cannot enjoy the right to education like her brother Tom. Not only does Maggie’s intelligence and vitality fail to win the approval of those around her, but she is also considered to be rude and lacking in feminine virtues such as good manners and gentleness. Unrecognized, Maggie tries to find a key to solve all the problems and reveal the mystery of life. Maggie’s temporary fulfillment is predicated on giving up her independence and her pursuit of individuality in an effort to harmonize with the social values of the entire town of St. Ogg. She could have left St. Aug’s with Stephen and gone to the elegant life with knowledge, books, and music that she had been pursuing, but she was bound by her family responsibilities, and when she returned to her family, she was disallowed by her family and society. Maggie’s transition from an innocent childhood to self-restraint and maturity is aptly mapped out the transformation of British society from a peaceful and self-contained traditional society to an industrial society, and Maggie’s personal pain and loss during this process is a true reflection of the fault line in the development of society.

With the advancement of industrial civilization, love and marriage also changed in industrial society. Undying love, such as Wade’s, is the kind of love that “craves the rare, hates the ready-made. Rare the far away, hate the near.” Tessa’s love for Clem was based on the luxury of Parisian life, which she thought she could enjoy in Clem’s person, “music, poetry, a thousand revolutions, a thousand battles, all the throbbing and jolting of the world’s arteries”, but when all this did not turn out as she had hoped after her marriage, and Clem was wounded in the eye, and was willing to be a laborer who cuts ivy and digs peat. At this time, you Tessa feel like life as a waste, can not help but mope all day, love is also gradually in this dissatisfaction slowly disappeared.

In the deteriorating natural environment and the distorted and unbalanced social ecology, people’s spirits became lost, confused, and entangled. In the face of the rapidly changing natural and social ecology, Victorian writers tried to find ways to solve their bewilderment and confusion, and endeavored to propose solutions to these dilemmas in their own works.