Retracted: Comparison and Research on the Applicability of Combining Information of Chinese and Foreign Literary Genres Based on Data Analysis

,


Introduction
In terms of race, the Chinese are more practical than metaphysical, with moral creeds being the most developed and systematic metaphysics being silent; in terms of literature, works on human and social issues are the most developed, while works based on imaginary structures are rare.Chinese nationality is the most "practical" and "humane."+is is its strength and weakness.Its weaknesses are also reflected here.Its strength lies in the importance it places on human relations, enabling a disorganized society to remain stable for more than two years, but its weakness lies in its overemphasis on humanism and existentialism and its inability to pursue a higher status [1].+e first, the most superficial, is "sensualism," the second begins with tacit love and joy, and the third is pantheism, in which nature is seen as a manifestation of God and his mysterious and superhuman sense of power, and the worship of nature becomes a religion.+is is the attitude of most Western poets toward nature, and few Chinese poets have reached this level.In terms of philosophy and religion, Zhu Guangqian argues that "Western poetry is deeper and broader than Chinese poetry because it has a deeper and broader philosophy and religion that nurtures its roots and backbone [2]." As an example, language textbooks have limited pedagogical capacity, and most of the foreign literature is chosen based on the principles of genre complementarity and adequate argumentation.+e authors use Hamlet (excerpt), text in the human education version of the high school language, as an example, where the references to Christian imagery and biblical structure can be clearly seen and the text moves towards mysticism from the appearance of the old king's ghost.However, in contrast, the selection of domestic literature in the textbook, such as the large proportion of ancient poems in the textbook, reflects the recognition of the "practicality" and solidity of tradition in contemporary Chinese literature, in addition to the realistic content and moral connotations of the ancient poems themselves.
In fact, the more active a literary work is, the less it will fade away from the heraldic symbols of national life and spirit, and the less it will sever its natural connection with the national literary tradition [3], as opposed to sticking to tradition.Both Chinese literature and the history of the national literature of the world provide such indications.It is hard to imagine that a particular country or nation could build its own sublime literary temples on the ruins of its culture.It is hard to imagine that any one nation or people could build its own sublime literary temples on cultural ruins [4,5].+e argument that the so-called "yellow culture" should be transformed into "blue culture" has become a theoretical fashion and guide for the practice of literary creation.+is trend of thought is also a symbol of the era of reform and opening up, which increases the complexity of distinguishing right from wrong [6].
However, the dichotomy between "yellow culture" and "blue culture" has led to the lack of a close relationship between the language materials of Chinese and foreign literature, which are scattered in the language materials of primary and secondary schools.In the author's opinion, foreign literature in language textbooks should strengthen the horizontal connection with Chinese literature, absorb the favorable factors that keep foreign literature alive, and strengthen the means of traditional culture education [7], so as to make use of the clear framework, long history, and coherent development of Chinese literature and establish corresponding connections, so that the study of foreign literature from the perspective of Chinese traditional culture not only enables us to analyze foreign literature in a more profound way but also allows us to analyze foreign literature from the perspective of Chinese traditional culture.Works in a more profound way, but also to transmit traditional culture and improve the proportion of traditional culture in the teaching of secondary school [8], the authors also hope to use the framework of experimental research on traditional culture and traditional literature to provide a reference for better dissemination of traditional culture in teaching and learning and for further integration of language teaching and traditional culture transmission, textbook texts and group reading [9].
Since the 21st century, artificial intelligence techniques have evolved as computer performance has improved and massive amounts of data have become available.In 2012, AlexNet, designed by Hinton and his student Alex Krzyzewski, won the image classification competition [10], and since then, deep learning has become increasingly well known.Deep learning is a branch of machine learning that has achieved great success in breaking through traditional machine learning bottlenecks such as image recognition, audio processing, and natural language processing and has opened a new era of artificial intelligence.Natural language processing (NLP) usually includes linguistics, mathematics, computer science, and other related knowledge, is a fusion of multiple disciplines, and is an important research direction in the field of artificial intelligence [11].
In this paper, based on the existing research results of literature generation, we adopt a literary poetry generation approach based on deep learning related techniques.+e purpose of this study is to achieve better teaching results for literature students by comparing Chinese and foreign literature and creating poetry at the level of human creativity.

Related Work
Literature is a representative poetry genre in Chinese traditional poetry, and the study of literature generation-related work can be seen as a study of poetry generation-related work.+e study of literary production can be seen as the study of poetry production, so this section focuses on the current state of research on literary production.+erefore, this section discusses the current state of research on literary production, mainly in terms of the current state of research on poetry production.[12] proposed a poetry generation method, but it only uses the random combination of phrases to generate poetry.+e main rule-and template-based generation methods are template-based methods [13,14] and instance-based reasoning methods [15], which are more likely to generate poems by filling in the blanks and combinations through template settings, and the generated poems are very incoherent and cannot even be called poems in the traditional sense.In the stage based on statistical machine learning, [16] introduced genetic algorithms in Song word generation and regarded Song word generation as an optimization problem; [17] regarded the poetry generation problem as a summary generation problem specifying the writing intention by retrieving words in a poetry corpus for sorting and then combining them to form verses; [7] [11] generates the whole poem line by line by user-supplied keywords, then expands the keywords, given the relevant metrical restrictions, and then selects the best-ranked one as the first line of the poem.Reference [6] combines the attention-based mechanism of neural network machine translation method.Attention based Neural Machine Translation Network (ANMT) [12] is applied to the generation of song lyrics, which treats song lyrics generation as a translation problem, and takes the generated verses as the input and the verses to be generated as the target output for poetry generation.iPoet is a poetry generation system based on the encoder-decoder framework [13], which generates poems by Hafez and it is a program that generates any number of poems based on a user-provided theme [14], which implements the generation of English poems based on the encoder-decoder framework.
+e Planning based Poetry Generation (PPG) model [15] imitates the human action of writing a poem to outline and divide the poetry writing process into two stages, poetry planning and poetry generation, by planning subthemes for poetry generation, which solves the problem of theme drift to some extent, but its theme planning is based on modern vernacular.It is easy to cause the problem of poor semantics.+e Natural Language Processing and Social Humanities Computing Laboratory of Tsinghua University proposed a

R E T R A C T E D
poetry generation method based on Salient-Clue Mechanism [16], which extracts words with salient features from generated verses and restricts the type of generated poems to generate poems, which has good contextual coherence and better overall effect.

Design of the Literature Generation Model
Based on Keyword Transformation Extension Quatrain Generation Model (KTEQG), which first extracts the unique topic keywords through the user's input of writing intention and then transforms the keywords into a literary language to generate the poem.+e transformed keywords are subjected to literary transformation, and then the transformed keywords are subjected to thematic expansion, and each line of verses by assigning relevant topic words, and an encoder-decoder model based on the attention mechanism, which takes each keyword and history generation as input for a literary generation.+e model is divided into three stages: keyword conversion, keyword expansion, and literature generation.In the keyword conversion stage, keyword extraction is first performed on the user's writing intention (user input topic words, sentences, or text passages) to determine the unique topic keywords, and then the unique keywords are converted into literary keywords words.+e keyword expansion stage expands the literary keywords to four subtopic keywords based on the literary corpus, which are assigned to each line of literature.In the literature generation phase, the subthemes generated in the keyword expansion phase and the historically generated verses are used as input to output the next line of poetry and repeated until the whole poetry generation is completed.

Keyword Conversion.
In this study, since there is no keyword annotation data in the collected literary corpus, and the supervised and semisupervised methods need to annotate the poetic corpus with keywords, if the literary corpus is manually annotated, the workload is huge, too tedious, and too costly.+erefore, the unsupervised method is chosen for user writing intention keyword extraction in this study.+e commonly used unsupervised methods include the TF-IDF algorithm [8], LDA topic model [4], and TextRank algorithm [9], etc. +e TF-IDF algorithm judges the importance of words purely by word frequency, which usually works better in long texts but is not suitable for short texts like literature.
+e TextRank algorithm is derived from Google's Pag-eRank algorithm, which is a graph-based ranking algorithm.+e TageRank algorithm slices the text into several component units and then builds a graph model and ranks the words in the text through a voting mechanism.+e algorithm only needs the information of a single text itself to achieve keyword extraction.+is algorithm only requires information from a single text to achieve keyword extraction, and it is used by Google Let's Rank to rank web pages by calculating the number and quality of web links.+e algorithm is used by Google to rank web pages by calculating the number and quality of web links to measure the importance of web pages.+e TextRank algorithm, on the other hand, performs keyword extraction by first dividing the text into multiple text units.+e similarity between several text units is the edge between nodes.+e graph model is formed by using the PageRank algorithm to iterate the graph model until it converges.
+e TextRank algorithm model is usually denoted by G � (V, E), where V denotes the set of nodes in the graph, E denotes the set of edges in the graph, and E is a subset of V × V. +e score of node V i is shown in the following: When using the TextRank algorithm, the initial value is usually set to 1 for all nodes, and the convergence threshold is usually set to 0.0001, i.e., convergence is achieved when the error rate is less than 0.0001, and the iteration is stopped.
When applied to keyword extraction specifically, the TextRank algorithm focuses on keyword extraction through cooccurrence relationships between lexical items.+e steps are as follows [9]

R E T R A C T E D
keyword extraction, there is no weight size between words, and the initial value is assumed to be 1.+erefore, formula ( 2) is modified and adjusted as follows: (5) Output keywords: all nodes are sorted, and the word with the highest score is the unique keyword for poetry theme extraction.
After the unique keyword extraction of writing intention is completed by the above way, since the extracted keywords are prone to vernacular words in modern Chinese, and in literature generation, it is based on ancient literary words in literary corpus verses, so the mismatch between modern vernacular words keywords and it is not conducive to literature generation, which may result in poor semantics of generated poems.To solve this problem, the literary generation model proposed in this paper will perform literary conversion after extracting the unique topic keywords and then perform topic expansion based on the converted literary words to generate literary subtopic keywords words before literary generation.+is study will directly use the relevant interface of the Baidu translation open platform to perform keyword conversion.
Baidu Translation Open Platform is a platform for Baidu Translation to provide open services for developers, relying on Baidu's powerful natural language processing technology and high level translation technology and providing powerful, simple and easy-to-use translation APIs and SDKs and other interface services for developers.Its universal translation API supports 28 languages such as Mandarin, Chinese, English, Japanese, and Korean translation for free.Simply call the universal translation API, input the corresponding translated content, and set the source and target languages to get the desired translation results [2].+erefore, the extracted unique keywords can be translated by using the universal translation API to realize the translation service of converting modern vernacular words to literary words.+e Baidu Universal Translation API is used as follows [8]: pass in the content to be translated, specify the source language and target language, and get the corresponding translation results.Among them, Input: access the service by sending a field to the Universal Translation API HTTP address via the POST or GET method.

Literary Generation. +e literary generation phase,
where the subject keywords are transformed and expanded, is achieved by assigning subsubject keywords, which can be seen as a sequence-to-sequence generation process, which will be based on the previously extracted and transformed expanded subsubject keywords.+e literary generation will be based on the previously extracted and transformed extended subtopic keywords and the historically generated content, and this paper will use an attention mechanism based encoder-decoder model for literary poem generation.+is paper will use an attention-based encoder-decoder model for literary poem generation.+e encoder-decoder is a sequence-to-sequence model, which transforms the input sequence into a fixed vector in the encoding phase and a fixed vector in the decoding phase.Into a fixed vector, the decoding stage decodes the fixed vector output from the coding side into an output sequence, and its input sequence and output sequence are the same length.+is model is widely used in image processing, translation, and other fields.It is a framework class model rather than a specific model.+e input and output can be various data forms, such as text, audio, image data, etc., and can also be various deep learning models, such as CNN, RNN, LSTM, GRU, etc.Its model framework can be GRU, etc., and its model framework can be abstractly represented as in Figure 2.
In the field of natural language processing, the encoderdecoder framework can be understood as a processing

R E T R A C T E D
model that turns a sentence into another form but without changing its meaning into a processing model.For a sentence pair <X, Y>, given a sentence input sequence X, we can obtain the generated target Y by the model, where X and Y are composed of different word sequences each: +e encoder encodes X and transforms the input sequence into a fixed-length vector of intermediate semantic expressions C using the model f, where +e decoder uses the intermediate vector C with the generated y 1 , y 2 , • • • , y i−1 information to generate the words to be generated at time i y i , where In practical applications, in the field of translation, such as English to Chinese, the input is an English sentence and the output is a Chinese sentence, while for speech recognition, the input is a speech signal and the output is text.
+e encoder-decoder framework is widely used, but it uses a semantic encoding vector C that is always constant when generating words in the target sentence, and the vector C is generated by encoding the words in sentence X through the encoder, so it has the same effect on generating words in the target sentence Y no matter what words are generated.As a result, the framework has little impact on the input short sequence data, but for long sequence data, the semantic information is always represented by a unique intermediate semantic encoding vector C, which makes the detailed information easily lost.
In order to solve the above problem, the attention model is introduced, and the attention model is changed from the original fixed middle semantic vector C to C i , which can be dynamically changed according to the current output words, and the encoder-decoder framework with the added attention model is shown in Figure 3.
At this point, the word y i � g(C, y 1 , y 2 , • • • , y i−1 ) is generated for time i, where +e length of the input sequence is T x , a ij indicates the attention weight distribution of the input jth word at the output ith word, and h j indicates the semantic encoding of the input jth word.+e probability distribution of the input word y i can be compared with the node state x 1 , x 2 , x 3 of the RNN implicit layer at the output i − 1 moment and the node state s i−1 of the RNN implicit layer for each input word; i.e., the function h j is used to obtain the alignment between the target word and the input word.+e alignment of the target word F(h j , s i−1 ) and the input word is finally normalized by the softmax activation function to obtain the value of the attention allocation probability distribution.+e above is the common method for calculating the probability of attention allocation for most attention models, where the definition of the F-function may differ for different attention models, as shown in Figure 4.
+e traditional encoder-decoder model based on the attention mechanism has only one input, and in literary generation studies, in addition to the subject keywords, there is also historically generated content as input.Here we refer to the design of the PPG model and modify it based on the encoder-decoder model based on the attention mechanism in order to realize both keywords and already generated poems as poetry input.

A Comparative Guide of Traditional Chinese and Western Literature
In  Mathematical Problems in Engineering

R E T R A C T E D
+ird, to distinguish which myths were influenced by Buddhism.
He argues that if Chinese mythological materials are studied according to these three principles, then myths that express the primitive beliefs and living conditions of the Chinese people will come to the fore.+ese three principles, especially the last two, emphasize the diversity of literature in the comparative study of Chinese mythology and foreign mythology.Qian Zhongshu , who, like Zhu Guangqian and Liang Zongdai, had the same experience of studying abroad and the same solid training in Chinese and Western studies, also stepped into the field of comparative literature in his own way.
Like Zhu Guangqian and Liang Zongdai, his research interests were mainly in the field of Chinese and Western poetry, but unlike Zhu Guangqian's logical and systematic research, Qian was more inclined to make comparisons between the West and the East in the manner of SU.g essays, somewhat similar to Liang Zongdai in this respect.However, he was an extremely documentary researcher who was not tired of citing literature, in contrast to Liang Zongtai's straightforward and brisk style of criticism.In this period, apart from a few articles in Chinese and English, such as "A Feature of Chinese Inherent Literary Criticism" (Literary Magazine, vol. 1, no. 4, 1937), Qian's main work in academic research was a collection of essays and journals centered on the evaluation of classical Chinese poetry, "Tangyilu" [18].
+e book was written between 1939 and 1942 and published by Kaiming Bookstore in 1948.+e entire text is written in Mandarin, and the text is rather archaic, with some quotations in English interspersed in between, making the quotations seem piled up, complicated, and trivial, and not easy to read for nonspecialist readers.+ere are ninetyone journals in +e Book of Talks and Art, and eighteen "addenda" were added later.It is a collection of fragmentary texts with no logical connection between chapters, few terms and concepts, no theoretical propositions, and no systematic construction.+e structure is loose, the content is mixed, and the text is free and loose, all of which make "Tal-Yi Lu" resemble traditional Chinese poetic discourse.After the modern era, this traditional poetic style of writing and research is almost unheard of.However, this style of writing seems to be very suitable for Qian Zhongshu's academic personality, which does not follow the fashionable trend and does not follow the crowd, and which is free and spontaneous, so it seems to be comfortable to use.More importantly, the "art of speaking" is not a poetic discourse, as it contains a large amount of indirect material from foreign literature and culture while extensively citing material from ancient and modern China, composing a special structure.Discourse text: It is for this reason that those who study comparative literature have reason to regard it as the fruit of comparative literature.When Qian Zhongshu talks about the topic of Chinese cuisine, he must use similar or relative multimaterial from the West as supporting evidence or circumstantial evidence, in order to emphasize his overall view of world literature, as he says in the preface: "+e East and the West are the same in psychology; the South and the North are the same in learning, but not in the art of Taoism."

Experiment
Due to the special structure of literature and its metricality, it is usually difficult to conduct a simple quantitative evaluation of literature generation models.In this paper, automatic evaluation, manual evaluation, and Turing test are designed to verify the validity of the model.+e following is a description of the relevant evaluation methods in this paper.

R E T R A C T E D
5.1.BLEU Automated Evaluation.BLEU (bilingual evaluation understudy) is a common evaluation algorithm for machine translation, which is used to define the similarity between machine translation and reference translation.
When the machine translation is closer to the human reference translation, the translation quality is higher.BLEU uses the N-gram matching rule to compare the similarity ratio of n words between machine translation and reference translation.
If the human reference translation is represented by reference and the machine translation is represented by candidate, the earliest BLEU algorithm can be expressed as follows: BLEU � the number of words in the machine translation that appear in the reference translation total number of words in machine translation .(7) Example: Original text: I ate an apple today Machine translation: I eat an apple today Reference translation: I ate an apple today If there are 5 words in the machine translation, and 3 words are the same as the reference translation, the BLEU value is 3/5.
As shown in Figure 5, because the literature generation model proposed in this paper is based the previous sentence history to generate poems and subthemes for generation, its generation process is to some extent similar to machine translation because literature generation is unlikely to produce poems that are exactly the same as the training poetry prediction, which is not the working principle of the objective function; BLEU can only roughly calibrate the generation ability of poems, reflecting the similarity of the generated poems to the training prediction, but the results are also instructive and can reflect the fluency of the generated poems to some extent.+erefore, for the evaluation of the literary generation model, the BLEU algorithm was borrowed for automatic evaluation [18,19].In this method, BLEU takes a value between 0 and 1, and its value can only be 1 when the translation is exactly the same as the reference translation.Hence, a larger BLEU value indicates a better translation model.When the generated verses are close to the reference next lines of the training corpus, then a larger BLEU value is considered a better poetry generation.

Manual
Assessment. +e quality of poetry production is mainly assessed manually, usually based on four evaluation criteria: fluency, coherence, metricality, and meaning [6,10,11].+e existing assessment of the quality of poetry generation is mainly conducted manually, usually based on four evaluation criteria: fluency, coherence, metricality, and meaning [6,10,11], which mainly consider the fluency, coherence, metricality, and meaning of the generated poems, i.e., whether the metricality is qualified, whether the verses are fluent, whether the verses are coherently related, and whether the verses have content and meaning.Combined with studies related to the literary aspects of literature, it is found that the evaluation of literature from these four criteria only is one-sided, as shown in Figure 6 the probability of the distribution of literary vocabulary; the above four criteria can be said to be the most basic requirements for generating sentences that can be called poetry, and a qualified poem, in addition to the basic requirements such as semantic coherence, flatness, and rhyme, the context/ emotion of the poem is more important, and contextual interplay is the most appropriate criterion for evaluating poetry +e most appropriate criterion for evaluating a poem is the blending of moods and emotions [5].+erefore, in this paper, based on the existing manual evaluation criteria for poetry, we add two evaluation indicators of mood/emotion and tangency +e improved manual evaluation criteria are shown in Table 1.
In the specific manual evaluation process, the evaluator was required to have a professional background in poetry, and then, according to the above table, he or she looked at "fluency," "coherence," "metricality," "meaning," "mood/ emotion," and "tangibility" respectively."meaning," "mood/ emotion," and "relevance."+e final score of the poetry generation model is then taken as the average value of the evaluation scores.+en, the average score of the evaluation is taken as the final score of the poetry generation model, which is used to evaluate the goodness of the generated literature.

Turing
Test. +e concept of the Turing test was introduced by Alan Matheson Turing in his 1950 article "Computers and Intelligence" [1].Turing used the test as a criterion for determining whether a computer possessed normal human intelligence.Turing's test is administered by isolating a human from a computer and then asking the participant questions to determine which one is the correct one.If the error rate exceeds 30%, the computer passes the test and has human intelligence, and this test is called the Turing test [2].+e effect of different vocabulary clustering is shown in Figure 7, so in the literary generation study of this paper, if it passes the Turing test, it can prove that the literary generation model proposed in this paper reaches the level of ordinary human creativity.+is experiment has feasibility and reference value for assessing the good and bad literary generation.
Firstly, by extracting any number of human-created literature for theme generalization, the computer-generated poems and the human-created poems are composed into test cards, randomly labeled as Literature I and Literature II, and the evaluator makes judgments on the literature in the cards Mathematical Problems in Engineering

R E T R A C T E D
in turn.As shown in Figures 2-7, the test participants were asked to make judgments from the following three options: A. Literature 1 was written by humans; B. Literature 2 was written by humans; C. Both could not be judged.When the test results were tallied, if the computer-generated literary poem was judged to have been composed by humans, it was recorded as incorrect; otherwise, it was recorded as correct.
By performing statistics on the test cards, the final judgment of the correct rate, the error rate, and the inability to judge rate were obtained to perform Turing test result analysis.

Conclusions
In this paper, we first describe the deep learning problem of literature generation, in general; then, we describe the rulebased and template-based generation approach, the statistical machine learning-based approach, and the deep learning-based approach; Rule-and template-based generation methods, statistical machine learning-based methods, and deep learning-based methods; then the three phases.+e evaluation system of literary generation in this study is introduced, and this study will be conducted from automatic evaluation, manual evaluation, and Turing test.
+is study evaluates the model from three aspects: automatic evaluation, manual evaluation, and Turing test.Table 1: Manual assessment criteria details.

Evaluation criterion Explain Fluency
Are the lines that generate poetry natural and smooth?Continuity Are the sentences that generate poetry related?Metricity Do the generated poems meet the metrical requirements of ancient poetry?Significance Does the generated poetry really express the content and specific practical significance?Artistic conception/emotion Does the generated poetry have emotion and poetic artistic conception?To the point Does the generated poetry fit the writing theme?
Mathematical Problems in Engineering

( 1 )
Login to the Baidu translation open platform (http:// api.fanyi.baidu.com)with a Baidu account.(2) Register as a developer, obtain the APPID of universal translation, and perform developer authentication to open the universal translation API service and open the link.(3) Call the translation API through the HTTP interface,

Figure 1 :
Figure 1: +e overall flow chart of literature generation.

Figure 4 :
Figure 4: Schematic diagram of the calculation of attention allocation probability.

Figure 6 :
Figure 6: Distribution probability of literary terms.

Figure 7 :
Figure 7: Effect of different vocabulary clustering.
3.1.LiteraryGenerationProblemDescription.+estudy of literature generation in this paper is based on deep learning techniques, and its overall process is shown in Figure1.3.2.Design of the Literature Generation Model Based on Keyword Transformation Extension.From the above studies of literature generation-related work, we can find that the generation of poetry in literature and other genres generally faces two problems: (1) thematic drift.+at is, in the process of literary generation, as the length of the generated verses increases, the subsequent generated verses appear to deviate from the theme of the writing; (2) semantic incoherence, i.e., the generated verses have semantic incoherence, such as vernacular language and incoherence in the verses.+is study addresses the above problems.Based on the existing literary generation methods, we propose a literary poem generation model based on keyword transformation extension.Based on the existing literature generation methods, we propose a Keyword Transformation and Expansion