Introduction

Computational literary studies have examined the features of specific genres (Underwood 2019(Underwood 2019, Piper 2018, Long et al. 2018, Jockers 2013, Moretti 2013). In Chinese studies, genre analysis mainly discusses premodern materials (Chen et al. 2023, Clifford 2018, Liu et al. 2025, Hou & Zhang 2024, Broadwell et al. 2019). However, most genre‑focused Digital Humanities (DH) studies still focus on canonical texts, and quantitative studies of born‑digital literary genres such as fan fiction and web novels are still emerging (Chen et al. 2023, Clifford 2018, Liu et al. 2025, Hou & Zhang 2024, Broadwell et al. 2019). However, most genre‑focused Digital Humanities (DH) studies still focus on canonical texts, and quantitative studies of born‑digital literary genres such as fan fiction and web novels are still emerging (Chen & Yan 2025, Johnson et al. 2025, Lei 2023, Nguyen et al. 2024, Pianzola et al. 2020, Yu & Pianzola 2024, Zhan 2019). (Chen & Yan 2025, Johnson et al. 2025, Lei 2023, Nguyen et al. 2024, Pianzola et al. 2020, Yu & Pianzola 2024, Zhan 2019).

Addressing this gap, this study centers Chinese web novels as a distinct born‑digital genre. Chinese web novels are commercial serialized fiction circulated on online platforms (Wu 2023).(Wu 2023).¹ They are divided into nanpin (male‑oriented) and nüpin (female‑oriented) channels. Historically, male-oriented novels concentrated on science fiction and action, while female-oriented novels centered around romance (Xiao 2024). However, female-oriented novels are increasingly challenging this stereotype as female‑oriented science fiction, especially cyberpunk, gains more visibility (Xiao 2024). However, female-oriented novels are increasingly challenging this stereotype as female‑oriented science fiction, especially cyberpunk, gains more visibility (SFW 2024, Bao 2023, Xiao 2023, 2022)(SFW 2024, Bao 2023, Xiao 2023, 2022). Nevertheless, related DH analyses remain scarce.

This study uses computational methods to investigate female‑oriented cyberpunk² web novels, comparing them with male‑oriented counterparts. Using supervised classification, BERTopic, and emotional intensity analysis, I identify three female‑oriented cyberpunk features: 1) multi‑genre fusion without reliance on a single IP (Intellectual Property); 2) more realist concerns; and 3) higher overall emotional intensity.

This study challenges the assumptions that cyberpunk is male‑centered and that female‑oriented work is reducible to romance, providing empirical support for understanding the diversity of Chinese science‑fiction web novels. More broadly, this project shows that working at the micro‑scale using DH tools will help track how everyday sociotechnical concerns circulate through web novels and into the wider public sphere. It also aligns with DH2026’s theme “Interpretation with Small Data,” showing that a small, well‑curated corpus can yield interpretable cultural claims and foreground Chinese contemporary literature and female‑oriented science fiction as underrepresented domains in DH studies.

Data and Workflow

This study draws on two web‑fiction platforms: Jinjiang (n.d. female‑oriented) and Qidian (n.d. female‑oriented) and Qidian (n.d. male‑oriented), and collects metadata(n.d. male‑oriented), and collects metadata³ and partial texts⁴ (See Table 1). Figure 1 represents the overall workflow of this study.

Table 1

Data Sources and Preprocessing

Figure 1

Overall Workflow

Supervised Classification Model

Figure 2

Top 30 Predictive Features of the Classification Model for Jinjiang/Qidian Cyberpunk Novel Summaries

Figure 2 represents the top coefficient features used to classify male- and female-oriented cyberpunk web novels from summaries.⁵ In female-oriented features, terms such as “game,” “game-instance,” and “guide” represent general game vocabulary. In male-oriented features, “Night City,” “Legend,” and “Cyberpunk: 2077” mark reliance on the video-game IP Cyberpunk: 2077.⁶ This shows that female‑oriented cyberpunk merges the game genre as flexible scaffolding, whereas male‑oriented novels are heavily reliant on Cyberpunk: 2077, showing less diversity and originality.

Topic Modeling

Table 2

Top 30 BERTopic Themes of Jinjiang/Qidian Cyberpunk Novel Summaries (Translated)

Building on previous findings, I use BERTopic to further differentiate male- and female-oriented novels’ topics (see Table 2).⁷ On the female‑oriented side (left), topics that discuss campus lives (Topic 21), live‑stream (Topic 28), and everyday lives (Topic 23) indicate the fusion of different genres and richer realist concerns. Also, female-oriented cyberpunk foregrounds relational and affective lexicon, with family and other relationships ((Topics 13, 24, 26) signaling stronger emotional expression. By contrast, male‑oriented themes (right) include words from the IP Cyberpunk: 2077 (Topics 0, 1, 6, 21), alongside conflict-related expressions (Topics 5, 20), confirming single‑IP anchoring and a focus on externally driven conflict.

Emotional Intensity

Figure 3

Emotional Intensity Distribution for Partial Texts from Jinjiang/Qidian

Building on BERTopic’s evidence of more emotional expressions in female‑oriented novels, I measure emotional intensity in partial texts using the NRC‑VAD lexicon (Mohammad 2025).(Mohammad 2025).⁸

Figure 3 shows that male‑oriented works have lower intensity values (p < 0.05) while female‑oriented works are distributed into the higher‑intensity range.⁹ These results illustrate that female‑oriented cyberpunk not only thematically emphasizes relational ties, but also deploys emotion more strongly in textual expression.

Conclusion and Prospects

Compared to male-oriented novels, female-oriented cyberpunk web novels show more thematic diversity, combining technological imagination with profound realist concerns with a focus on characters’ emotions. Going forward, I will use close reading to verify and complicate these findings, and I hope this study invites wider scholarly focus on Chinese web novels and catalyzes more DH work on born‑digital genres.

See Also

Table 3

Top 30 BERTopic Themes of Jinjiang/Qidian Cyberpunk Novel Summaries (Chinese)

Notes

1 Chinese web novels feature rapid serialization, interactive reader engagement, and a VIP paid system that monetizes advanced chapters. They have become a major cultural and economic force in China, with TV adaptations achieving global success such as Love Between Fairy and Devil (苍兰诀) and The Untamed (陈情令).

2 Cyberpunk is a science fiction genre that depicts near‑future worlds where pervasive computing, AI, and cybernetics intersect with corporate control, urban precarity, and marginalized lives, often summarized as “high tech, low life.”

3 Metadata comprises title, author, first‑publication date, ranking, summary, and author‑applied tags. It includes 791 Jinjiang works tagged “Cyberpunk” and 989 Qidian works retrieved with the keyword “赛博” (Cyber) in titles/summaries.

4 For partial texts, this study retains works whose free-of-charge chapters exceed 30,000 Chinese characters and analyzes the first 30,000 characters per work, yielding 282 female‑oriented and 482 male‑oriented texts.

5 In literary studies, supervised classification models have been used to distinguish the features of genres and author styles (Sharmaa et al. 2020, Underwood 2016, Olsen 2005, Koppel et al. 2002). Building on prior study, my classification model uses TF‑IDF (term frequency–inverse document frequency) to vectorize Chinese texts into numerical features, segments words with the jieba tokenizer, and augments the vocabulary with cyberpunk domain terms (e.g., “Cyberpunk: 2077”) to improve segmentation accuracy; it filters stop words, punctuation, and platform‑specific terms, and extracts 5,000 features. Logistic Regression serves as the classifier in a binary setup (0 = Qidian, 1 = Jinjiang), using an 80/20 train–test split. The linear classifier learns the mapping between textual features and platform labels, and coefficient analysis identifies the features that contribute most to classification, thereby quantifying stylistic differences between male and female oriented cyberpunk web novels. A null‑hypothesis baseline accuracy of 51.67% confirms the substantive validity of the original model’s around 84% accuracy.(Sharmaa et al. 2020, Underwood 2016, Olsen 2005, Koppel et al. 2002). Building on prior study, my classification model uses TF‑IDF (term frequency–inverse document frequency) to vectorize Chinese texts into numerical features, segments words with the jieba tokenizer, and augments the vocabulary with cyberpunk domain terms (e.g., “Cyberpunk: 2077”) to improve segmentation accuracy; it filters stop words, punctuation, and platform‑specific terms, and extracts 5,000 features. Logistic Regression serves as the classifier in a binary setup (0 = Qidian, 1 = Jinjiang), using an 80/20 train–test split. The linear classifier learns the mapping between textual features and platform labels, and coefficient analysis identifies the features that contribute most to classification, thereby quantifying stylistic differences between male and female oriented cyberpunk web novels. A null‑hypothesis baseline accuracy of 51.67% confirms the substantive validity of the original model’s around 84% accuracy.

6 Cyberpunk: 2077 is an open-world role-playing video game developed by CD Projekt Red released in 2020. Set in the dystopian Night City, it explores themes of transhumanism, corporate power, and cybernetic augmentation through player-driven narratives.

7 Literary scholars use topic modeling to understand topics across languages and genres (Mengyuan 2024, Goldstone & Underwood 2012, Blei 2012). In my study, I use BERTopic (Mengyuan 2024, Goldstone & Underwood 2012, Blei 2012). In my study, I use BERTopic (Grootendorst 2022) for topic modeling. It is an advanced Transformer‑based model that uses sentence embeddings to capture deep semantic relations in text and clusters semantically similar passages into topics. I use novel summaries rather than full texts for topic modeling because summaries function as condensed authorial abstractions of a work’s central themes and genre positioning. In this analysis, the workflow first splits each novel summary into individual sentences as the units of analysis. Preprocessing includes jieba for Chinese tokenization; for embeddings I use the paraphrase‑multilingual‑MiniLM‑L12‑v2 sentence model to handle Chinese, and in BERTopic I set language= “Chinese (simplified)” to optimize keyword extraction. Because Chinese cyberpunk web novels are heavily code‑mixed, with many core concepts appearing in English, I deliberately did not filter English tokens. A multilingual sentence embedding model allows these elements to be preserved. The pipeline is applied separately to the Jinjiang and Qidian corpora and compare topical differences. I choose sentences as units rather than whole summaries because many single sentences already constitute independent thematic units; analyzing them separately allows us to more precisely capture and disentangle the multiple themes contained within a summary. Please see Table 3 for original Chinese topics and keywords.(Grootendorst 2022) for topic modeling. It is an advanced Transformer‑based model that uses sentence embeddings to capture deep semantic relations in text and clusters semantically similar passages into topics. I use novel summaries rather than full texts for topic modeling because summaries function as condensed authorial abstractions of a work’s central themes and genre positioning. In this analysis, the workflow first splits each novel summary into individual sentences as the units of analysis. Preprocessing includes jieba for Chinese tokenization; for embeddings I use the paraphrase‑multilingual‑MiniLM‑L12‑v2 sentence model to handle Chinese, and in BERTopic I set language= “Chinese (simplified)” to optimize keyword extraction. Because Chinese cyberpunk web novels are heavily code‑mixed, with many core concepts appearing in English, I deliberately did not filter English tokens. A multilingual sentence embedding model allows these elements to be preserved. The pipeline is applied separately to the Jinjiang and Qidian corpora and compare topical differences. I choose sentences as units rather than whole summaries because many single sentences already constitute independent thematic units; analyzing them separately allows us to more precisely capture and disentangle the multiple themes contained within a summary. Please see Table 3 for original Chinese topics and keywords.

8 The lexicon is built on the three‑dimensional VAD model and provides a quantitative score for each word on Valence, Arousal, and Dominance. Valence measures the positivity or negativity of affect; Arousal measures energy level from calm to excited; Dominance measures a sense of control versus being controlled. Studies have used the VAD model to examine emotions in literary texts (Neugarten 2025, Qiu 2025, Vishnubhotla et al. 2024, Yuri 2024, Pascale 2024). This study uses the Chinese version of the NRC VAD Lexicon, provided as automatic translations of the English VAD norms into 108 languages, and prior work shows that “most affective norms are broadly stable across languages” (Neugarten 2025, Qiu 2025, Vishnubhotla et al. 2024, Yuri 2024, Pascale 2024). This study uses the Chinese version of the NRC VAD Lexicon, provided as automatic translations of the English VAD norms into 108 languages, and prior work shows that “most affective norms are broadly stable across languages” (Mohammad 2022).(Mohammad 2022).

Because my focus is the intensity of emotional expression, I primarily use Valence and Arousal, the two dimensions most directly related to intensity. My calculation is grounded in James A. Russell’s Circumplex Model of Affect (Russell 1980), which locates emotions in a two‑dimensional circular space with Valence on the horizontal axis and Arousal on the vertical axis, where the center point (0,0) represents a completely neutral state. In this space, the intensity of an emotion is the distance from the neutral origin: the farther the distance, the stronger the intensity. Concretely, I rescale the mean Valence and Arousal scores from [0, 1] to [-1, 1], aligning neutrality with the origin. Emotional intensity for a document is then computed as the Euclidean distance of the document’s average affect coordinate from the origin, as shown in Equation:(Russell 1980), which locates emotions in a two‑dimensional circular space with Valence on the horizontal axis and Arousal on the vertical axis, where the center point (0,0) represents a completely neutral state. In this space, the intensity of an emotion is the distance from the neutral origin: the farther the distance, the stronger the intensity. Concretely, I rescale the mean Valence and Arousal scores from [0, 1] to [-1, 1], aligning neutrality with the origin. Emotional intensity for a document is then computed as the Euclidean distance of the document’s average affect coordinate from the origin, as shown in Equation:

The calculated intensity values range from 0 to 1.414, with higher numbers indicating stronger overall emotional expression in the text.

9 The calculation shows that the average emotional intensity for female‑oriented (Jinjiang) works is 0.1321, while male‑oriented (Qidian) works average 0.1150. The female‑oriented mean is 14.9% higher, and this difference is statistically significant (p < 0.05).