Abstract
Literary therapy theory provides a new field for contemporary literature research and is of greater significance for maintaining the physical and mental health of modern people. The quantitative evaluation of psychotherapy effects in Japanese healing literature is a hot research topic at present. In this study, a text convolutional neural network (Text-CNN) was selected to extract psychological therapy features with different levels of granularity by using multiple convolutional kernels of different sizes. Bidirectional threshold regression neural network (BiGRU) can characterize the relationship between literature research and the psychotherapy effect. On the basis of the CNN-BilSTM model, a parallel hybrid network integrated with the attention mechanism was constructed to analyze the correlation between literature and psychotherapy. Through experimental verification, the model in this study further improves the accuracy of correlation classification and has strong adaptability.
Introduction
The concept of "literary therapy"—the idea that reading specific types of literature can maintain or improve mental health—is well-established in the humanities. However, evaluating these effects has traditionally been a subjective, qualitative endeavor. This paper, published in Frontiers in Psychology, explores the intersection of computational linguistics and literary therapy by proposing a deep learning architecture designed to quantitatively analyze the therapeutic potential of Japanese "healing literature."
Methodology
The authors introduce a Parallel Hybrid Network that processes text through two specialized branches simultaneously. This parallel approach ensures that the unique feature extraction capabilities of both CNNs and RNNs are preserved without one filtering the other too early.
Parallel Hybrid Architecture
- Text-CNN Layer: This branch acts like a set of magnifying glasses of different sizes. By using multiple convolutional kernels (sizes 3, 4, and 5), the model extracts features at different levels of granularity—essentially "reading" short, medium, and long phrases to find emotional triggers.
- BiGRU Layer (Bidirectional Gated Recurrent Unit): While the CNN looks at local patterns, the BiGRU looks at the flow of the sentence. It processes text in both directions (forward and backward) to capture long-distance dependencies. The authors chose BiGRU over the more common BiLSTM because it is computationally simpler and faster while maintaining similar performance for this specific task.
Attention Mechanism
The attention mechanism acts as the "judge" of the model. It takes the outputs from both the CNN and BiGRU and assigns "weights" to specific words. If a word carries significant psychological weight, the attention mechanism ensures the model prioritizes it for the final classification.
Experimental Validation
While the paper’s motivation is Japanese literature, the authors validated their architecture using two gold-standard English-language sentiment analysis benchmarks: the IMDB dataset (50,000 movie reviews) and the SST-2 (Stanford Sentiment Treebank). This allowed for a direct comparison against established baselines in the NLP field.
Key Results
The hybrid model (Text-CNN + BiGRU + Attention) consistently outperformed standalone models and simpler combinations:
- Accuracy: Reached 91.46% on the IMDB dataset and 89.48% on SST-2.
- Comparison: It outperformed the standard Text-CNN by roughly 4% and the BiLSTM by about 1.4% in accuracy.
- Stability: The model showed a more stable learning curve (lower loss rates over fewer epochs) compared to non-hybrid versions, suggesting it is more "adaptable" to different types of literary text.
Conclusion
By integrating Text-CNN and BiGRU in parallel with an attention mechanism, this research provides a robust "diagnostic" algorithm capable of handling the complex emotional layers found in literature. Future work will focus on cross-language validation with native Japanese corpora to account for unique linguistic features like kanji and honorifics, as well as testing the model's predictions against actual patient outcomes in clinical settings.
Summary updated