This study aims to construct a comprehensive feature system for identifying artificial intelligence–generated content (AIGC) in online Q&A communities, thus uncovering the key factors and mechanisms influencing the identification of AIGC. First, based on the theory of systemic functional linguistics (SFL) and information quality (IQ), this article extracts vocabulary, content, structure, and emotional features from the text, and identifies the AIGC through nine mainstream machine learning algorithms. Subsequently, three widely used resampling strategies are exploited to address the category imbalance problem. The grid search optimisation algorithm fine-tunes different combinations of parameters to improve the performance of the identification classifier. Finally, SHAP values are introduced to evaluate and elucidate the global feature importance and feature influence mechanism. A Chinese corpus from the Zhihu Q&A community is constructed to verify the validity of these methods. The experimental results show that the eXtreme Gradient Boosting (XGBoost) model optimised with hybrid sampling and grid search parameters exhibits excellent performance in identifying AI-generated text, which achieves an F 1 -score of 0.9935, an improvement of 0.11 percentage points over the original model. In addition, all four dimensions of features constructed in this article contribute to AI-generated text identification, and the results of feature interpretability analysis show the greatest impact of features that focus on content readability. The study facilitates the identification and labelling of AIGC in online Q&A communities, thereby enhancing transparency and accountability of information shared online.
To detect multiple coexisting emotions from public emergency opinions, this article proposes a novel two-stage multiple coexisting emotion-detection model. First, the text semantic feature extracted through bidirectional encoder representation from transformers (BERT) and the emotion lexicon feature extracted through the emotion dictionary are fused. Then, the emotion subjectivity judgement and multiple coexisting emotion detection are performed in two separate stages. In the first stage, we introduce synthetic minority oversampling technique (SMOTE) to enhance the balance of data distribution and select the optimal classifier to recognise opinion texts with emotion. In the second stage, the label powerset (LP)-SMOTE is proposed to increase the number of the minority category samples, and multichannel emotion classifiers and the decision mechanism are employed to recognise different types of emotions and determine the final coexisting emotion labels. Finally, the Weibo data about coronavirus disease 2019 (COVID-19) are collected to verify the effectiveness of the proposed model. Experiment results indicate that the proposed model outperforms state-of-the-art models, with the F1_macro of 0.8532, the F1_micro of 0.8333, and the hamming loss of 0.0476. The emotion detection results are conducive to decision-making for public emergency departments.
This study characterized wood liquefaction by the fractal geometry method. Chinese fir ( Cunninghamia lanceolata Hook.) fine powderwas liquefied under various conditions of phenol-to-wood ratio (3:1, 4:1, and 5:1), catalyst content (4, 6, and 8%), and temperature (130, 150, and 170°C). The surface fractal dimension of liquefiedwood residueswas determined at different liquefaction time levels (30, 60, 90, 120, 150, and 180min) by software based on the cubic coveringmethod. The relationship between fractal dimension and residue content was examined quantitatively. Results indicated that 1) surface fractal dimensions of liquefied wood residues were between 2.27 and 2.30 under all liquefaction conditions; 2) surface fractal dimension was inversely related to liquefaction time, and it decreased faster at early liquefaction stages; 3) surface fractal dimension was inversely related to phenol-to-wood ratio, catalyst content, and liquefaction temperature; and 4) the relationship between surface fractal dimension and residue content could be described by a linear function with high R 2 values. This study provides a new alternative to the arsenal of wood liquefaction characterization methods and also sheds light on some fundamental aspects of wood liquefaction research.
Murals unearthed in China have outstanding regional characteristics and one of the largest period spans in scale and variety. To explore the visual distinguishability and topic autocorrelation of murals unearthed in China from the spatial perspective, multiple classification models are employed to classify murals unearthed in China through visual features. Then, the k-means is employed to mine topics, and they are analysed through topic intensities (TIs), Moran’s Index (MI) and spatial topic concentration degrees (STCDs). In addition, the characteristics of topic distribution and evolution are summarised and revealed in the spatial dimension. From a spatial perspective, it verifies the distinguishability of visual features of murals through ViT_BOW_GNB, and the precision of this model is 98.17%. Thirteen topics are clustered through k-means, and the distribution of mural topics is spatial autocorrelation according to MI. Besides, the topic evolves from the political centre to the surrounding area, and the topics with high intensities are highly concentrated in spatial. This study reveals the spatial characteristics of the mural at the level of visual features and semantics, which facilitates the digital management, conservation and knowledge discovery of cultural heritage resources.
Murals are important resources for carrying cultural heritage, historical evidence and artistic memory. The sentiment of a mural is the transmission of its inner thoughts, closely related to the region and dynasty to which the mural belongs. To explore the sentiment evolution patterns of temple murals, we construct a spatio-temporal evolution analysis framework based on sentiment recognition. This framework mainly includes feature extraction, sentiment recognition and sentiment evolution analysis. First, we extract the colour features, local features, global semantic features, patch features and structure relations to represent the visual features of temple murals. Second, the semantics of spatio-temporal attributes and titles of murals are extracted through the fine-tuned BERT (Bidirectional Encoder Representations from Transformers) to enhance the feature discrimination for sentiment recognition. Third, we introduce the SMOTE (Synthetic Minority Oversampling Technique) to reduce the influence of imbalanced data and select RF (random forest) as the optimal classifier. The F1 score of the fine-grained sentiment recognition model is up to 81.37%. Finally, we collect the temple murals and reveal the characteristics and patterns of sentiment evolution from the spatial, temporal and spatio-temporal perspectives.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)