Search

The paper industry is a high-carbon emission and energy-intensive industry. From the perspective of low-carbon trade and carbon neutrality, its energy conservation and emission reduction are worthy of attention. This study used the input-output model to calculate the embodied carbon emissions of China’s paper industry’s export trade to the United States from 2006 to 2020 and used the logarithmic mean division index (LMDI) method to analyze influencing factors of the change of embodied carbon emissions. The study found that the embodied carbon emissions of China’s paper industry’s export trade to the United States generally shows a stable downward trend after reaching the peak with the increase of export trade scale; scale effect is the main factor that causes the embodied carbon emissions, while technological progress, policy support, and environmental regulations are important driving forces to promote carbon emission reduction. The research results of this paper not only can test and guide China’s paper industry trade policies and industrial policies, but they can also provide decision-making reference for China and the United States to promote the carbon emission reduction of the paper industry.

Textual network embedding aims to learn meaningful low-dimensional representations for vertices with the consideration of the associated texts. When learning the representations for texts in network embedding, existing methods mostly only exploit information from neighboring texts (i.e., contexts), while rarely taking advantages of the valuable network topological (structural) information. To bridge the gap, in this paper, a model based on adaptive-filter convolutional neural networks (CNN) is developed, in which the filters are adapted to local network topologies, rather than clamped to fixed values as in traditional CNNs. The dependency enables the learned text representations to be aware of local network topologies. It is shown that the proposed topology-aware representations can be viewed as a complement to existing context-aware ones. When the two are used together, experimental results on three real-world benchmarks demonstrate that significant performance improvements on the tasks of link prediction and vertex classification.

Network embedding has recently emerged as a promising technique to embed nodes of a network into low-dimensional vectors. While fairly successful, most existing works focus on the embedding techniques for static networks. But in practice, there are many networks that are evolving over time and hence are dynamic, e.g., the social networks. To address this issue, a high-order spatio-temporal embedding model is developed to track the evolutions of dynamic networks. Specifically, an activeness-aware neighborhood embedding method is first proposed to extract the high-order neighborhood information at each given timestamp. Then, an embedding prediction framework is further developed to capture the temporal correlations, in which the attention mechanism is employed instead of recurrent neural networks (RNNs) for its efficiency in computing and flexibility in modeling. Extensive experiments are conducted on four real-world datasets from three different areas. It is shown that the proposed method outperforms all the baselines by a substantial margin for the tasks of dynamic link prediction and node classification, which demonstrates the effectiveness of the proposed methods on tracking the evolutions of dynamic networks., Accepted by COLING 2020 : The 28th International Conference on Computational Linguistics

Deepfake detection, the task of automatically discriminating machine-generated text, is increasingly critical with recent advances in natural language generative models. Existing approaches to deepfake detection typically represent documents with coarse-grained representations. However, they struggle to capture factual structures of documents, which is a discriminative factor between machine-generated and human-written text according to our statistical analysis. To address this, we propose a graph-based model that utilizes the factual structure of a document for deepfake detection of text. Our approach represents the factual structure of a given document as an entity graph, which is further utilized to learn sentence representations with a graph neural network. Sentence representations are then composed to a document representation for making predictions, where consistent relations between neighboring sentences are sequentially modeled. Results of experiments on two public deepfake datasets show that our approach significantly improves strong base models built with RoBERTa. Model analysis further indicates that our model can distinguish the difference in the factual structure between machine-generated text and human-written text., Comment: EMNLP2020;10 pages

Thermal environment has many influences on people who use outdoor space. The thermal environment of university campus is an important part of educational living space. In campus planning and design, the impact of landscape design on the thermal environment should be considered. This paper measures the thermal environment of the campus of Xi’an Eurasia University and conducts questionnaire survey to obtain the Library and Nanshan Road, East Gate Square and Parking lot 4 different underlying surface characteristics of the impact on the thermal environment. The results show that the SET* correlation of different underlying surfaces: Library> East Gate Square> Nanshan Road > Parking lot. Shade, grassland, and water bodies can effectively reduce the average radiant temperature. Among them, the humidification and cooling effects of the shade are the best, and the cooling ability of the grass is second only to the shade. In the process of campus environment design, reasonable adjustment of the relationship between the shade of the grass and other underlying surfaces can effectively improve the outdoor thermal comfort of the campus in winter and create a comfortable outdoor activity place on the campus.

Textual network embeddings aim to learn a low-dimensional representation for every node in the network so that both the structural and textual information from the networks can be well preserved in the representations. Traditionally, the structural and textual embeddings were learned by models that rarely take the mutual influences between them into account. In this paper, a deep neural architecture is proposed to effectively fuse the two kinds of informations into one representation. The novelties of the proposed architecture are manifested in the aspects of a newly defined objective function, the complementary information fusion method for structural and textual features, and the mutual gate mechanism for textual feature extraction. Experimental results show that the proposed model outperforms the comparing methods on all three datasets., Comment: To appear at EMNLP-IJCNLP 2019 (Conference on Empirical Methods in Natural Language Processing & International Joint Conference on Natural Language Processing 2019)

Traditional methods of annotating the sentiment of a document are based on sentiment lexicons, which have been proven quite efficient. However, such methods ignore the effect of supplementary features (e.g., negation and intensity words), while only consider the counts of positive and negative words, the sum of strengths, or the maximum sentiment score over the whole document primarily. In this paper, we propose to use convolutional neural network (CNN) and long short-term memory network (LSTM) to model the role of negation and intensity words, so as to address the limitations of lexicon-based methods. Results show that our model can not only successfully capture the effect of negation and intensity words, but also achieve significant improvements over state-of-the-art deep neural network baselines without supplementary features.

Sentiment analysis has important applications in many areas, including marketing, recommendation, and financial analysis. Since topic modeling can discover hidden semantic structures, researchers put forward sentiment analysis models based on topic models. These models have been successfully applied on long texts, but analysis for short text is a challenging task because of the sparsity of features in short texts. We observe that the textual context has been widely considered on text analysis task, but on sentiment analysis area, most sentiment analysis models still lack of consideration and integration of sentimental context. Thus, by taking the speciality of sentiment analysis task and short text into consideration, we propose the sentimental context to enrich the characteristics and improve the performance of sentiment classification over short text. We first put forward the concept of sentimental context, which is extracted from the text body and sentiment lexicon, and then we integrate the sentimental context and propose two sentiment classification models based on word-level and topic-level respectively. We present results on real-world datasets from various sources, validating the effectiveness of the proposed models.

Low dispersity star-like polyethylene with a “silica” core was synthesized via a simple yet efficient sol–gel process using trimethoxysilane-terminated polyethylene, which was prepared using radical-mediated thiol-ene click chemistry between low molecular weight vinyl-terminated polyethylene and (3-mercaptopropyl)trimethoxysilane.