[Objective] This paper explores the research progress of the process-level semantic representation of experimental protocols. It aims to discover the key issues to be addressed and identify development trends. [Coverage] We used related topics to retrieve the relevant literature from Web of Science, arXiv, Engineering Village, CNKI, Wanfang, and VIP. We also examined the requirements of the submission requirements and evaluation principles of renowned journals on experimental protocols. [Methods] First, we defined the concepts of experimental protocols and their semantic representation at the process-level. Then, we examined the representation methods, representation element extraction, and application of representative data. [Results] The research on process-level semantic representation is in the early development stages. The representation framework was not unified, and the elements were different. The experimental protocols were mainly written in natural language, which were difficult to extract the representation elements automatically. Some studies explored the application of process-level semantic representation data, which leaves more knowledge gaps to be filled. [Limitations] This paper does not thoroughly discuss the technical details of extracting representation elements from literature and data application methods. [Conclusions] We need to establish a unified representation framework for more complete elements by integrating various representation methods. We should also explore automatic extraction methods based on advanced intelligent technology and application using the process-level semantic representation data.

[Objective] This paper summarizes the developments of early warning research for online public opinion. [Coverage] We searched the Web of Science and CNKI with related terms such as public opinion warning, online public opinion, and public opinion risks. A total of 52 articles representing the foundations of the disciplines and the development trends were selected for a comprehensive review. [Methods] We summarized the foundations of early warning studies from the perspective of online public opinion characteristics and risk evaluations. Then, we examined the status quo of current research on early warning for online public opinion. [Results] Currently, most research focuses on expert empowerment, machine learning, communication process, and sentiment analysis. All of them can accurately predict the risk level of online public opinion, which is significant to the online environment and social stability. [Limitations] More research is needed to review early warning technology. [Conclusions] The existing research does not provide universal concepts for online public opinion. The risk evaluation method needs to be improved. We should also establish authoritative and unified standards to compare the performance of different monitoring systems.

[Objective] This paper addresses the issues of the traditional single classification method, which cannot effectively identify high-quality “bottleneck” technology patents. [Methods] We developed a multi-category polling model (LSTM-Seq-BERT) with LSTM, Word2Vec, and BERT to identify high-quality “bottleneck” patents from the application documents. Moreover, we constructed a corresponding multi-level label system for the model with IPC number as the primary classification labels and authorization status as the secondary classification labels. [Results] The accuracy of identifying high-quality “bottleneck” technology patents was increased to 88.1%. [Limitations] We only utilized patents from the Hongkong-Macau-Guangdong Greater Bay Area, resulting in data imbalance. [Conclusions] The proposed model can enhance the accuracy of identifying high-quality “bottleneck” technology patents and possesses practical value.

[Objective] This paper uses neural networks to detect facial expressions in real-time video streams, aiming to explore the correlation between CEO’s emotional characteristics at product launch events and media attention. [Methods] A total of 566 product launch event videos from 34 electronics companies were collected. Facial expressions of CEOs during the events were detected using models like MTCNN. Then, we investigated the patterns of CEO’s emotional expressions and explored the influence of their characteristics on media attention with correlation analysis. [Results] CEOs of different companies exhibited distinct emotional expression patterns during the launch events, which could be clustered closely associated with the main product types of the companies. Each cluster also had significant emotional inertia expression and influence trends. The proportion of anger was positively correlated with media attention during the launch events at a confidence level of 95% (with Pearson’s correlation coefficients exceeding 0.21). [Limitations] This study focuses on electronic product launch events, and the collected data from various companies were not unevenly distributed. [Conclusions] Deep learning enable the rapid detection of CEO facial expressions based on video streams. This study analyzed CEO’s emotional expression patterns and their influence and provided suggestions for CEO’s emotional management in brand communication.

[Objective] This paper aims to accurately identify technology opportunities using a knowledge network and multidimensional map of technology innovation, which will enhance the enterprises’ core competitiveness. [Methods] Firstly, we extracted technology keywords based on existing patent data and created innovation dimensions. Secondly, we constructed a knowledge network to analyze the importance of keywords and innovation dimensions. Finally, we identified technology opportunities and determined their priority with the multidimensional map of technology innovation. [Results] We examined the new method with patent data of barium sulfate preparation from titanium dioxide waste acid (from 2012 to 2021). We found that the five types of technological opportunities identified by this method can provide helpful theoretical decision-making support for enterprises to choose innovative directions. [Limitations] We only examined the new method with existing patents and technology keywords. We should have comprehensively studied technology development trends. [Conclusions] Identifying technology opportunities based on knowledge networks and multidimensional maps of technology innovation can improve the accuracy of identification results.

[Objective] This paper addresses the issues of insufficient consideration of evolution patterns and factors in the analysis of financial events evolution. It focuses on modeling and representing the evolution of financial risk events based on event correlation and evolution. This study also constructs an event evolution graph. [Methods] We combined event evolution pattern modeling to analyze evolution conditions and proposed a graph generation algorithm for event evolution based on the nearest neighbor query Ball-Tree. This algorithm enables an adequate representation of financial risk events. [Results] We analyzed the risk events related to “Evergrande Group”. We found that when the strength of event evolution relationships was set at 0.2, 489 correct evolutionary relationships were detected among all 629 event pairs with evolutionary relationships, with an accuracy rate of 77.74%. [Limitations] Due to the space limitation, identifying financial risk events was not extensively described, and the dynamic updating of financial events was not considered. [Conclusions] The proposed modeling approach can analyze various potential association relationships among events, recreate significant scenarios during the development of risk events, and provide effective technical support for understanding potential evolution paths and patterns.

[Objective] This paper develops a new deep learning model to decide the usefulness of travel reviews, which provides valuable insights for consumers and hotel managers. [Methods] We proposed a usefulness identification model based on multi-dimensional graph convolutional networks and multi-modal fusion. Then, we used BERT and MAE models to pre-train texts and images, and adopted multi-view graph convolutional networks to model multi-modal features. Third, we captured the interactive information between different modalities with the attention mechanism. Finally, we integrated text features to identify valuable reviews. [Results] We conducted comparative experiments on the Yelp dataset. The accuracy of this method reached 73.21%, which was 10% higher than the traditional single-modal and existing multi-modal models. [Limitations] This paper only explores the text and image modalities on the Yelp dataset. More research is needed to investigate other data fusion and modalities. [Conclusions] The proposed model could effectively identify helpful online reviews with multi-dimensional graph convolutional networks and multi-modal features.

[Objective] This paper aims to grasp the core content of social science academic literature accurately and improve the structure recognition effect of literature abstracts. [Methods] An experiment was conducted on the bilingual abstract data of several core periodicals in the field of library and information science by using pre-training language model, and an enhanced learning method was proposed by using domain data in the stages of pre-training, fine-tuning and model's output layer. [Results] Enhancement pre-training, fine-tuning, and fusion of bilingual sentence classification probability could improve the F1 values of abstract structure recognition by 1 to 2, 1, and 0.5 to 1 percentage point on single journal data, respectively. [Limitations] Due to limited computing resources, the field bilingual text continued pre-training and performance test were not conducted on the cross-language pre-training model. [Conclusions] This research makes full use of bilingual resources in academic literature and effectively improves the recognition effect of abstract structure, which is of certain significance to quickly understand the content of literature and promote scientific communication.

[Objective] This paper improves the Reinforced Latent Factor Model with user multi-adaptive preference feature temporal weighting, aiming to improve the accuracy of recommendations. [Methods] Building upon the Temporal Potential Factor Model, we further integrated user preferences from different periods, such as interest forgetting features, publication interest overlap, and semantic similarity of comments. The user rating matrix is weighted and decomposed based on preference weights to capture the multiple preference changes of users towards different publications at different times. [Results] We conducted comparison experiments with four baseline methods based on temporal matrix factorization with three datasets. The proposed model’s precision was 9.26% higher than TDMF, 17.35% higher than TMRevCo, 38.63% higher than BPTF, and 26.24% higher than TCMF. This demonstrates that the proposed model is more accurate in extracting user temporal features. [Limitations] The analysis of interest drift evolution depends on historical user data. When the data is too sparse, alternative user information is required for a cold start. [Conclusions] The proposed model considers user forgetting and comment evolution features, effectively capturing user temporal interest drift and reflecting the evolving relationship of users’ interest in publications. It improves the accuracy of recommendations.

[Objective] This study proposes an improved paper recommendation framework based on Ensemble Learning and Factorization Machine. It addresses the issues of the existing methods, such as difficulties in processing sparse data and representing features. [Methods] First, we used Convolutional Neural Network, Network Embedding, and other algorithms to obtain feature representations, which were processed by Factorization Machine learners. Homogeneous weak Factorization Machine learners are then trained based on Ensemble Learning. We integrated these weak learners into a stronger learner through the voting mechanism and generated the final recommendations. [Results] We examined the new model with the CiteULike dataset, and the Precision, Accuracy, and F-Measure reached 72.6%, 69.7%, and 76.2%, respectively, 20%, 15%, and 9% higher than the benchmark algorithms. [Limitations] The input, sampling strategy, and processing mode need to be further explored. [Conclusions] The proposed Ensemble Factorization Machine enables effective representation and utilization of sparse data features, enhancing the recommendation performance.

[Objective] This paper utilizes deep learning to recommend medical services for patients, which helps them choose doctors during online diagnosis and treatment. [Methods] First, we used the Hierarchical Attention Network and patient consultation records to construct doctor-patient models. Then, we designed doctor recommendation schemes based on the “doctor-patient” compatibility and patient “rating”. Both schemes incorporated the HAN deep learning framework to build doctor-patient models and used attention mechanisms to enhance the interaction of “doctor-patient”. Patients with similar conditions to those inquiring about treatments receive higher weights, which helped us calculate the doctor’s recommendation score. [Results] The HAN model could extract the critical information representing the patient’s condition from their disease descriptions. The recommendation hit rate was improved by 16.45% compared to the classical Word2Vec model by improving the modeling quality. For the recommendation score, the “rating” scheme based on the attention mechanism achieved the highest hit rate (79.7%), which is significantly outperforming the cosine similarity-based scheme (74.9%). [Limitations] This study only utilized historical patient consultation data under each doctor’s name to model the doctors, and the model did not include information such as the doctor’s reputation, credentials, and expertise. [Conclusions] Constructing user and recommendation objects is crucial in designing recommendation systems. Enhancing feature interaction between the users and recommendation objectives can improve recommendation quality. This study validates the advantages of deep learning modeling techniques in recommendation tasks.

[Objective] This paper conducts a meta-analysis of current empirical research on the adoption of health device users, aiming to explore the relationship between various factors and the users’ adoption. [Methods] Based on the Comparison Standards Paradigm (CSP), we divided the antecedents of users’ adoption into three stages and five dimensions (standard establishment-experience perception-comparison and results). We also utilized meta-analysis to re-analyze 56 independent studies. [Results] We found that all variables in the process of the standard establishment had positive impacts on user adoption, of which social influence has a more substantial effect. In the perception stage, the ease of use had a strong positive correlation with user adoption of wearable devices for medical purposes. In the comparison and results stage, trust had the most substantial influence on the adoption of users among multiple antecedent variables. [Limitations] The sample size of this study needs to be expanded, which might generate consistent moderating effects of some variables. [Conclusions] In addition to consumer innovation and perceived loss, the study verifies the actual effect size of other user adoption factors, laying a foundation for new theoretical models in the future.

[Objective] This paper proposes an improved academic trace method and quantitatively evaluates the academic impacts of traditional Chinese medicine scholars. It aims to improve the evaluation system of traditional Chinese medicine talents. [Methods] Based on the academic trace method of author position weight, we proposed a weight calculation method of academic trace using the contribution of corresponding authors and journal impact factors. [Results] We compared the performance of four academic trace methods. The proposed one yielded significantly better results and rankings than the others. We also verified the rationality and effectiveness of the proposed method with the award-winning information and Q-index. [Limitations] The evaluation research based on weight academic trace is only carried out for the English papers of traditional Chinese medicine scholars, and the subsequent experiments will be carried out to verify the Chinese papers of scholars and other scholars in the field of traditional Chinese medicine.[Conclusions] The proposed weighted academic trace method could effectively evaluate the impacts of scholars in traditional Chinese medicine.