A Multimodal Framework Embedding Retrieval-Augmented Generation with MLLMs for Eurobarometer Data

Public opinion software was used to collect data in the context of internet big data, aiming to analyze the situation of resident standardized training. The results indicated that over 820 thousand public opinion items regarding resident standardized training were collected and the greatest attention paid to this issue came from medical college students. 51% (418 200/820 000)of the public opinion items were neutral, 32% (262 400/820 000)of the items was negative, and 17%(139 400/820 000)was positive. Specifically, 39% (102 336/262 400)of the media focused on the policy of resident standardized training, 27% (70 848/262 400)on the training period, and 26% (68 224/262 400)on the training income. Both the media and cyber citizen paid close attention to resident standardized training, but their attitudes towards it were polarized. Government should create public opinion research and judge mechanism, information dissemination mechanism and public opinion guidance mechanism to orient internet public opinion tracking in a scientific and objective way. Key words: Internet; Big data; Residents; Standardized training; Public opinion analysis

Natural disasters are usually sudden and unpredictable, so it is too difficult to infer them. Reducing the impact of sudden natural disasters on the economy and society is a very effective method to control public opinion about disasters and reconstruct them after disasters through social media. Thus, we propose a public sentiment feature extraction method by social media transmission to realize the intelligent analysis of natural disaster public opinion. Firstly, we offer a public opinion analysis method based on emotional features, which uses feature extraction and Transformer technology to perceive the sentiment in public opinion samples. Then, the extracted features are used to identify the public emotions intelligently, and the collection of public emotions in natural disasters is realized. Finally, through the collected emotional information, the public's demands and needs in natural disasters are obtained, and the natural disaster public opinion analysis system based on social media communication is realized. Experiments demonstrate that our algorithm can identify the category of public opinion on natural disasters with an accuracy of 90.54%. In addition, our natural disaster public opinion analysis system can deconstruct the current situation of natural disasters from point to point and grasp the disaster situation in real-time.

This study analyzes the public's opinions and assessments regarding the candidates' capabilities for Governor of Riau, especially regarding candidates who are suitable to lead Riau Province. In addition, people's predictions regarding the candidates who will win the second re-election. This study uses descriptive qualitative methods to thoroughly, broadly, and in-depth explore social situations. The results found that the community's assessment of candidates more suitable to lead Riau Province is balanced. Of the 20 informants, eight people (40%) considered Herman Abdullah more suitable, eight people (40%) stated that Anas Ma'amun was more appropriate, and the remaining four people (20%) considered the two candidates for Riau Governor equally suitable. It means that the two candidates are still balanced when viewed from the context of appropriateness. Besides that, the informants generally sided with the AM figure for the community's prediction of the candidate being promoted as the winner. Of the 20 informants, the majority, namely 16 (80%), favoured AM as the winner, three (15%) favoured the chances of both being equal (fifty-fifty), and one (5%) predicted HA would be the winner. This study concluded that the two candidates for governor have balanced capabilities, but according to the majority prediction, AM will win the gubernatorial election. Furthermore, this study seeks to explain the process of forming public opinion from various perspectives.

Traditional public opinion analysis methods mostly focus on single modal data, and it is difficult to make full use of the correlation and complementarity between multi-modal data. In this paper, a multi-modal news data mining and intelligent public opinion analysis model (MM-OpinionNet) is proposed. By fusing multi-modal data, a more comprehensive and accurate public opinion analysis model is constructed. In the research, text data is preprocessed by word segmentation and stop words removal, image and video data are processed by standardization and object detection, and audio data is processed by speech-to-text and acoustic feature extraction. In the feature extraction stage, BERT embedding and TF-IDF are used for text, ResNet-50 for image, 3D-CNN for video and MFCC and prosodic features for audio. The model realizes feature fusion through cross-modal attention mechanism and graph neural network, and introduces cross-modal comparative learning loss to solve the problem of data heterogeneity. The experiment was based on a synthetic multimodal news dataset containing 100,000 news samples encompassing text, images, videos, and audio. The results demonstrated that MM-OpinionNet outperformed unimodal baseline methods and traditional multimodal approaches in event detection, sentiment analysis, and public opinion prediction. It achieved an accuracy of 89.7%, reduced the Root Mean Square Error (RMSE) for sentiment analysis by 34%, and attained a Mean Absolute Percentage Error (MAPE) of 8.3% and a Mean Squared Error (MSE) of 4.1×10³ for public opinion prediction. Ablation studies further confirmed the importance of the cross-modal attention mechanism and multitask learning in enhancing model performance.

Take-away food (also referred to as “take-out” food in different regions of the world) is a very convenient and popular dining choice for millions of people. In this article, we collect online textual data regarding “take-away food safety” from Sina Weibo between 2015 and 2018 using the Octopus Collector. After the posts from Sina Weibo were preprocessed, users’ emotions and opinions were analyzed using natural language processing. To our knowledge, little work has studied public opinions regarding take-away food safety. This paper fills this gap by using latent Dirichlet allocation (LDA) and k-means to extract and cluster topics from the posts, allowing for the users’ emotions and related opinions to be mined and analyzed. The results of this research are as follows: (1) data analysis showed that the degree of topics have increased over the years, and there are a variety of topics about take-away food safety; (2) emotional analysis showed that 93.8% of the posts were positive; and (3) topic analysis showed that the topic of public discussion is diverse and rich. Our analysis of public opinion on take-away food safety generates insights for government and industry stakeholders to promote the healthy and vigorous development of the food industry.

Using data analysis and case anatomy, this paper interprets the public opinion under the outbreak of “COVID-19 pneumonia” from the perspective of “presence”, and deconstructs the generation mechanism of public opinion under the current media ecology. At the initial stage of the “COVID-19 pneumonia” epidemic, the “media time and space” based on cyberspace was built under the active “presence” of some people. In the following month, people were “present” one after another through active information acquisition, and the energy in the field continued to increase. On January 20, the statement of “human to human” was announced, and the amount of epidemic coverage was “blowout”, which immediately activated the public opinion ecology in “media space-time”. The generation of public opinion under the epidemic situation has exposed the weakness of the current media social early warning and the public's negative resistance to the traditional “agenda setting”. Under the new public opinion ecology, the way people get information and participate in discussion has changed, and a new double public opinion field ecology has been constructed. In the new ecological environment, mainstream media and social media, which are in the “one light and one dark” mode, should make good use of their own advantages and shoulder their social responsibilities. They should not only do a good job in guiding public opinion, but also pay attention to the emotional satisfaction of the people, so as to guide the people to complete the “presence” ceremony.

It is a commonly held belief that the foreign policy issue to which the American public is most sensitive is the use of military force. Because American public opinion regarding the use of force is highly palpable, salient, and organized, and because decisions regarding the use of force are some of the most important decisions the nation ever has to consider, the analysis of public opinion regarding military involvement is academically significant and policy relevant. The indication from policymakers is that American military operations require public support. As a result, scholars and analysts have come to realize that public opinion is the “essential domino” of military operations. The relationship between mass American public opinion and the use of military force has become, therefore, the focus of numerous studies and surveys. There are currently several competing explanations—schools of thought—in the literature on why the mass public supports the use of force. This article is an attempt to identify the most prominent schools of thought on public opinion and the use of force, and the central factors associated with each school. Such a review is important to generating policy–relevant guidance pertaining to public opinion and the use of force—an objective made more pressing by the current war against terrorism.

This paper proposes a semantic information extraction technology model system based on BERT, aiming to solve the problem of semantic understanding and information extraction in online public opinion analysis. The system design includes data collection, preprocessing, semantic information extraction and result display modules, using the BERT model to capture the deep semantic relationship of the text and enhance the ability to understand complex semantics and contextual information. The data collection module covers multiple sources such as social media, news comments and forum posts, and ensures the high quality and diversity of data through real-time crawling and cleaning. The semantic information extraction module is based on the multi-head attention mechanism of BERT, which can effectively capture the emotional tendency, topic information and potential intention in the public opinion text. The experiment verifies the system performance based on 1 million online public opinion data. The results show that in the semantic information extraction task, the accuracy of the BERT-based model is improved by 22.4%, the recall rate is improved by 19.8%, and the F1 value reaches 91.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, which is significantly better than traditional methods such as TF-IDF and Word2Vec. The sentiment classification module has an accuracy of 93.1%, achieving accurate recognition of positive, negative and neutral emotions. The processing time of a single piece of public opinion text is reduced by an average of 25.6%, indicating the high efficiency of the system in large-scale data processing. The simulation test verifies the system's stable operation capability in the scenario of 100,000 new public opinion data per day, with a delay of less than 1 second, which can meet the needs of real-time public opinion analysis.

Network public opinion analysis, which is through the analysis of the situation of public opinion, can reflect the true society and public opinion, and on this basis, a reasonable forecast and recommendations decision-making can be provided for relevant managers. The paper studies on agricultural science and technology related website, and realize internet public opinion analysis for agriculture science and technology site. The network public opinion hotspots found technology is studied. It is achieved internet public opinion hotspots monitoring and found through the network information automatically grab and information classification. It is provided an analytical basis for managers to grasp the ideological trends in the network and make the correct guidance.

With the rapid development of the automotive industry and the growing influence of social media, public opinion on automotive safety tests has become a key factor affecting consumer trust and the reputation of corporations. In 2025, the collision experiment between Li Auto i8 and CHENGLONG truck caused controversy on Weibo, which authenticity of the test results and the safety performance of the vehicles were questioned by public. However, traditional public opinion analysis methods struggle to accurately capture both thematic focuses and emotional tendencies in large-scale unstructured comment data. Against this backdrop, this study focused on analyzing public opinion regarding the controversial collision experiment between Li Auto i8 and CHENGLONG truck, which happened in 2025. This research used a fusion method of LDA, BERT and data augmentation. It explored public opinion themes and the effectiveness of the model in automotive public opinion text analysis, which provided insights for car companies' responses. Data was selected from 599 Weibo comments, after cleaning, tokenizing, stopword filtering and augmenting, data reached 2803 samples. The LDA-BERT (Latent Dirichlet's Assignment - Bidirectional Encoder Representation) fusion model extracted four core topics, which achieved 98.93% sentiment classification accuracy. Findings showed the model effectively managed automotive public opinion. It helped car companies find the direction of public questioning. Limitations included insufficient negative emotion extraction in LDA and errors in recognizing niche slang. In the future, more research can be done on optimizing the subject title filtering logic. Although the research had limitations, it can still contribute methodologically and practically to automotive public opinion analysis.

At present, emergencies occur frequently. How to better adapt to the changes brought by big data in the analysis of network public opinion for emergencies is an important issue of emergency management in the era of big data. In this paper, combining the actual needs and the shortcomings of the existing public opinion pre-warning system, data mining technology, emotion computing technology and artificial intelligence methods are introduced into network public opinion pre-warning system, build a network public opinion pre-warning system model based on Hadoop big data architecture, which includes four layers: public opinion data collection layer, data preprocessing layer, public opinion analysis layer, public opinion pre-warning and decision support layer. The model integrates the important functions of the whole process of network public opinion analysis and pre-warning, in order to provide accurate pre-warning information and scientific auxiliary decision-making scheme for emergency management.

Does the organization of the assembly affect whether governments deliver policy that reflects the public's changing preferences? Cross‐national analyses of public opinion and policy outputs for policies concerning welfare and immigration show that governments respond to shifts in public opinion in systems with a dominant chamber but not where bicameralism is strong. Our theory's emphasis on the distribution of power between chambers further explains differences within bicameral systems: constraints on policy change mean that responsiveness is weaker where power is equally distributed between chambers but more robust where power is concentrated in the lower house. Evidence from institutional change in Belgium, where the fourth state reform shifted power away from the senate and disproportionately toward the lower house, provides corroborating evidence that policy becomes more responsive when constitutions concentrate legislative power. This study's findings have implications for our understanding of how bicameralism matters for government responsiveness to public opinion.

Technology social is the fastest and most up-to-date source of information. A model that can provide mapping will help in sorting out information more precisely and quickly. Public opinion in the mass media always develops quickly to talk about an issue in just a few days or even hours, so we do not know what the opinions of the people in the mass media are on the issue. In this study, the author applied topic modeling to the results of sentiment analysis on PPKM. The source of data in this study was obtained from twitter using SNScrape. The collected data was analyzed sentiment using the Long Short-term Memory (LSTM) method, so that public opinion was obtained with positive, negative, and neutral sentiments. The classification obtained from the results of the sentiment analysis process is continued with the topic modeling process using the Latent Dirichlet Allocation (LDA) method and visualized in the form of a wordcloud to find out the relationship between one topic and another. The sentiment analysis process produces a model with an accuracy rate of 90.8% and the topic modeling process successfully presents topics that are easy to interpret so that conclusions can be known about an issue.

With the explosive growth of social media data, Weibo public opinion analysis faces challenges such as short text sparsity, semantic complexity, and sentiment dynamics. This paper proposes an analytical framework based on BERTopic topic discovery and ALBERT-TextCNN sentiment classification, achieving multi-level public opinion analysis. First, the BERTopic and HDBSCAN algorithms are combined, and UMAP dimensionality reduction is applied to Weibo text data to address the sparsity issue in short text clustering. Second, the ALBERT-TextCNN hybrid model is used to improve the fine-grained performance of sentiment classification. Experimental results show that the framework can dynamically extract topics in real-time from Weibo text data and accurately classify sentiments. Finally, a public opinion spatiotemporal visualization system is used to display hot topics and their sentiment variations. This research effectively tracks and analyzes public opinion, providing strong technical support for governmental and organizational public opinion supervision and decision-making, with broad application prospects.

Developing a highly efficient transformer from an embedding public opinion network into a low-dimensional vector space contributes a lot to many research areas such as vertex classification, community detection and public opinion analysis, etc. Most existing network embedding methods have chosen to analysis in social networks. However, constructing a social network from public opinions is very sparsely, which would serve as an effective way to capture and process public opinions. On top of that, social network can only reflect the social relationships between nodes while the information derived from opinions is neglected. Hence, a network that incorporates opinion features of nodes into social networks is reported. This study evaluates the similarity of opinions from different nodes and connects them with enough similarity. The final public opinion network would certainly be denser than the social network. Experimental results show that researchers might give top priority to use the approach of public opinion network embedding compared with the regular social network methods, especially when the sentiment orientation of opinions is explicit.