social media data collection

Social media corpora for clean-energy sentiment analysis suffer from high noise, poor timeliness, and low relevance due to keyword-dependent filtering. Method: This paper proposes a lightweight fine-tuned pre-trained Transformer-based binary classification framework tailored for fine-grained topic discrimination in socially generated texts—where class boundaries are inherently ambiguous. Unlike conventional rule-based or coarse-grained retrieval approaches, our method adapts BERT-like models via supervised fine-tuning on human-annotated data, achieving strong generalization with minimal computational overhead. Contribution/Results: Evaluated on a real-world Twitter dataset, the model achieves an F1-score of 0.95, significantly outperforming baseline methods. The resulting high-quality, temporally responsive corpus substantially enhances downstream sentiment analysis in both accuracy and real-time capability. Our approach provides a scalable, cost-effective technical pathway for large-scale public opinion modeling in energy-related domains.

To address the scarcity of multi-platform social data, high acquisition costs, and platform-imposed restrictions on real-world data, this paper proposes a topic-driven cross-platform synthetic data generation framework. Methodologically, it designs a multi-platform-adaptive prompt engineering strategy to leverage Llama-3, Claude-3, and GPT-4 for batch generation of platform-specific textual content (e.g., Twitter, Reddit, Instagram). It further introduces the first fidelity evaluation framework tailored for cross-platform social data, integrating lexical and semantic similarity analysis with post-hoc calibration. Experiments demonstrate that the synthetic data closely approximates real data in both lexical distribution and semantic structure (average similarity: 0.82), while revealing significant disparities among LLMs in modeling platform-specific stylistic conventions. The evaluation framework is open-sourced, providing high-quality synthetic data to support downstream tasks such as misinformation detection and influence operation analysis.

Amid escalating information overload and media bias, the public urgently requires trustworthy, interpretable bias detection tools. This paper proposes the first dual-path analysis framework integrating large language models (LLMs) with explainable bias indicators to jointly model bias types, intensity, and contextual evidence in news texts. We develop an online analytical system supporting fine-grained, metric-driven automated identification and human-in-the-loop annotation, incorporating vector retrieval and an interactive web interface. Extensive evaluation across four cross-platform news datasets demonstrates the method’s effectiveness. The system’s source code is fully open-sourced, and the platform is publicly accessible. Our core innovation lies in injecting structured bias knowledge into the LLM’s reasoning process—thereby enhancing both predictive accuracy and interpretability without compromising either.

Existing machine-generated text detection research primarily targets long-form English texts, struggling with the linguistic irregularities—such as informal language, grammatical errors, emojis, and hashtags—prevalent in social media short texts, and lacks multilingual benchmarks. Method: We introduce the first social-media-oriented, multilingual (22 languages), multi-platform (5 platforms) benchmark for short-text detection, comprising 470K human-written and LLM-generated (from 7 models) short posts. Contribution/Results: This work systematically fills critical gaps in short-text, multilingual, and informal-language detection. We empirically reveal that platform choice significantly impacts detector performance. Through zero-shot transfer and fine-tuning experiments, we demonstrate strong cross-lingual and cross-platform generalization—fine-tuning yields substantial gains, while platform-specific adaptation proves essential for robustness, underscoring the decisive role of domain alignment in real-world deployment.

Constructing structured knowledge from multilingual (English/French/Spanish) social media texts in tourism faces challenges of linguistic diversity and prohibitively high annotation costs. Method: This paper proposes a multitask NLP framework tailored for few-shot learning, introducing the first fine-grained, multilingual, domain-specific dataset for tourism. It systematically evaluates and integrates few-shot learning, pattern-based prompting, and parameter-efficient fine-tuning, jointly leveraging sequence labeling and semantic resource alignment. Contribution/Results: The framework achieves performance on par with fully supervised baselines using only 15 samples for sentiment analysis, 160 for location recognition, and 200 for topic concept extraction (315 classes), effectively overcoming low-resource bottlenecks. It establishes a scalable, low-dependency automation paradigm for cross-lingual tourism sentiment analysis and knowledge graph construction.

This study addresses the limitations of traditional social media data collection methods, which are often cumbersome, context-disconnected, and prone to failure due to frequent changes in page structures—thereby compromising ecological validity. To overcome these challenges, this work proposes a self-healing browser extension that enables non-technical researchers to conduct code-free annotation and intervention experiments directly within native platform interfaces. The system innovatively integrates a large language model–driven self-healing mechanism capable of automatically detecting and repairing selectors to support robust capture of dynamic content. It also features a no-code form designer and a cross-platform injection framework. Validated across twelve major social media platforms, the approach significantly reduces data collection and maintenance costs while empowering researchers to carry out high-quality studies efficiently.

This study addresses the challenges of cross-platform social media data analysis—namely data heterogeneity, API restrictions, and privacy compliance—stemming from the absence of standardized, reproducible workflows. To overcome these limitations, the authors propose an open-source Python framework featuring a unified data model that harmonizes multi-source social data across five core dimensions: communities, accounts, posts, behaviors, and entities. The framework incorporates a configurable personally identifiable information (PII) anonymization module to ensure regulatory compliance and integrates an LLM-driven analytical layer that enables semantic enrichment without requiring code modifications. Evaluated through four case studies, the framework demonstrates significant improvements in fairness, reproducibility, and scalability for cross-platform textual and network analyses.

Existing datasets struggle to simultaneously preserve audiovisual content and hierarchical dialogue structures, limiting fine-grained analysis of political stance expression on short-video platforms. This work introduces a multimodal, context-aware dataset tailored to TikTok political discourse, uniquely integrating video audiovisual features, parent-child comment tree structures, and multi-target ternary stance annotations (support/oppose/neutral). The dataset comprises 161 videos related to the three leading candidates in the 2024 U.S. presidential election and 13,876 associated comments, with human annotations achieving a Krippendorff’s α inter-rater reliability above 0.72. It establishes a high-quality benchmark for multimodal political stance detection and reveals significant differences across political targets in both stance distribution and interaction depth.

Existing political discourse analysis lacks structured cross-modal alignment between parliamentary debates and social media content, hindering comparative studies of institutional versus public political communication. Method: We introduce MultiParTweet—the first framework systematically aligning multilingual parliamentary Twitter data with the German parliamentary corpus GerParCor. We develop TTLABTweetCrawler, a reproducible cross-platform crawler, collecting 39,546 tweets and 19,056 associated media items. We integrate nine multilingual text models and one vision-language model (VLM) to automate multimodal annotation of sentiment, emotion, and topics, validated on human-annotated subsets. Contribution/Results: (1) First structured mapping between parliamentary proceedings and social media discourse; (2) A VLM-driven multimodal annotation paradigm—human preference tests show statistically significant superiority over text-only models; (3) Empirical validation of cross-modal predictability among model outputs, enabling robust consistency analysis across modalities.

This work addresses the lack of scalable, service-oriented frameworks for continuously monitoring user engagement with and stance toward harmful content—such as conspiracy theories—at the scale of social media platforms. We propose the first modular system that integrates a service-based architecture with a multi-stage AI analysis pipeline, combining real-time data stream processing, topic modeling, sentiment analysis, and stance detection to enable sustained, large-scale monitoring of user interactions. A key strength of our approach lies in its capacity to capture early amplification dynamics of such content. Evaluation on over 7 million comments reveals that 70% of conspiracy-related interactions occur within the first week of posting; our system effectively identifies highly active user clusters, most of whom express supportive stances, thereby demonstrating the framework’s scalability and practical utility.