Introduction
NewsLens investigates how automated sentiment models and AI-scored news lenses can support exploratory media analysis. The project focuses on transparent comparison rather than definitive judgments about outlet quality.
Poster and Paper Materials
NewsLens Research Writeup Hub
This project presents a proof of concept for a dynamic, exploratory framework for discourse analysis applied to interpretive media studies. Rather than adjudicating truth claims or advancing normative prescriptions, the system is designed to support exploratory inquiry into how meaning, rhetoric, and power are constructed and circulated within large-scale media environments.
Discourse analysis has long been central to literary and media studies, relying primarily on qualitative and interpretive methodologies. Over recent decades, increased computational accessibility has enabled the incorporation of quantitative techniques, such as sentiment analysis and automated text classification, into these domains. More recently, advances in natural language models and large language models have further expanded the capacity to process, label, and organize large corpora of textual data at relatively low cost.
Building on emerging best practices in rubric-driven AI classification, this project proposes a dynamic analytical tool that integrates interpretive frameworks with scalable computational methods. The system is not intended to produce definitive conclusions or guide decision-making in a conventional data science sense. Instead, it functions as an exploratory environment in which researchers can iteratively interrogate relationships among discourse, framing, rhetoric, and institutional power.
By positioning computational analysis as a complement to interpretive inquiry, this work foregrounds the role of data-driven exploration in supporting ontological and critical questions about media systems, artificial intelligence, and the production of meaning within contemporary information ecosystems.
Project Summary
One-Sentence Contribution
NewsLens is a public research dashboard that combines sentiment baselines, AI-assisted lens scoring, and controlled media analytics to make exploratory news-framing comparisons more transparent and auditable.
Object of StudyRSS news articles scored across interpretable lenses.
Primary MethodBackend-derived comparisons by source, topic, tag, event, and latent lens geometry.
Interpretive FrameExploratory signal discovery, not final judgment or quality ranking.
Poster Section
Poster Content Outline
Use these blocks as poster panels or slide sections.
Research Question
Can a public dashboard combine local sentiment models, rubric-based AI scoring, and controlled source comparisons to reveal interpretable differences in news framing?
Methodology
The system collects RSS articles, normalizes article metadata, scores articles across interpretable lenses, and computes backend-derived summaries for sources, topics, tags, events, and latent-space views.
Analysis
Current analysis includes lens distributions, source-by-lens matrices, lens correlations, PCA/MDS projections, topic-controlled source comparisons, tag-controlled source comparisons, event-controlled comparisons, drift diagnostics, and tag momentum.
Key Results to Present
The poster should highlight one or two clear examples: a topic-controlled source comparison, a lens PCA visualization, a tag momentum example, and an audit trail from aggregate metric to article-level evidence.
Limitations
Scores are automated and should be interpreted as exploratory signals. Topic mix, event selection, prompt changes, model drift, and missing article text can affect results.
Future Work
Planned work includes individual lens detail pages, stronger calibration, richer event matching, temporal centroid movement, Postgres-backed analytics snapshots, and additional uncertainty indicators.
Paper Section
Paper Content Outline
A draft structure for a research paper or extended project report.
Abstract
This project presents NewsLens, a public research dashboard for comparing sentiment models and AI-assisted news analytics. The system combines RSS ingestion, rubric-based scoring, backend-derived statistical views, and a Next.js/FastAPI interface for interpretable exploratory analysis.
Introduction
The paper should motivate the difficulty of comparing news sources when articles differ by topic, event, and editorial mix. NewsLens is framed as an analytic instrument for generating inspectable hypotheses about framing patterns.
Related Work
Discuss sentiment analysis, media framing analysis, computational journalism tools, dimensionality reduction for exploratory data analysis, and the risks of automated scoring without calibration.
Data
Describe the RSS-based corpus, article metadata, source labels, publication dates, topic tags, AI tags, scraped text, summaries, and scoring records. Include data-quality exclusions and missingness treatment.
Methods
Explain the pipeline: article ingestion, normalization, local sentiment baselines, rubric/lens scoring, source aggregation, topic/tag duplication policies, event clustering, PCA/MDS latent views, FDR correction, drift diagnostics, and precomputed snapshots.
Evaluation Strategy
Separate model evaluation from news interpretation. Report local sentiment model metrics where available, then describe exploratory validation checks such as topic controls, tag controls, same-event comparisons, stability diagnostics, and auditability.
Results
Use selected dashboard outputs as figures: score distributions, source effects, topic-controlled differences, tag lens PCA, source/tag composition, trend charts, and event-controlled comparisons where data is sufficient.
Discussion
Interpret outputs as patterns in lens-score space, not as direct measures of journalistic quality. Discuss where pooled differences collapse under topic or event control and where controlled differences persist.
Limitations
Address automated scoring reliability, lack of human-grounded calibration, source selection bias, topic tagging noise, event-clustering uncertainty, model/prompt drift, and small-sample slices.
Future Work
Add lens detail pages, improve event clustering, persist derived metrics in Postgres, add human or proxy calibration, track temporal movement of centroids, and expand uncertainty labels in the public interface.
Conclusion
NewsLens demonstrates a practical architecture for transparent, controlled, exploratory news analysis while making the interpretive boundary conditions visible to users.
Figures
Suggested Figures and Tables
- System pipeline diagram from RSS feed to scoring to FastAPI to Next.js dashboard.
- Lens PCA or MDS map showing article/source positions in lens-score space.
- Topic-controlled source differentiation table or chart.
- Tag momentum chart showing recently accelerating tags.
- Source-tag composition matrix to contextualize source comparisons.
- Event-controlled comparison example showing same-story source differences.
Guardrails
Claim Boundaries
- Use: source framing pattern, score distribution, topic-controlled difference, event-controlled comparison.
- Avoid: outlet quality ranking, truthfulness ranking, or definitive ideological classification.
- Always report sample size, control mode, missingness, and whether a view is pooled or controlled.
- Treat automated rubric scores as exploratory measures until calibration work is added.