Fast takeaway: The PLOS corpus-assisted discourse study (published July 9, 2026) compared Weibo and X posts from top government space accounts and found cross-national differences in themes, addressing terms, and co-occurrence patterns; see the original PLOS ONE. If you analyze transcripts, social posts, or multilingual corpora, this post explains how to reproduce that study's rigour using AI-enabled corpus analysis and which Evidano features to use (ingest, multilingual preprocessing, thematic and co-occurrence networks, and cross-segment comparisons).
Key Takeaways
Evidano is an AI-powered qualitative data analysis platform that ingests multilingual corpora, preserves reproducible preprocessing logs, and produces co-occurrence, thematic, and addressing-term analyses.
This post shows how to reproduce a PLOS corpus-assisted discourse study (published July 9, 2026) on Weibo and X using those features and a two-week pilot workflow that covers data ingest, preprocessing, co-occurrence mapping, coding, and cross-segment comparisons.
- The PLOS study compared Weibo (895 posts, 24, 769 tokens) and X (1, 204 posts, 25, 063 tokens) collected Apr 1–Oct 1, 2023.
- Core analyses include co-occurrence networks (Jaccard), topic clustering, and addressing-term word associations around self/audience/nation.
- Follow the 7-step reproducible workflow to run a two-week pilot and produce exportable tables, concordances, and audit logs; Evidano stores data encrypted and supports role-based access without using your data to train third-party models.
Dataset snapshot
What the PLOS study did (in plain terms)
The PLOS study built two specialized corpora (Weibo and X) and applied quantitative text-mining with KH Coder plus critical discourse analysis to examine three layers: topics, addressing terms, and word associations.
The PLOS study examined (1) topics and themes using co-occurrence networks, (2) addressing terms such as self, audience, and nation, and (3) words co-occurring with self-addressing terms, finding that China’s official accounts blend state affiliation, patriotism, and interpersonal warmth while U.S. accounts emphasize professional expertise, mission execution, and interest-based audience segmentation.
- Published: July 9, 2026; data period: Apr 1–Oct 1, 2023.
- Chinese corpus: 895 posts / 24, 769 tokens. English corpus: 1, 204 posts / 25, 063 tokens.
- Analyses: co-occurrence networks (Jaccard), topic clusters, and word-association around self-addressing terms.
How to run AI-enabled corpus analysis (primary keyword: ai-enabled corpus analysis)
Ingest & multilingual preprocessing
To ingest posts and preprocess multilingual data, Evidano ingests posts, transcripts, PDFs, and spreadsheets in batch and applies language detection plus custom dictionaries to preserve domain tokens.
For this use case: scrape Weibo/X (public posts), or upload CSVs of exported posts and add project dictionary entries to protect multi-word tokens like “Artemis II” or language-specific tokens like 航小科.
Outcome: reproducible token lists and traceable preprocessing steps, with segmentation corrections logged for auditability.
Automatic thematic & co-occurrence mapping
To surface themes and term relationships, Evidano runs automated thematic clustering and co-occurrence networks (Jaccard or PMI) to surface clusters equivalent to the PLOS study’s themes.
Evidano generates interactive network visualizations and exportable cluster tables that you can filter by time, account, or language.
Outcome: interactive co-occurrence graphs you can filter by time, account, or language.
Addressing-term and word-association analysis
To quantify addressing patterns, Evidano lets teams define code categories (self, audience, nation) and run frequency, concordance, and word-association analyses around query terms.
Evidano saves and versions codebooks so counts are reproducible across runs and provides concordance lines to support qualitative interpretation.
Outcome: a quantitative breakdown (for example, percentage self-addressing versus nation terms) and supporting concordance lines for qualitative interpretation.
Cross-segment and time-series comparisons
To spot trends across groups and time, Evidano compares segments (country, account, month) with cross-segment analysis to reveal shifts in themes or addressing patterns.
Evidano produces heatmaps and exportable segment-by-theme tables suitable for stakeholder briefings, for example comparing launch-window posts to outreach posts.
Outcome: heatmaps and exportable segment-by-theme tables for stakeholder briefings.
Secure collaboration and audit trail
To keep data secure and auditable, Evidano stores data encrypted, supports role-based access, and does not use your data to train third-party models.
Evidano enables sharing interactive reports while preserving raw concordances and preprocessing logs to meet academic and policy auditability requirements.
Outcome: reproducible, privacy-aware workflows suitable for policy and academic audiences.
7-step reproducible workflow (two-week pilot)
This 7-step workflow outlines how to reproduce a CADS like the PLOS paper in a two-week pilot using AI-enabled corpus analysis.
- 1) Collect: export or scrape public posts for target accounts (1–6 months).
- 2) Ingest: upload raw CSV/JSON to Evidano and enable language detection.
- 3) Normalize: apply lemmatization/segmentation and add custom dictionary entries (project names, mascots).
- 4) Tag: import a simple codebook (self/audience/nation) and run automated coding.
- 5) Explore: generate topic clusters, co-occurrence networks, and word associations (top 150 edges like the study).
- 6) Compare: run cross-segment frequency and time-series analyses; export tables and visualizations.
- 7) Synthesize: use Evidano’s chat-over-documents to draft an evidence brief and extract exemplars (concordance lines) for qualitative quotes.
FAQ: AI-enabled corpus analysis
How do I preserve domain tokens (e.g., “Artemis II” or 航小科)?
Add domain tokens to a project dictionary before tokenization to preserve them intact during preprocessing.
Evidano supports custom dictionaries so multi-word or language-specific tokens remain intact during preprocessing and tokenization.
Can I compare languages and still be rigorous?
Yes, you can compare languages rigorously by applying parallel preprocessing and aligning analytic units.
Use language-specific tokenization and manual checks, align analytic units at the post level, and compare normalized frequencies and co-occurrence measures as the PLOS study did for comparability.
Is this approach appropriate for sensitive data?
This approach can be appropriate for sensitive data when you apply redaction and follow ethics approvals and data controls.
For research with PII or sensitive content, apply redaction at ingest and use Evidano’s encryption and role controls; treat non-public data under your ethics approvals.
Wrapping up & next steps
AI-enabled corpus analysis scales corpus-assisted discourse studies across languages and platforms by automating preprocessing, mapping, coding, and cross-segment statistics while preserving qualitative depth via concordances and coded exemplars.
- Try a secure demo on Evidano to import two corpora, add a custom dictionary, generate co-occurrence networks, and produce a downloadable evidence brief in under 48 hours.
- If your work requires ethics oversight (human subjects or sensitive datasets), treat results as research-only and secure necessary approvals before analysis.
Ready to reproduce the PLOS methods on your own social corpora? Start a pilot at Try Evidano for free and get a reproducible, audit-ready workflow for AI-enabled corpus analysis.