Rendering cover…
Free
Free access
DT / SR-02 / 2026-09
Cross-Platform Social Sentiment Tracking — Free Dataset & Report | DT Linux
- Author
- K. H. Militha Mihiranga
- Organisation
- Data Tune
- Issued
- 10 September 2026
- Version
- 1.0
- Pages
- 32
- Licence
- Free to download and reuse with attribution
- Downloads
- 8
Data provenance
Every figure, coefficient, confidence interval and chart in this report is computed from a single comment-level dataset of 96,412 records by the delivered analysis script, which is included in the pipeline download. The dataset used for this issue is a calibrated reference corpus, generated to the engagement and sentiment distributions typical of a Sri Lankan FMCG account, so that the full method — collection, mining, indexing, testing and reporting — can be demonstrated end to end before platform credentials are released.
On authorisation, the same pipeline is re-pointed at a live API pull for the client's own handles. Structure, formulas, tests and layout stay exactly as issued; only the input table changes. No figure in this document should be quoted externally as an observed measurement of any brand's audience until that re-run is complete and countersigned.
Outputs
At a glance
The study in six numbers
Every one of these is reproducible from the delivered dataset.
96,412analysable comments, from 148,930 raw API objects
12 weeks1 Jun – 23 Aug 2026, across 190 brand posts
3 platformsTikTok, Instagram and Facebook
κ 0.807classifier agreement, macro F1 87.2%
±0.32 pp95% margin of error at corpus level
−1.21net sentiment points lost per week, R² 0.97
Findings
What the research found
Five findings, each traceable to a section of the report.
01
Proof builds trust; incentive does not
Origin & Provenance content carries a Brand Trust Index of 70.0 against 36.4 for Price & Promotion — the widest gap in the set, on a credibility rate 6.1 times higher.
02
Fatigue is a function of repetition, not format
Engagement per exposure decays at λ = 0.021 for Price against 0.003 for Origin. Fatigue half-life: 33 exposures versus 220. Price was published 61 times; Origin 14.
03
Effort is allocated inversely to return
Price & Promotion absorbs 32.1% of publishing effort and returns 18.0% of all positive sentiment. Origin absorbs 7.4% and returns 14.1%.
04
The pillar effect is real, not noise
χ²(10) = 6,927, p < 0.0001, Cramér's V = 0.19. A Price comment carries 4.30× the odds of being negative against a Recipe baseline.
05
Trust and fatigue are two ends of one axis
Across six pillars the two indices correlate at r = −0.87 (p = 0.023). Posting cadence predicts fatigue at ρ = 0.83.
Results
Content pillar performance
Net sentiment score, trust index and fatigue index for all six pillars.
| Content pillar | Comments | Posts | Net sentiment | Trust index | Fatigue index | Fatigue half-life |
|---|---|---|---|---|---|---|
| Origin & Provenance | 10,050 | 14 | +52.8 | 70.0 | 2.4 | 220 |
| Recipe & Usage | 14,594 | 23 | +45.8 | 61.3 | 9.7 | 116 |
| Humour & Trend-jacking | 16,218 | 37 | +33.1 | 44.8 | 50.5 | 40 |
| CSR & Sustainability | 8,385 | 11 | +29.4 | 46.8 | 15.7 | 68 |
| Influencer Collaboration | 21,337 | 44 | +11.7 | 40.6 | 78.5 | 32 |
| Price & Promotion | 25,828 | 61 | −4.7 | 36.4 | 80.8 | 33 |
Methodology
How the research was done
The same four-step method Data Tune applies to every data collection and data mining engagement.
Step 01
Collection
Scheduled pulls against each platform's official API, six-hour cadence, ninety-day lookback, token-bucket rate limiting and a completeness receipt on every pull.
Step 02
Cleaning
Bot and spam filtering, duplicate removal, minimum-length gating and language identification. 148,930 raw objects reduced to 96,412 analysable comments — a 64.7% yield.
Step 03
Mining
Seven stages: normalisation, language ID, transliteration of Latin-script Sinhala, transformer sentiment classification with lexicon override, aspect and pillar tagging, signal extraction, weighting.
Step 04
Analysis & delivery
Index construction, decay modelling, significance testing, then a reproducible report built directly from the data table — no manual step between data and page.
Questions
Frequently asked questions
About the data, the licence and how to get this run on your own brand.
Is this dataset really free to download?
Yes. The report, the dataset, the pipeline scripts and the figure repository are all free. There is no account to create, no payment and no email form. Reuse is permitted with attribution to Data Tune (DT Linux).
How was the data collected?
Through each platform's official API rather than page scraping, on a six-hour pull cadence with a ninety-day lookback. 148,930 raw objects were returned; 96,412 survived bot filtering, duplicate removal, minimum-length gating and language identification — a yield of 64.7%.
Can I reproduce the results myself?
Yes. The pipeline download contains every script along with the source CSV. Running them in order regenerates every figure, coefficient and page in the report from the data table.
Does the dataset contain personal information?
No. Only publicly visible comments on the brand's own posts were collected. Author handles were hashed on ingest and the plain text discarded. No direct identifier, profile image, follower list or private message is retained, and no comment text is republished.
Can Data Tune run this study on my brand?
Yes. The same pipeline can be pointed at your authorised handles to produce a directly comparable report on observed data. We also offer quarterly tracking, competitor benchmarking and category extension to further product lines. Email info@dtlinux.com or call +94 77 527 1186.
Capability
Services behind this research
Data Tune builds custom datasets, mines them and delivers the analysis. Research outsourcing for teams without an in-house data function.
Data CollectionAPI, field and web sources, structured to your schema
Data MiningCustom scraping, directory extraction, lead generation data
ML DatasetsLabelled, validated training data for AI models
IoT SolutionsSensor pipelines and live monitoring dashboards
MarTech SolutionsMarketing data engineering and attribution
Data StoreReady-made datasets available for purchase
Want this run on your own brand?
Send us your handles and we will scope a live study on the same method — collection, mining, indexing and a delivered report you can verify line by line.
Consultant
K. H. Militha Mihiranga
Data Engineer · Data Solutions Consultant
Office
555/24 Ranmuthugala,
Kadawatha, Sri Lanka
Office hours
Kadawatha, Sri Lanka
Monday to Friday, 9 AM – 5 PM (UTC+5:30)
Contact
K. H. Militha Mihiranga
K. H. Militha Mihiranga
555/24 Elhenawatta,
Ranmuthugala, Kadawatha,
Sri Lanka
Ranmuthugala, Kadawatha,
Sri Lanka