From input to interpretation.
Input format
One already cleaned Chinese post per row, with four columns. Raw-post cleaning is not provided.
| Column | Purpose |
|---|---|
id | Your row identifier, returned unchanged. |
timestamp | Returned unchanged. The service does not parse or modify it. |
text | The analysis-ready, already cleaned Chinese post text to be evaluated. |
referents | The referents you specify. Separate multiple referents with an English comma: Company T,Company L. |
For CSV, quote a field that contains commas. Keep identifiers as text to preserve leading zeros. No timestamp format is imposed or parsed.
A worked example will be added after a real run has been verified.
Output fields
One row is returned for each post–referent pair. A post with N specified referents produces N output rows.
| Column | Meaning |
|---|---|
id | The input identifier, unchanged. |
timestamp | The input timestamp, unchanged. |
text | The input text, unchanged. |
referent | The specified referent for this row. |
sentiment_strength | The signed, adjusted sentiment-strength value. |
status | ok, referent not found, timeout or a safe error description. Failed pairs have a blank sentiment_strength. |
source | Actual result source: live or archive; current official route is live. |
lexicon | event, general, or event+general — the selection actually used. |
Lexicon selection and provenance
Event is selected by default. General and event+general are exploratory configurations, not the configuration used in the paper. With both selected, event-specific entries override general entries, including cross-polarity conflicts; each word is counted once.
Every result and CSV row retains status, source and lexicon. The current official service returns source=live. Earlier local validation results are marked unusable. A score of 0 does not by itself establish neutrality: limited lexicon coverage or cancellation can produce zero.
Scores display six decimal places; nonzero values below 1e-6 in absolute magnitude use scientific notation. Downloads use the identical representation. Failed scores stay blank.
Meaning of the value
Positive values indicate supportive evaluation; negative values indicate oppositional evaluation.
Sentiment strength is a continuous value, not a polarity label.
Scope and boundaries
Only already cleaned text is accepted. Referents must appear in the text. Long-text chunking is not yet verified; unsupported text returns an error without silent truncation. The processing timeout is 20 seconds per post–referent pair.
Maximum 1000 input rows per submission, automatically divided into requests of at most 200 post–referent pairs. Each newly issued reviewer credential lasts 60 days and permits 1000 pairs in total, including failed pairs. Deleting results or restarting the service does not restore quota.
Referents are specified by the user. Entity extraction and alias normalization are not exposed by this service. The instance uses the study’s event-specific sentiment lexicon.
Terms of use
Access is for non-commercial academic research only. Scraping, redistribution and reverse engineering are not permitted.
Submitted input files and temporary processing files are deleted when a task completes, fails, or is cancelled. Results include the submitted text and are retained for download for 7 days after completion, unless deleted earlier by the user. After that period, the stored results are automatically deleted. Submitted texts are used only to process the requested analysis.