¶ Description
Allow to connect to the Rosette API for advanced Text Mining capabilities.
¶ Parameters
¶ Standard tab
Parameters:
- Text analytics powered by Rosette Text Analytics
- Document ID
- Text to process
- Operation
- Source text language
- API key
- Output column name prefix
¶ Advanced tab
Parameters:
- Rosette limitation
- The maximum payload size is 600KB, with a maximum character count of 50,000
- Sentiment and category extractions are only working for English
- Throttling
- Max throughput (0 = unlimited)
- On lost server connection
- Time interval between each connection attempt
- Retries on connection error
- TCP/IP network server time-out
- Process all rows
¶ About
Rosette applies cloud NLP from Rosette Text Analytics to each incoming row. It supports four operations:
- Sentiment analysis
- Parts-of-Speech (POS) tagging
- Entity extraction (named entities)
- Categorization (topic labels)
Limits (as shown in the box): max payload 600 KB and max 50,000 characters per row. Sentiment and Categorization are English-only.
¶ Prerequisites
-
A valid Rosette API key (keep it secret; paste masked in screenshots).
-
Outbound Internet allowed from the worker to Rosette’s API.
-
Input table with:
- a unique Document ID column,
- a Text column containing the text to analyze (UTF-8).
-
Each row’s text fits both limits: ≤600 KB and ≤50,000 chars.
-
Optional: choose the Source language or keep AUTO (recommended for mixed data; note: Sentiment & Categorization still require English).
¶ Input Requirements
| Column | Type | Required | Notes |
|---|---|---|---|
| Document ID | string | Yes | Unique per row; used to correlate outputs. |
| Text to process | string | Yes | UTF-8 text; HTML should be pre-cleaned if you don’t want tags analyzed. |
| (implicit) Lang | string | No | Use Source text language parameter or AUTO. |
General file assumptions: UTF-8, header row present.
¶ Parameters
¶ Standard
| Name | Type | Required | Default | Description |
|---|---|---|---|---|
| Document ID | Column selector | Yes | — | Column containing a unique identifier for the row/document. |
| Text to process | Column selector | Yes | — | Column containing the text to send to Rosette. |
| Operation | Enum | Yes | sentiment analysis |
One of: sentiment analysis, parts-of-Speech (POS), entity extraction, categorization. |
| Source text language | Enum | No | AUTO |
Language hint for the text. Options include AUTO, Arabic (ara), Chinese (zho), Dutch (nld), English (eng), French (fra), German (deu), Hebrew (heb), Indonesian (ind), Italian (ita), Japanese (jpn), Korean (kor), Pashto (pus), Persian/Dari/Farsi (fas), Portuguese (por), Russian (rus), Spanish (spa), Urdu (urd). (Sentiment & Categorization: English only.) |
| API key | Secret string | Yes | — | Your Rosette API key. Do not commit to source control. |
| Output column name prefix | String | No | (blank) | An optional prefix added to all generated output columns (e.g., ros_). |
Inter-parameter rules
- If Process all rows = OFF, the Only rows where … equals … filter is applied; otherwise it’s ignored.
- Max throughput > 0 enables throttling;
0leaves it unbounded. - Operation controls the output schema (see below).
- Source text language may be AUTO, but Sentiment and Categorization still require English text to return results.
¶ Output
New columns are appended to each row. Names begin with your Output column name prefix (if provided). Exact column set depends on Operation:
- Sentiment analysis
Adds a sentiment label (e.g., positive/neutral/negative) and a confidence/probability value. - Parts-of-Speech (POS)
Adds tokens and their POS tags (typically arrays or a structured JSON string). - Entity extraction
Adds recognized entities (e.g., people, organizations, locations) with entity types and positions; commonly represented as a JSON array. - Categorization
Adds category label(s) with scores indicating assignment strength.
Column names follow the box’s internal naming with your chosen prefix (e.g.,
ros_sentiment,ros_confidence,ros_entities, etc.). The exact suffixes may vary by version.
¶ Example Workflow
-
Upstream input
Ensure your table hasidandtextcolumns (UTF-8). Remove HTML if not desired. -
Drop the Rosette box and connect it after your input/cleaning steps.
-
Standard tab
- Document ID:
id - Text to process:
text - Operation:
sentiment analysis(pick others as needed) - Source text language:
AUTO(orEnglish (eng)for sentiment/categorization) - API key:
********(masked) - Output column name prefix:
ros_
- Document ID:
-
Advanced tab (safe defaults)
- Max throughput:
0(unlimited; raise to e.g.60to self-throttle) - Time interval between attempts:
10 - Retries on connection error:
5 - TCP/IP timeout:
5000 - Process all rows:
ON
- Max throughput:
-
Run the box.
-
Downstream write/inspect
Use a viewer or a writer to confirm new columns withros_prefix appear and contain valid values for several sample rows.
Validation checklist
- For N rows in, N rows out (unless you deliberately filtered).
- New
ros_*columns exist and are non-null for English text. - Spot-check 3–5 rows for sensible labels/entities/tags.
- No row exceeds 600 KB or 50,000 chars.
- No 4xx/5xx errors in the run log.
¶ Troubleshooting
| Symptom | Likely Cause | Fix |
|---|---|---|
| 401/403 in log | Invalid/missing API key; wrong scope | Re-paste a valid key; confirm it’s not expired; avoid trailing spaces. |
| 413 / “payload too large” | Row text exceeds 600 KB or 50k chars | Truncate or split long documents upstream. |
| Empty sentiment results | Non-English text or undetected language | Ensure English text; set Source text language = English (eng) explicitly. |
| 422 / unsupported language for operation | Using Sentiment/Categorization on non-English | Restrict those ops to English; for other languages, use POS or Entity extraction. |
| Timeouts / connection errors | Network egress blocked; timeout too low | Allow outbound HTTPS; increase TCP/IP timeout (e.g., 15000 ms); set retries. |
| 429 / rate limited | Throughput too high | Set Max throughput (e.g., 60 req/min); add backoff at the pipeline level. |
| No new columns | Output prefix + schema expectation mismatch | Remove the prefix temporarily to inspect names; verify downstream writer schema refresh. |
| Filter returned 0 rows | Process-all OFF with strict filter | Turn Process all rows ON or adjust filter expression/value. |
¶ Use Cases
- Support ticket triage with per-ticket sentiment.
- Customer feedback mining via entities (brands, products) and categories.
- Document preprocessing: POS tagging for downstream NLP modeling.
- Compliance/news monitoring: extract entities to drive alerts and joins.
Notes
- Sentiment & Categorization are English-only (per box notice).
- Respect the 600 KB / 50k-char limit per row to avoid truncation or errors.
- Use Output column name prefix to keep schemas tidy when chaining multiple NLP steps.
