Business Context
A global e-commerce giant monitors reviews for thousands of products. They need to extend the standard English sentiment model with thousands of product-specific slang terms and positive/negative indicators specific to their niche (e.g., 'DOA' is negative, 'BIFL' is positive). The test validates performance on a larger rule set.
Data Preparation
Generating a synthetic dataset with 5,000 custom sentiment rules to test volume handling.
Copied!
| 1 | DATA casuser.retail_sentiment_rules; |
| 2 | LENGTH id $20 LITI_rule $100; |
| 3 | DO i=1 to 2500; |
| 4 | id=cats('POS_RULE_', i); |
| 5 | LITI_rule=cats('CONCEPT:AmazingProduct_', i); |
| 6 | OUTPUT; |
| 7 | id=cats('NEG_RULE_', i); |
| 8 | LITI_rule=cats('CONCEPT:BrokenItem_', i); |
| 9 | OUTPUT; |
| 10 | END; |
| 11 | RUN; |
Étapes de réalisation
1
Compiling a large rule set while extending the existing SAS Sentiment model.
Copied!
| 1 | PROC CAS; |
| 2 | textRuleDevelop.compileConcept / |
| 3 | TABLE={caslib="casuser", name="retail_sentiment_rules"}, |
| 4 | ruleId="id", |
| 5 | config="LITI_rule", |
| 6 | predefinedSentiment=TRUE, |
| 7 | casOut={caslib="casuser", name="retail_large_model", replace=TRUE}; |
| 8 | RUN; |
Expected Result
The compilation should complete within a reasonable time frame without memory errors. The output table 'retail_large_model' is created, containing a binary that merges the 5,000 custom rules with the base English sentiment logic.