Business Context
An e-commerce platform needs to tag named entities (Product Names, Brands) in thousands of daily customer reviews to analyze trends. Performance and stability on larger datasets are key.
Data Preparation
Generation of a dataset containing 10,000 synthetic reviews and reuse of the generic model structure.
Copied!
| 1 | DATA mycas.reviews_large; |
| 2 | LENGTH review_id $10 content $300; |
| 3 | DO i=1 to 10000; |
| 4 | review_id=cats('REV', i); |
| 5 | content='This product from AcmeCorp is amazing and durable.'; |
| 6 | OUTPUT; |
| 7 | END; |
| 8 | RUN; |
| 9 | |
| 10 | /* Reusing model structure from previous examples or creating dummy model tables for volume test */ |
| 11 | DATA mycas.v_label; LENGTH _label_ $20 _type_ $20; INPUT _label_ $ _type_ $; DATALINES; |
| 12 | B-BRAND ORGANIZATION |
| 13 | O OTHER |
| 14 | ; RUN; |
| 15 | DATA mycas.v_attr; LENGTH _attr_ $50 _value_ $50; INPUT _attr_ $ _value_ $; DATALINES; |
| 16 | WORD[0] AcmeCorp |
| 17 | ; RUN; |
| 18 | DATA mycas.v_feat; LENGTH _feature_ $50; INPUT _feature_ $; DATALINES; |
| 19 | U00:AcmeCorp |
| 20 | ; RUN; |
| 21 | DATA mycas.v_attr_feat; INPUT _attrid_ _featureid_ _weight_; DATALINES; |
| 22 | 1 1 2.0 |
| 23 | ; RUN; |
| 24 | DATA mycas.v_temp; LENGTH _template_ $100; INPUT _template_ $; DATALINES; |
| 25 | U00:%w[0] |
| 26 | ; RUN; |
Étapes de réalisation
1
Execution of scoring on the volume dataset.
Copied!
| 1 | PROC CAS; |
| 2 | conditionalRandomFields.crfScore |
| 3 | TABLE={name='reviews_large'}, |
| 4 | model={ |
| 5 | attr={name='v_attr'}, |
| 6 | attrfeature={name='v_attr_feat'}, |
| 7 | feature={name='v_feature'}, |
| 8 | label={name='v_label'}, |
| 9 | template={name='v_template'} |
| 10 | }, |
| 11 | casOut={name='reviews_tagged', replace=true}, |
| 12 | target='entities'; |
| 13 | RUN; |
Expected Result
The action executes successfully without timeout. The output table 'reviews_tagged' contains 10,000 rows, with the 'entities' column populated.