Printed Work Classification - Automation

[emilyyan04]

For 11/9:

• Make best guess at threshold (+ reasoning)
• Use pillow library to manually label correct category
• After manual labels assigned, loop through dataset; adjust thresholds by increments, calculate error rate for each iteration, determine actual best cutoff
• Evaluate highest-performing cutoff to determine viability of this process
• Optional: relate general trends in scores to characteristics of posters vs prints

[emilyyan04]

Application of #108

[emilyyan04]

Dataframe schema:

• accession_number: accession number
• qid: QID
• locale: language (from text detection)
• description: text output (from text detection)
• label_en: title of work (manipulated for string matching)
• partial_ratio, ratio: scores from string matching
• prop: ratio/partial_ratio
• category: classification of work based on prop
• title: title of work (official)
• type: actual classification of work (print, poster)

[emilyyan04]

• Data: gallery_buchanan/text_works/ratio_scoring.csv
• Dataset: 619 English works (206 prints, 413 posters)
• Max accuracy: 68.3%