AI Cookbook for Libraries

AI Recipes

The recipes are classified by technical domain, and for each recipe, classical use cases in libraries are provided.

Natural Language Processing (NLP)

Recipe #1: Named entity extraction & linking

• enrichment of digital collections (creating new metadata such as person names, organizations, locations) for information retrieval, scientific objectives, etc.
• establishing links between documents and authority data

Recipe #2: Topic Modeling

• understanding large collections of unstructured text documents (text mining use case)
• enrichment of digital collections with topics (information retrieval)

Recipe #3: Text Classification, computational semantics

• enrichment of digitized collections with genres (novel, poetry, science, ...) or other classification schema (Dewey...)
• cataloguing of born-digital materials

Recipe #4: Language models

• creation and use of language models for NLP tasks

Recipe #5: OCR Post-correction

• correction of ocerized collections

---

Optical Character Recognition (OCR)

Recipe #1: Training an OCR on a specific corpora

Recipe #2: Evaluating OCR quality

---

Handwritten Text Recognition (HTR)

Recipe #1: HTR

• HTR for full text indexing
• HTR for transcription

Recipe #2: Authorship attribution

• Attribution of authorship based on handwriting
• Attribution of style of writing (uncial, carolingian, etc.)

---

Document Analysis

Recipe #1: Document Classification

• pre-treatment of uncatalogued collections (filtering, preindexing, ...) based on the document type (letter, typewritten, map, etc.)
• enrichment of digital catalogued collections with document types

Recipe #2: Page Segmentation

• extraction of text from heritage documents for full text indexing
• segmentation of illustrations from heritage documents

Recipe #3: Article Recognition for newspapers, dictionaries, sales catalogues (arts, coins/medals...)

Recipe #4: Tables Recognition (census tables, stock market, etc.)

Recipe #5: Date Attribution

---

Computer vision

Recipe #1: Images Classification

• pre-treatment of uncatalogued collections (filtering, preindexing, ...)
• enrichment of digital catalogued collections (creating new metadata) for information retrieval, scientific objectives, etc.

Recipe #2: Object Detection and Face Detection, Instance Search

• enrichment of digital collections
• data analysis, visual studies

Recipe #3: Image Similarity

• information retrieval based on visual similarity
• navigating massive digital collections
• curation of digital collections (duplicate detection, variant detection)

Recipe #4: Video Indexing

• pre-treatment of uncatalogued collections (cutting into sequences, ...)
• enrichment of digital catalogued collections (creating new metadata): object detection, scene classification, etc.
• subtitles transcription (OCR)

Recipe #5: Audio Transcription

• speech to text
• speaker detection

---

To add a new AI recipe, use the recipe template