QBIC Vector Based Model

Content-Based Document Retrieval

Like content-based image retrieval, content-based document retrieval goes through similar process of data flow. Documents are processed and features were extracted. A similarity function is applied based on the features extracted to determine relevance of document.

Pre-Processing:

  • Extract text and structure from raw document (Word document, html, etc.)
  • Remove stop words from document (the, at, all, etc.)
  • Stemming - Converting words to stems (processing => process)

Features Extraction:
- Words are projected into vectors.

Computing Relevance (Similarity):

  • Vector Based Model
    - Document and query is represented by vector.
    - Elements of vector are frequency of words.
    - Similarity is calculated by calculating distance between the query vector and document vector.
  • Probabilistic Model
    - Relevance is based on users feedback or ranking of the vector model.
    - Relevance of the document is represented as probability.