Section 5 : Information retrieval and information extraction

Commentary

Section Goals

• To introduce two active research and application fields related to natural language text information processing (i.e., information retrieval and information extraction), to which a great deal of attention has recently been given as a result of popular applications, such as Web search engines (e.g., Google and Yahoo) and Web mining.

Learning Objectives

Learning Objective 1

• Outline the information retrieval (IR) and information extraction (IE) tasks, and compare the differences between them.
• Explain the principles of the language modeling approach based on "bag of words," and naive Bayes models for IR, produced from a Boolean keyword model.
• Describe the issues related to the IR systems' evaluation, such as precision, recall, and ROC curve measures.
• Summarize how NLP can help refine IR systems, and how to present result sets by clustering and relevance feedback.
• Describe the data structures and components in real IR systems, such as inverted index and vector space model.
• Describe the main steps and techniques involved in information extraction systems.
• Explain the following concepts or terms:
  • Information retrieval
  • Boolean keyword model
  • Bag of words
  • Precision
  • Recall
  • ROC curve
  • Relevance feedback
  • Document classification
  • Document clustering
  • Agglomerative clustering
  • K-means clustering
  • Vector space model
  • Inverted index
  • Stop words
  • TREC
  • Information extraction
  • Tokenization
  • Cascaded finite-state transducer

Objective Readings

Required readings:

Reading topics:

Information Retrieval, Information Extraction (see Sections 22.3 and 22.4 of AIMA3ed)

--. (2006). Special issue on Web information retrieval. Journal of Information Retrieval, 9(2). Springer (ISSN: 1573-7659)

Objective Questions

• What are the widely adopted main data structures and models in practical IR systems?
• What are the main techniques involved in information extraction?

Objective Activities

• Explore the most recent novel methods for IR systems, other than vector space models and inverted index. Report your findings and discuss them in the course conference.
• Explore recent, well-known open source tools for IR and IE, such as Lucene and GATE, and test them on your computer. Try to implement prototype systems based on them for testing.