Information Extraction

1. Approaches
   • Manually crafted patterns
   • Machine learning Approaches
2. Tasks and subtasks
3. Evaluation

Intelligent Agent Summary

Information Extraction is the task of automatically extracting structured information from a unstructured and semi-structured documents.

Approaches

Manually crafted patterns

1. regular expression

2. Automata

3. NLP-based

Rule-based/knowledge based

[person][office]of[org]
[org][person][office]
[org]in[loc]

Machine learning Approaches

1. Label documents

2. Learn rules and patterns

Tasks and subtasks

Typical subtasks of IE include:

1. Named entity recognition:

   a Named entity Extraction

   b Coreference Resolution (指代消解)

   c Relationship Extraction

2. Semi-structured information Extraction:

   • Table extract: Finding and extracting tables from documents

   • Comment extract

3. Language and vocabulary analysis

   • Terminology extraction

Evaluation

1. Model solutions as labelled facts

2. Precision: $\frac{correctFoundFacts}{allFoundFacts}$

3. Recall: $\frac{correctFoundFacts}{allFacts}$

4. F-measure

$F-measure = \frac{2 \times precision \times recall}{Precision + Recall}$