4. NLP
Natural language processing (NLP)
a field of computer science … concerned with the
interactions
between computers and human (natural)
languages.
!
https://en.wikipedia.org/wiki/Natural_language_processing
5. NLP
“between computers and human (natural)
languages”
1. computer -> human language
2. human language -> computer
6. NLP trend
• Internet is huge and easily accessible resource
of information
• BUT - information is mainly unstructured
• usually simple scraping (scrapy) is sufficient, but
sometimes it is not
• NLP solves or helps in converting free text
(unstructured information) to structural form
13. Word, term, feature
• word <> term
• document or text chunk is an unit / entity / object!
• terms are features of the document!
• each term has properties:
• normalized form -> term.baseform + term.transformation
• position(s) in the document -> term.position(s)
• frequency -> term.frequency
14. Text, document, chunk
• what is document?
• text segmentation
• hard problem
• usually we consider whole document as one
unit (entity)
15. Terms, features
• converting words -> terms
• term frequency is usually the most important feature!
• how to get the list of terms with frequencies:
• preprocessing - e.g. remove all but words, remove stopwords,
tokenization (regexp)
• word normalization
dog ~ dogs zeleno ~ najzelenijih
• .tolower(), regexp, stemming, lemmatization
• much harder for inflectional languages, e.g. Croatian, see text-hr :)
16. Term weight - TF-IDF
• term frequency – inverse document frequency
• variables:
• t - term,
• d - one document
• D - all documents
• TF - is term frequency in a document function - i.e. measure on how
much information the term brings in one document
• IDF - is inverse document frequency of the term function - i.e.
inversed measure on how much information the term brings in all
documents (corpus)
17. Terms position, syntax
• sometimes term position is important
• neighbours, collocation, phrase extraction, NER
• from regexp to parsers
• syntax trees
• complex, cpu intensive
18. Terms position, syntax
In their public lectures they have even claimed that the only
evidence that Khufu built the pyramid is the graffiti found in the five
chambers.
20. Bag of words
• simplified and effective way to process
documents by:
• disregarding grammar (term.baseform?)
• disregarding word order (term.position)
• keeping only multiplicity (term.frequency)
21. Bag of words
• sparse matrix
• numbers can be:
• binary - 0/1
• simple term frequency
• weight - e.g. TF-IDF
22. Bag of words
• very simple -> very fast
• frequently used:
• in index servers
• in database for simple full-text-search
operations
• for processing of large datasets
24. Machine learning
• one of the Machine learning application is NLP
• after text is converted to entities with features,
machine learning techniques can be applied
25. Machine learning
• ML algorithm families categorisation
• supervised - classification (distinct), regression (numerical)
• unsupervised - clustering
• A lot of various methods/algorithm families, statistical,
probabilistic, …
decision trees, neural networks / deep learning,
support vector machines, bayesian networks,
markov models, genetic algorithms