Natural Language Processing (NLP), Top Exam Questions and answers, verified.

Natural Language Processing (NLP), Top Exam Questions and answers, verified. Artificial Intelligence - -A computer performing tasks that a human can do NLP Sentiment analysis is a form of... - -classification NLP topic modeling is a form of... - -Dimensionality reduction Tokenization - -Splitting raw text into small, indivisible units for processing. Units can be words, sentences, n-grams (n-word combos), other characters defined by regex Stop words - -Words that have very little semantic value Stemming and Lemmatization - -Cut word down to base form Stemming- uses rough heuristics to reduce words to base Lemmatization- uses vocabulary and morphological analysis (makes run, runs, running, and ran all the same) Named Entity Recognition - -Identifies and tags named entities in text (people, places, organizations, phone numbers, emails, etc) aka entity extraction Compound term extraction - -extracting and tagging compound words or phrases in text Levenshtein distance - -Minimum number of operations to get from one word to another. One way of quantifying word similarity Levenshtein operations - -Deletions (delete a character) Insertions (insert a character) Mutation (change a character) Corpus - -Collection of texts Bag of words model - -- Simplified representation of text, where each document is recognized as a bag of its words - Grammar and word order are disregarded, but multiplicity is kept Cosine similarity - -Way to quantify the similarity between documents 1. Put each document in vector format 2. Find the cosine of the angle between the documents Term frequency-inverse document frequency - -(term frequency) * (inverse document frequency) Term frequency - -Term count/total terms Inverse document frequency - -- Considers how common a word is among all the documents - Rare words get additional weight Which classification models suffer from curse of dimensionality? - -KNN, SVM, linear models (linear/logisitic regression), decision trees Distance-based models feature selection - -Removes features that aren't helpful (might not be predictive of target and may not have a lot of variation) Art (try fitting with some features and changing it and comparing, regularization, feature importance scores) Feature extraction - -Uses information from all features, but creates artificial new features that are composites (uses information from all features, may put more weight on certain features) SVD - -Singular value decomposition, type of matrix decomposition (just multiplication) easy to compute & doesn't require square matrix matrices are our data, rows are observations, columns are features High level, what does SVD do? - -Generalization of eigendecomposition for rectangular matrices PCA - -Principal Components Analysis unsupervised technique care about the direction of maximal variation b/c that represents the differences in our observations and that helps us in our clustering/classification tasks What is PCA used for? - -Dropping the components that explain the least variance (uses SVD behind the scenes). Dimensionality reduction Document-term matrix - -rows = documents