5-Contextual Information

LSA stands for Latent Semantic Analysis. It factorizes the term-document-matrix to obtain lower-dimensional dense features as word representations.

One approach is to use the term-document-matrix to represent words based on their occurrences in different documents.

Neighboring words can be used to create cooccurrence vectors that represent the relationships between words based on their proximity in text.

<p>An n-gram is a subsequence of contiguous items.</p> Signup and view all the answers

<p>Example bigrams on text: Introduction to text mining, mining.</p> Signup and view all the answers

<p>N-grams are more tolerant to declension and cases, and composed words have a similar vector to the part words.</p> Signup and view all the answers

<p>The bag-of-words representation does not capture word similarity.</p> Signup and view all the answers

<p>To have word representations that are of lower dimensionality (100-500) than our vocabulary.</p> Signup and view all the answers

<p>Document vectors can be interpreted as a term-document matrix.</p> Signup and view all the answers

<p>Train a neural network to predict a word given the preceding and following words (CBOW) or predict the preceding and following words given a word (Skip-Gram).</p> Signup and view all the answers

<p>Continuous Bag of Words (CBOW) and Skip-Gram.</p> Signup and view all the answers

<p>100 to 1000 dimensions.</p> Signup and view all the answers

<p>To use this layer as a feature for predicting words based on similarity to a 'target' vector.</p> Signup and view all the answers

<p>Words, not documents.</p> Signup and view all the answers