Text Analytics Toolbox
Analyze and model text data
Text Analytics Toolbox™ provides algorithms and visualizations for preprocessing, analyzing, and modeling text data. Models created with the toolbox can be used in applications such as sentiment analysis, predictive maintenance, and topic modeling.
Text Analytics Toolbox includes tools for processing raw text from sources such as equipment logs, news feeds, surveys, operator reports, and social media. You can extract text from popular file formats, preprocess raw text, extract individual words, convert text into numerical representations, and build statistical models.
Using machine learning techniques such as LSA, LDA, and word embeddings, you can find clusters and create features from high-dimensional text datasets. Features created with Text Analytics Toolbox can be combined with features from other data sources to build machine learning models that take advantage of textual, numeric, and other types of data.
Get Started:
Extract Text Data
Import text data into MATLAB® from single files or large collections of files, including PDF, HTML, and Microsoft® Word® and Excel® files.
Extracting text from a collection of Microsoft Word documents.
Visualize Text
Visually explore text datasets using word clouds and text scatter plots.
Text scatter plot showing the relative frequency of words using font size and color.
Language Support
Text Analytics Toolbox provides language specific preprocessing capabilities for English, Japanese, German, and Korean. Most functions also work with text in other languages.
Import, prepare, and analyze Japanese text.
Clean Text Data
Apply high-level filtering functions to remove extraneous content such as URLs, HTML tags, and punctuations, and correct spellings.
Simplify raw text (left) to work with the most meaningful words (right).
Filter Stop Words and Normalize Words to Root Form
Prioritize meaningful text data in your analysis by filtering out common words, words that appear too frequently or infrequently, and very long or very short words. Reduce the vocabulary and focus on the broader sense or sentiment of a document by stemming words to their root form or lemmatizing them to their dictionary form.
Removing stop words like “a” and “of” from documents.
Identify Tokens, Sentences, and Parts-of-Speech
Automatically split raw text into a collection of words using a tokenization algorithm. Add sentence boundaries, part-of-speech details, and other relevant information for context.
Financial charts and technical indicators.
Word and N-Gram Counting
Calculate word frequency statistics to represent text data numerically.
Identify and visualize the most frequently occurring words in a model.
Word Embedding and Encoding
Train word-embedding models such as word2vec continuous bag-of-words (CBOW) and skip-gram models. Import pretrained models including fastText and GloVe.
Visualize clusters in a text scatter plot using word embedding.
Topic Modeling
Discover and visualize underlying patterns, trends, and complex relationships in large sets of text data using machine learning algorithms such as latent Dirichlet allocation (LDA) and latent semantic analysis (LSA).
Identifying topics in storm report data.
Document Summarization
Extract summary from one or more documents automatically and evaluate similarity and importance of documents.
Extract summary from text.
Deep Learning with Text Data
Perform sentiment analysis and classification with deep learning networks such as long short-term memory networks (LSTMs).
Sentiment Analysis
Identify the attitudes and opinions expressed in text data to categorize statements as being positive, neutral, or negative. Build models that can predict sentiment in real time.
Identifying words that predict positive and negative sentiment.
Text Classification
Classify text descriptions using word embeddings that can identify categories of text through deep learning.
Training a deep neural network to classify text data.
Text Generation
Use deep learning to generate new text based on observed text.
Text generation using Jane Austen’s Pride and Prejudice and a deep learning LSTM network.
Document Summarization
Extract summaries from text
Document Importance
Evaluate text importance using TextRank, LexRank, and Maximal Marginal Relevance (MMR)
Document Similarity
Evaluate text similarity using BM25, Cosine similarity, BLEU, or ROUGE scoring algorithms
Spelling Correction
Correct spelling of English, German, and Korean text
See release notes for details on any of these features and corresponding functions.
Sentiment Analysis with Deep Learning
Analyze the sentiment of live Twitter data to understand how a given term is perceived.
