Annotators

Annotators Guideline

How to read this section

All annotators in Spark NLP share a common interface, this is:

• Annotation -> Annotation(annotatorType, begin, end, result, metadata, embeddings)
• AnnotatorType -> some annotators share a type. This is not only figurative, but also tells about the structure of the metadata map in the Annotation. This is the one refered in the input and output of annotators.
• Inputs -> Represents how many and which annotator types are expected in setInputCols. These are column names of output of other annotators in the dataframe.
• Output -> Represents the type of the output in the column setOutputCol.

There are two types of annotators:

• Approach -> AnnotatorApproach extend Estimators, which are meant to be trained through fit()
• Model -> AnnotatorModel extend from Transfromers, which are meant to transform dataframes through transform()

Model suffix is explicitly stated when the annotator is the result of a training process. Some annotators, such as Tokenizer are transformers, but do not contain the word Model since they are not trained annotators.

Model annotators have a pretrained() on it’s static object, to retrieve the public pretrained version of a model.

• pretrained(name, language, extra_location) -> by default, pretrained will bring a default model, sometimes we offer more than one model, in this case, you may have to use name, language or extra location to download them.

The types are:

• DOCUMENT = “document”
• TOKEN = “token”
• CHUNK = “chunk”
• POS = “pos”
• WORD_EMBEDDINGS = “word_embeddings”
• SENTENCE_EMBEDDINGS = “sentence_embeddings”
• DATE = “date”
• ENTITY = “entity”
• CATEGORY = “category”
• SENTIMENT = “sentiment”
• NAMED_ENTITY = “named_entity”
• DEPENDENCY = “dependency”
• LABELED_DEPENDENCY = “labeled_dependency”

There are annotators freely available in the Open Source version of Spark-NLP. More are available in the licensed version of Spark NLP. Visit www.johnsnowlabs.com for more information about getting a license.

Spark-NLP Open Source

Tokenizer

Identifies tokens with tokenization open standards. A few rules will help customizing it if defaults do not fit user needs.
Output type: Token
Input types: Document
Reference: Tokenizer|TokenizerModel
Functions:

• setExceptions(StringArray): List of tokens to not alter at all. Allows composite tokens like two worded tokens that the user may not want to split.
• addException(String): Add a single exception
• setExceptionsPath(String): Path to txt file with list of token exceptions
• caseSensitiveExceptions(bool): Whether to follow case sensitiveness for matching exceptions in text
• contextChars(StringArray): List of 1 character string to rip off from tokens, such as parenthesis or question marks. Ignored if using prefix, infix or suffix patterns.
• splitChars(StringArray): List of 1 character string to split tokens inside, such as hyphens. Ignored if using infix, prefix or suffix patterns.
• splitPattern (String): pattern to separate from the inside of tokens. takes priority over splitChars.
• setTargetPattern: Basic regex rule to identify a candidate for tokenization. Defaults to \\S+ which means anything not a space
• setSuffixPattern: Regex to identify subtokens that are in the end of the token. Regex has to end with \\z and must contain groups (). Each group will become a separate token within the prefix. Defaults to non-letter characters. e.g. quotes or parenthesis
• setPrefixPattern: Regex to identify subtokens that come in the beginning of the token. Regex has to start with \\A and must contain groups (). Each group will become a separate token within the prefix. Defaults to non-letter characters. e.g. quotes or parenthesis
• addInfixPattern: Add an extension pattern regex with groups to the top of the rules (will target first, from more specific to the more general).
• minLength: Set the minimum allowed legth for each token
• maxLength: Set the maximum allowed legth for each token

Note: all these APIs receive regular expressions so please make sure that you escape special characters according to Java conventions.

Example:

Refer to the Tokenizer Scala docs for more details on the API.

tokenizer = Tokenizer() \
    .setInputCols(["sentences"]) \
    .setOutputCol("token") \
    .setSplitChars(['-']) \
    .setContextChars(['(', ')', '?', '!']) \
    .addException("New York") \
    .addException("e-mail")

val tokenizer = new Tokenizer()
    .setInputCols("sentence")
    .setOutputCol("token")
    .setContextChars(Array("(", ")", "?", "!"))
    .setSplitChars(Array('-'))
    .addException("New York")
    .addException("e-mail")

Normalizer

Text cleaning

Removes all dirty characters from text following a regex pattern and transforms words based on a provided dictionary
Output type: Token
Input types: Token
Reference: Normalizer | NormalizerModel
Functions:

• setCleanupPatterns(patterns): Regular expressions list for normalization, defaults [^A-Za-z]
• setLowercase(value): lowercase tokens, default true
• setSlangDictionary(path): txt file with delimited words to be transformed into something else

Example:

Refer to the Normalizer Scala docs for more details on the API.

normalizer = Normalizer() \
    .setInputCols(["token"]) \
    .setOutputCol("normalized")

val normalizer = new Normalizer()
    .setInputCols(Array("token"))
    .setOutputCol("normalized")

Stemmer

Returns hard-stems out of words with the objective of retrieving the meaningful part of the word
Output type: Token
Input types: Token
Reference: Stemmer

Example:

Refer to the Scala docs for more details on the API.

stemmer = Stemmer() \
    .setInputCols(["token"]) \
    .setOutputCol("stem")

val stemmer = new Stemmer()
    .setInputCols(Array("token"))
    .setOutputCol("stem")

Lemmatizer

Retrieves lemmas out of words with the objective of returning a base dictionary word
Output type: Token
Input types: Token
Input: abduct -> abducted abducting abduct abducts
Reference: Lemmatizer | LemmatizerModel
Functions: –

• setDictionary(path, keyDelimiter, valueDelimiter, readAs, options): Path and options to lemma dictionary, in lemma vs possible words format. readAs can be LINE_BY_LINE or SPARK_DATASET. options contain option passed to spark reader if readAs is SPARK_DATASET.

Example:

Refer to the Lemmatizer Scala docs for more details on the API.

lemmatizer = Lemmatizer() \
    .setInputCols(["token"]) \
    .setOutputCol("lemma") \
    .setDictionary("./lemmas001.txt")

val lemmatizer = new Lemmatizer()
    .setInputCols(Array("token"))
    .setOutputCol("lemma")
    .setDictionary("./lemmas001.txt")

StopWordsCleaner

This annotator excludes from a sequence of strings (e.g. the output of a Tokenizer, Normalizer, Lemmatizer, and Stemmer) and drops all the stop words from the input sequences.

Functions:

• setStopWords: The words to be filtered out. Array[String]
• setCaseSensitive: Whether to do a case sensitive comparison over the stop words.

Example:

Refer to the StopWordsCleaner Scala docs for more details on the API.

stop_words_cleaner = StopWordsCleaner() \
        .setInputCols(["token"]) \
        .setOutputCol("cleanTokens") \
        .setCaseSensitive(False) \
        .setStopWords(["this", "is", "and"])

val stopWordsCleaner = new StopWordsCleaner()
      .setInputCols("token")
      .setOutputCol("cleanTokens")
      .setStopWords(Array("this", "is", "and"))
      .setCaseSensitive(false)

NOTE: If you need to setStopWords from a text file, you can first read and convert it into an array of string:

# your stop words text file, each line is one stop word
stopwords = sc.textFile("/tmp/stopwords/english.txt").collect()
# simply use it in StopWordsCleaner
stopWordsCleaner = new StopWordsCleaner()
      .setInputCols("token")
      .setOutputCol("cleanTokens")
      .setStopWords(stopwords)
      .setCaseSensitive(false)

// your stop words text file, each line is one stop word
val stopwords = sc.textFile("/tmp/stopwords/english.txt").collect()
// simply use it in StopWordsCleaner
val stopWordsCleaner = new StopWordsCleaner()
      .setInputCols("token")
      .setOutputCol("cleanTokens")
      .setStopWords(stopwords)
      .setCaseSensitive(false)

RegexMatcher

Uses a reference file to match a set of regular expressions and put them inside a provided key. File must be comma separated.
Output type: Regex
Input types: Document
Input: the\\s\\w+, “followed by ‘the’”
Reference: RegexMatcher | RegexMatcherModel
Functions:

• setStrategy(strategy): Can be any of MATCH_FIRST|MATCH_ALL|MATCH_COMPLETE
• setRulesPath(path, delimiter, readAs, options): Path to file containing a set of regex,key pair. readAs can be LINE_BY_LINE or SPARK_DATASET. options contain option passed to spark reader if readAs is SPARK_DATASET.

Example:

Refer to the RegexMatcher Scala docs for more details on the API.

regex_matcher = RegexMatcher() \
    .setStrategy("MATCH_ALL") \
    .setOutputCol("regex")

val regexMatcher = new RegexMatcher()
    .setStrategy(strategy)
    .setInputCols(Array("document"))
    .setOutputCol("regex")

TextMatcher

Phrase matching

Annotator to match entire phrases (by token) provided in a file against a Document
Output type: Entity
Input types: Document, Token
Input: hello world, I am looking for you
Reference: TextMatcher | TextMatcherModel
Functions:

• setEntities(path, format, options): Provides a file with phrases to match. Default: Looks up path in configuration.
• path: a path to a file that contains the entities in the specified format.
• readAs: the format of the file, can be one of {ReadAs.LINE_BY_LINE, ReadAs.SPARK_DATASET}. Defaults to LINE_BY_LINE.
• options: a map of additional parameters. Defaults to {“format”: “text”}.

Example:

Refer to the TextMatcher Scala docs for more details on the API.

entity_extractor = TextMatcher() \
    .setInputCols(["inputCol"])\
    .setOutputCol("entity")\
    .setEntities("/path/to/file/myentities.txt")

val entityExtractor = new TextMatcher()
    .setInputCols("inputCol")
    .setOutputCol("entity")
    .setEntities("/path/to/file/myentities.txt")

Chunker

Meaningful phrase matching

This annotator matches a pattern of part-of-speech tags in order to return meaningful phrases from document

Output type: Chunk
Input types: Document, POS
Reference: Chunker
Functions:

• setRegexParsers(patterns): A list of regex patterns to match chunks, for example: Array(“‹DT›?‹JJ›*‹NN›”)
• addRegexParser(patterns): adds a pattern to the current list of chunk patterns, for example: “‹DT›?‹JJ›*‹NN›”

Example:

Refer to the Chunker Scala docs for more details on the API.

chunker = Chunker() \
    .setInputCols(["document", "pos"]) \
    .setOutputCol("chunk") \
    .setRegexParsers(["‹NNP›+", "‹DT|PP\\$›?‹JJ›*‹NN›"])

val chunker = new Chunker()
    .setInputCols(Array("document", "pos"))
    .setOutputCol("chunk")
    .setRegexParsers(Array("‹NNP›+", "‹DT|PP\\$›?‹JJ›*‹NN›"))

NGramGenerator

NGramGenerator annotator takes as input a sequence of strings (e.g. the output of a Tokenizer, Normalizer, Stemmer, Lemmatizer, and StopWordsCleaner). The parameter n is used to determine the number of terms in each n-gram. The output will consist of a sequence of n-grams where each n-gram is represented by a space-delimited string of n consecutive words with annotatorType CHUNK same as the Chunker annotator.

Output type: CHUNK
Input types: TOKEN
Reference: NGramGenerator
Functions:

• setN: number elements per n-gram (>=1)
• setEnableCumulative: whether to calculate just the actual n-grams or all n-grams from 1 through n
• setDelimiter: Glue character used to join the tokens

Example:

Refer to the NGramGenerator Scala docs for more details on the API.

ngrams_cum = NGramGenerator() \
            .setInputCols(["token"]) \
            .setOutputCol("ngrams") \
            .setN(2) \
            .setEnableCumulative(True)
            .setDelimiter("_") # Default is space

val nGrams = new NGramGenerator()
      .setInputCols("token")
      .setOutputCol("ngrams")
      .setN(2)
      .setEnableCumulative(true)
      .setDelimiter("_") // Default is space

DateMatcher

Date-time parsing

Reads from different forms of date and time expressions and converts them to a provided date format. Extracts only ONE date per sentence. Use with sentence detector for more matches.
Output type: Date
Input types: Document
Reference: DateMatcher
Reads the following kind of dates:

• 1978-01-28
• 1984/04/02
• 1/02/1980
• 2/28/79
• The 31st of April in the year 2008
• Fri, 21 Nov 1997
• Jan 21, ‘97
• Sun, Nov 21
• jan 1st
• next thursday
• last wednesday
• today
• tomorrow
• yesterday
• next week
• next month
• next year
• day after
• the day before
• 0600h
• 06:00 hours
• 6pm
• 5:30 a.m.
• at 5
• 12:59
• 23:59
• 1988/11/23 6pm
• next week at 7.30
• 5 am tomorrow

Functions:

• setDateFormat(format): SimpleDateFormat standard date output formatting. Defaults to yyyy/MM/dd

Example:

Refer to the DateMatcher Scala docs for more details on the API.

date_matcher = DateMatcher() \
    .setOutputCol("date") \
    .setDateFormat("yyyyMM")

val dateMatcher = new DateMatcher()
    .setFormat("yyyyMM")
    .setOutputCol("date")

SentenceDetector

Sentence Boundary Detector

Finds sentence bounds in raw text. Applies rules from Pragmatic Segmenter.
Output type: Document
Input types: Document
Reference: SentenceDetector
Functions:

• setCustomBounds(string): Custom sentence separator text
• setUseCustomOnly(bool): Use only custom bounds without considering those of Pragmatic Segmenter. Defaults to false. Needs customBounds.
• setUseAbbreviations(bool): Whether to consider abbreviation strategies for better accuracy but slower performance. Defaults to true.
• setExplodeSentences(bool): Whether to split sentences into different Dataset rows. Useful for higher parallelism in fat rows. Defaults to false.

Example:

Refer to the SentenceDetector Scala docs for more details on the API.

sentence_detector = SentenceDetector() \
    .setInputCols(["document"]) \
    .setOutputCol("sentence")

val sentenceDetector = new SentenceDetector()
    .setInputCols("document")
    .setOutputCol("sentence")

DeepSentenceDetector

Sentence Boundary Detector with Machine Learning

Finds sentence bounds in raw text. Applies a Named Entity Recognition DL model.
Output type: Document
Input types: Document, Token, Chunk
Reference: DeepSentenceDetector
Functions:

• setIncludePragmaticSegmenter(bool): Whether to include rule-based sentence detector as first filter. Defaults to false.
• setEndPunctuation(patterns): An array of symbols that deep sentence detector will consider as an end of sentence punctuation. Defaults to “.”, “!”, “?”

Example:

Refer to the DeepSentenceDetector Scala docs for more details on the API.

deep_sentence_detector = DeepSentenceDetector() \
    .setInputCols(["document", "token", "ner_con"]) \
    .setOutputCol("sentence") \
    .setIncludePragmaticSegmenter(True) \
    .setEndPunctuation([".", "?"])

val deepSentenceDetector = new DeepSentenceDetector()
    .setInputCols(Array("document", "token", "ner_con"))
    .setOutputCol("sentence")
    .setIncludePragmaticSegmenter(true)
    .setEndPunctuation(Array(".", "?"))

POSTagger

Part of speech tagger

Sets a POS tag to each word within a sentence. Its train data (train_pos) is a spark dataset of POS format values with Annotation columns.
Output type: POS
Input types: Document, Token
Reference: PerceptronApproach | PerceptronModel
Functions:

• setNIterations(number): Number of iterations for training. May improve accuracy but takes longer. Default 5.
• setPosColumn(colname): Column containing an array of POS Tags matching every token on the line.

Example:

Refer to the PerceptronApproach Scala docs for more details on the API.

pos_tagger = PerceptronApproach() \
    .setInputCols(["token", "sentence"]) \
    .setOutputCol("pos") \
    .setNIterations(2) \
    .fit(train_pos)

val posTagger = new PerceptronApproach()
    .setInputCols(Array("sentence", "token"))
    .setOutputCol("pos")
    .setNIterations(2)
    .fit(trainPOS)

ViveknSentimentDetector

Scores a sentence for a sentiment

Output type: sentiment
Input types: Document, Token
Reference: ViveknSentimentApproach | ViveknSentimentModel
Functions:

• setSentimentCol(colname): Column with sentiment analysis row’s result for training. If not set, external sources need to be set instead.
• setSentimentCol(colname): column with the sentiment result of every row. Must be ‘positive’ or ‘negative’
• setPruneCorpus(true): when training on small data you may want to disable this to not cut off infrequent words

Input: File or folder of text files of positive and negative data
Example:

Refer to the ViveknSentimentApproach Scala docs for more details on the API.

sentiment_detector = ViveknSentimentApproach() \
    .setInputCols(["sentence", "token"]) \
    .setOutputCol("sentiment")
    .setSentimentCol("sentiment_label")

val sentimentDetector = new ViveknSentimentApproach()
      .setInputCols(Array("token", "sentence"))
      .setOutputCol("vivekn")
      .setSentimentCol("sentiment_label")
      .setCorpusPrune(0)

SentimentDetector

Sentiment analysis

Scores a sentence for a sentiment
Output type: sentiment

Input types: Document, Token

Reference: SentimentDetector | SentimentDetectorModel
Functions:

• setDictionary(path, delimiter, readAs, options): path to file with list of inputs and their content, with such delimiter, readAs LINE_BY_LINE or as SPARK_DATASET. If latter is set, options is passed to spark reader.
• setPositiveMultiplier(double): Defaults to 1.0
• setNegativeMultiplier(double): Defaults to -1.0
• setIncrementMultiplier(double): Defaults to 2.0
• setDecrementMultiplier(double): Defaults to -2.0
• setReverseMultiplier(double): Defaults to -1.0

Input:

• superb,positive
• bad,negative
• lack of,revert
• very,increment
• barely,decrement

Example:

Refer to the SentimentDetector Scala docs for more details on the API.

sentiment_detector = SentimentDetector() \
    .setInputCols(["lemma", "sentence"]) \
    .setOutputCol("sentiment")

val sentimentDetector = new SentimentDetector
    .setInputCols(Array("token", "sentence"))
    .setOutputCol("sentiment")

WordEmbeddings

Word Embeddings lookup annotator that maps tokens to vectors

Output type: Word_Embeddings

Input types: Document, Token

Reference: WordEmbeddings | WordEmbeddingsModel
Functions:

• setStoragePath(path, format): sets word embeddings options.
  • path: word embeddings file
  • format: format of word embeddings files:
    • TEXT -> This format is usually used by Glove
    • BINARY -> This format is usually used by Word2Vec
• setCaseSensitive: whether to ignore case in tokens for embeddings matching

Example:

Refer to the WordEmbeddings Scala docs for more details on the API.

embeddings = WordEmbeddings()
      .setStoragePath("/tmp/glove.6B.100d.txt", "TEXT")\
      .setDimension(100)\
      .setStorageRef("glove_100d") \
      .setInputCols("document", "token") \
      .setOutputCol("embeddings")

val embeddings = new WordEmbeddings()
      .setStoragePath("/tmp/glove.6B.100d.txt", "TEXT)
      .setDimension(100)
      .setStorageRef("glove_100d") // Use or save this WordEmbeddings with storageRef
      .setInputCols("document", "token")
      .setOutputCol("embeddings")

There are also two convenient functions to retrieve the embeddings coverage with respect to the transformed dataset:

• withCoverageColumn(dataset, embeddingsCol, outputCol): Adds a custom column with word coverage stats for the embedded field: (coveredWords, totalWords, coveragePercentage). This creates a new column with statistics for each row.
• overallCoverage(dataset, embeddingsCol): Calculates overall word coverage for the whole data in the embedded field. This returns a single coverage object considering all rows in the field.

BertEmbeddings

BERT (Bidirectional Encoder Representations from Transformers) provides dense vector representations for natural language by using a deep, pre-trained neural network with the Transformer architecture

You can find the pre-trained models for BertEmbeddings in the Spark NLP Models repository

Output type: Word_Embeddings

Input types: Document, Token

Reference: BertEmbeddings

Refer to the BertEmbeddings Scala docs for more

How to use pretrained BertEmbeddings:

bert = BertEmbeddings.pretrained() \
      .setInputCols("sentence", "token") \
      .setOutputCol("bert")

val bert = BertEmbeddings.pretrained()
      .setInputCols("sentence", "token")
      .setOutputCol("bert")
      .setPoolingLayer(0) // 0, -1, or -2

ElmoEmbeddings

Computes contextualized word representations using character-based word representations and bidirectional LSTMs

You can find the pre-trained model for ElmoEmbeddings in the Spark NLP Models repository

Output type: Word_Embeddings

Input types: Document, Token

Reference: ElmoEmbeddings

Refer to the ElmoEmbeddings Scala docs for more

How to use pretrained ElmoEmbeddings:

# Online - Download the pretrained model
elmo = ElmoEmbeddings.pretrained()
      .setInputCols("sentence", "token") \
      .setOutputCol("elmo")

# Offline - Download the pretrained model manually and extract it
elmo = ElmoEmbeddings.load("/elmo_en_2.4.0_2.4_1580488815299") \
        .setInputCols("sentence", "token") \
        .setOutputCol("elmo")

val elmo = ElmoEmbeddings.pretrained()
      .setInputCols("sentence", "token")
      .setOutputCol("elmo")
      .setPoolingLayer("elmo") //  word_emb, lstm_outputs1, lstm_outputs2 or elmo

UniversalSentenceEncoder

The Universal Sentence Encoder encodes text into high dimensional vectors that can be used for text classification, semantic similarity, clustering and other natural language tasks.

Output type: SENTENCE_EMBEDDINGS

Input types: Document

Refer to the UniversalSentenceEncoder Scala docs for more

use = UniversalSentenceEncoder.pretrained() \
            .setInputCols("sentence") \
            .setOutputCol("use_embeddings")

val use = new UniversalSentenceEncoder()
      .setInputCols("document")
      .setOutputCol("use_embeddings")

SentenceEmbeddings

This annotator converts the results from WordEmbeddings, BertEmbeddings, or ElmoEmbeddings into sentence or document embeddings by either summing up or averaging all the word embeddings in a sentence or a document (depending on the inputCols).

Functions:

• setPoolingStrategy: Choose how you would like to aggregate Word Embeddings to Sentence Embeddings: AVERAGE or SUM

Output type: SENTENCE_EMBEDDINGS

Input types: Document

Refer to the SentenceEmbeddings Scala docs for more

sentence_embeddings = SentenceEmbeddings() \
            .setInputCols(["document", "embeddings"]) \
            .setOutputCol("sentence_embeddings") \
            .setPoolingStrategy("AVERAGE")

val embeddingsSentence = new SentenceEmbeddings()
      .setInputCols(Array("document", "embeddings"))
      .setOutputCol("sentence_embeddings")
      .setPoolingStrategy("AVERAGE")

NOTE:

If you choose document as your input for Tokenizer, WordEmbeddings/BertEmbeddings, and SentenceEmbeddings then it averages/sums all the embeddings into one array of embeddings. However, if you choose sentence as inputCols then for each sentence SentenceEmbeddings generates one array of embeddings.

TIP:

How to explode and convert these embeddings into Vectors or what’s known as Feature column so it can be used in Spark ML regression or clustering functions:

from org.apache.spark.ml.linal import Vector, Vectors
from pyspark.sql.functions import udf
# Let's create a UDF to take array of embeddings and output Vectors
@udf(Vector)
def convertToVectorUDF(matrix):
    return Vectors.dense(matrix.toArray.map(_.toDouble))


# Now let's explode the sentence_embeddings column and have a new feature column for Spark ML
pipelineDF.select(explode("sentence_embeddings.embeddings").as("sentence_embedding"))
.withColumn("features", convertToVectorUDF("sentence_embedding"))

import org.apache.spark.ml.linalg.{Vector, Vectors}

// Let's create a UDF to take array of embeddings and output Vectors
val convertToVectorUDF = udf((matrix : Seq[Float]) => {
    Vectors.dense(matrix.toArray.map(_.toDouble))
})

// Now let's explode the sentence_embeddings column and have a new feature column for Spark ML
pipelineDF.select(explode($"sentence_embeddings.embeddings").as("sentence_embedding"))
.withColumn("features", convertToVectorUDF($"sentence_embedding"))

ChunkEmbeddings

This annotator utilizes WordEmbeddings or BertEmbeddings to generate chunk embeddings from either Chunker, NGramGenerator, or NerConverter outputs.

Functions:

• setPoolingStrategy: Choose how you would like to aggregate Word Embeddings to Sentence Embeddings: AVERAGE or SUM

Output type: CHUNK

Input types: CHUNK, Word_Embeddings

Refer to the ChunkEmbeddings Scala docs for more

chunk_embeddings = ChunkEmbeddings() \
            .setInputCols(["chunk", "embeddings"]) \
            .setOutputCol("chunk_embeddings") \
            .setPoolingStrategy("AVERAGE")

val chunkSentence = new ChunkEmbeddings()
      .setInputCols(Array("chunk", "embeddings"))
      .setOutputCol("chunk_embeddings")
      .setPoolingStrategy("AVERAGE")

TIP:

How to explode and convert these embeddings into Vectors or what’s known as Feature column so it can be used in Spark ML regression or clustering functions:

from org.apache.spark.ml.linal import Vector, Vectors
from pyspark.sql.functions import udf

// Let's create a UDF to take array of embeddings and output Vectors
@udf(Vector)
def convertToVectorUDF(matrix):
    return Vectors.dense(matrix.toArray.map(_.toDouble))

import org.apache.spark.ml.linalg.{Vector, Vectors}

// Let's create a UDF to take array of embeddings and output Vectors
val convertToVectorUDF = udf((matrix : Seq[Float]) => {
    Vectors.dense(matrix.toArray.map(_.toDouble))
})

// Now let's explode the sentence_embeddings column and have a new feature column for Spark ML
pipelineDF.select(explode($"chunk_embeddings.embeddings").as("chunk_embeddings_exploded"))
.withColumn("features", convertToVectorUDF($"chunk_embeddings_exploded"))

ClassifierDL

Multi-class Text Classification

ClassifierDL is a generic Multi-class Text Classification. ClassifierDL uses the state-of-the-art Universal Sentence Encoder as an input for text classifications. The ClassifierDL annotator uses a deep learning model (DNNs) we have built inside TensorFlow and supports up to 50 classes

NOTE: This annotator accepts a label column of a single item in either type of String, Int, Float, or Double.

NOTE: UniversalSentenceEncoder and SentenceEmbeddings can be used for the inputCol

Output type: CATEGORY

Input types: SENTENCE_EMBEDDINGS

Refer to the ClassifierDLApproach Scala docs for more

Refer to the ClassifierDLModel Scala docs for more

docClassifier = ClassifierDLApproach()\
      .setInputCols("sentence_embeddings")\
      .setOutputCol("category")\
      .setLabelColumn("label")\
      .setBatchSize(64)\
      .setMaxEpochs(20)\
      .setLr(0.5)\
      .setDropout(0.5)

val docClassifier = new ClassifierDLApproach()
      .setInputCols("sentence_embeddings")
      .setOutputCol("category")
      .setLabelColumn("label")
      .setBatchSize(64)
      .setMaxEpochs(20)
      .setLr(5e-3f)
      .setDropout(0.5f)

Please refer to existing notebooks for more examples.

NER CRF

Named Entity Recognition CRF annotator

This Named Entity recognition annotator allows for a generic model to be trained by utilizing a CRF machine learning algorithm. Its train data (train_ner) is either a labeled or an external CoNLL 2003 IOB based spark dataset with Annotations columns. Also the user has to provide word embeddings annotation column.
Optionally the user can provide an entity dictionary file for better accuracy
Output type: Named_Entity
Input types: Document, Token, POS, Word_Embeddings
Reference: NerCrfApproach | NerCrfModel
Functions:

• setLabelColumn: If DatasetPath is not provided, this Seq[Annotation] type of column should have labeled data per token
• setMinEpochs: Minimum number of epochs to train
• setMaxEpochs: Maximum number of epochs to train
• setL2: L2 regularization coefficient for CRF
• setC0: c0 defines decay speed for gradient
• setLossEps: If epoch relative improvement lass than this value, training is stopped
• setMinW: Features with less weights than this value will be filtered out
• setExternalFeatures(path, delimiter, readAs, options): Path to file or folder of line separated file that has something like this: Volvo:ORG with such delimiter, readAs LINE_BY_LINE or SPARK_DATASET with options passed to the latter.
• setEntities: Array of entities to recognize
• setVerbose: Verbosity level
• setRandomSeed: Random seed

Example:

Refer to the NerCrfApproach Scala docs for more details on the API.

nerTagger = NerCrfApproach()\
    .setInputCols(["sentence", "token", "pos"])\
    .setLabelColumn("label")\
    .setOutputCol("ner")\
    .setMinEpochs(1)\
    .setMaxEpochs(20)\
    .setLossEps(1e-3)\
    .setDicts(["ner-corpus/dict.txt"])\
    .setL2(1)\
    .setC0(1250000)\
    .setRandomSeed(0)\
    .setVerbose(2)
    .fit(train_ner)

val nerTagger = new NerCrfApproach()
    .setInputCols("sentence", "token", "pos")
    .setLabelColumn("label")
    .setMinEpochs(1)
    .setMaxEpochs(3)
    .setC0(34)
    .setL2(3.0)
    .setOutputCol("ner")
    .fit(trainNer)

NER DL

Named Entity Recognition Deep Learning annotator

This Named Entity recognition annotator allows to train generic NER model based on Neural Networks. Its train data (train_ner) is either a labeled or an external CoNLL 2003 IOB based spark dataset with Annotations columns. Also the user has to provide word embeddings annotation column.
Neural Network architecture is Char CNNs - BiLSTM - CRF that achieves state-of-the-art in most datasets.
Output type: Named_Entity
Input types: Document, Token, Word_Embeddings
Reference: NerDLApproach | NerDLModel
Functions:

• setLabelColumn: If DatasetPath is not provided, this Seq[Annotation] type of column should have labeled data per token
• setMaxEpochs: Maximum number of epochs to train
• setLr: Initial learning rate
• setPo: Learning rate decay coefficient. Real Learning Rate: lr / (1 + po * epoch)
• setBatchSize: Batch size for training
• setDropout: Dropout coefficient
• setVerbose: Verbosity level
• setRandomSeed: Random seed

Note: Please check here in case you get an IllegalArgumentException error with a description such as: Graph [parameter] should be [value]: Could not find a suitable tensorflow graph for embeddings dim: [value] tags: [value] nChars: [value]. Generate graph by python code in python/tensorflow/ner/create_models before usage and use setGraphFolder Param to point to output.

Example:

Refer to the NerDLApproach Scala docs for more details on the API.

nerTagger = NerDLApproach()\
    .setInputCols(["sentence", "token", "embeddings"])\
    .setLabelColumn("label")\
    .setOutputCol("ner")\
    .setMaxEpochs(10)\
    .setRandomSeed(0)\
    .setVerbose(2)
    .fit(train_ner)

val nerTagger = new NerDLApproach()
        .setInputCols("sentence", "token", "embeddings")
        .setOutputCol("ner")
        .setLabelColumn("label")
        .setMaxEpochs(120)
        .setRandomSeed(0)
        .setPo(0.03f)
        .setLr(0.2f)
        .setDropout(0.5f)
        .setBatchSize(9)
        .setVerbose(Verbose.Epochs)
        .fit(trainNer)

Norvig SpellChecker

This annotator retrieves tokens and makes corrections automatically if not found in an English dictionary
Output type: Token
Inputs: Any text for corpus. A list of words for dictionary. A comma separated custom dictionary. Input types: Tokenizer
Train Data: train_corpus is a spark dataset of text content
Reference: NorvigSweetingApproach | NorvigSweetingModel
Functions:

• setDictionary(path, tokenPattern, readAs, options): path to file with properly spelled words, tokenPattern is the regex pattern to identify them in text, readAs LINE_BY_LINE or SPARK_DATASET, with options passed to Spark reader if the latter is set.
• setCaseSensitive(boolean): defaults to false. Might affect accuracy
• setDoubleVariants(boolean): enables extra check for word combinations, more accuracy at performance
• setShortCircuit(boolean): faster but less accurate mode
• setWordSizeIgnore(int): Minimum size of word before moving on. Defaults to 3.
• setDupsLimit(int): Maximum duplicate of characters to account for. Defaults to 2.
• setReductLimit(int): Word reduction limit. Defaults to 3
• setIntersections(int): Hamming intersections to attempt. Defaults to 10.
• setVowelSwapLimit(int): Vowel swap attempts. Defaults to 6.

Example:

Refer to the NorvigSweetingApproach Scala docs for more details on the API.

spell_checker = NorvigSweetingApproach() \
    .setInputCols(["token"]) \
    .setOutputCol("checked") \
    .setDictionary("coca2017.txt", "[a-zA-Z]+")

val symSpellChecker = new NorvigSweetingApproach()
      .setInputCols("token")
      .setOutputCol("checked")
      .setDictionary("coca2017.txt", "[a-zA-Z]+")

Symmetric SpellChecker

This spell checker is inspired on Symmetric Delete algorithm. It retrieves tokens and utilizes distance metrics to compute possible derived words
Output type: Token
Inputs: Any text for corpus. A list of words for dictionary. A comma separated custom dictionary. Input types: Tokenizer
Train Data: train_corpus is a spark dataset of text content
Reference: SymmetricDeleteApproach | SymmetricDeleteModel
Functions:

• setDictionary(path, tokenPattern, readAs, options): Optional dictionary of properly written words. If provided, significantly boosts spell checking performance
• setMaxEditDistance(distance): Maximum edit distance to calculate possible derived words. Defaults to 3.

Example:

Refer to the SymmetricDeleteApproach Scala docs for more details on the API.

spell_checker = SymmetricDeleteApproach() \
    .setInputCols(["token"]) \
    .setOutputCol("spell") \
    .fit(train_corpus)

val spellChecker = new SymmetricDeleteApproach()
    .setInputCols(Array("normalized"))
    .setOutputCol("spell")
    .fit(trainCorpus)

Dependency Parsers

Dependency parser provides information about word relationship. For example, dependency parsing can tell you what the subjects and objects of a verb are, as well as which words are modifying (describing) the subject. This can help you find precise answers to specific questions. The following diagram illustrates a dependency-style analysis using the standard graphical method favored in the dependency-parsing community.

Relations among the words are illustrated above the sentence with directed, labeled arcs from heads to dependents. We call this a typed dependency structure because the labels are drawn from a fixed inventory of grammatical relations. It also includes a root node that explicitly marks the root of the tree, the head of the entire structure. [1]

Dependency Parser

Unlabeled grammatical relation

Unlabeled parser that finds a grammatical relation between two words in a sentence. Its input is a directory with dependency treebank files.
Output type: Dependency
Input types: Document, POS, Token
Reference: DependencyParserApproach | DependencyParserModel
Functions:

• setNumberOfIterations: Number of iterations in training, converges to better accuracy
• setDependencyTreeBank: Dependency treebank folder with files in Penn Treebank format
• conllU: Path to a file in CoNLL-U format

Example:

Refer to the DependencyParserApproach Scala docs for more details on the API.

dependency_parser = DependencyParserApproach() \
            .setInputCols(["sentence", "pos", "token"]) \
            .setOutputCol("dependency") \
            .setDependencyTreeBank("file://parser/dependency_treebank") \
            .setNumberOfIterations(10)

val dependencyParser = new DependencyParserApproach()
    .setInputCols(Array("sentence", "pos", "token"))
    .setOutputCol("dependency")
    .setDependencyTreeBank("parser/dependency_treebank")
    .setNumberOfIterations(10)

Typed Dependency Parser

Labeled grammatical relation

Labeled parser that finds a grammatical relation between two words in a sentence. Its input is a CoNLL2009 or ConllU dataset.
Output type: Labeled Dependency
Input types: Token, POS, Dependency
Reference: TypedDependencyParserApproach | TypedDependencyParserModel
Functions:

• setNumberOfIterations: Number of iterations in training, converges to better accuracy
• setConll2009: Path to a file in CoNLL 2009 format
• setConllU: Path to a file in CoNLL-U format

Example:

Refer to the TypedDependencyParserApproach Scala docs for more details on the API.

typed_dependency_parser = TypedDependencyParserApproach() \
            .setInputCols(["token", "pos", "dependency"]) \
            .setOutputCol("labdep") \
            .setConll2009("file://conll2009/eng.train") \
            .setNumberOfIterations(10)

val typedDependencyParser = new TypedDependencyParserApproach()
    .setInputCols(Array("token", "pos", "dependency"))
    .setOutputCol("labdep")
    .setConll2009("conll2009/eng.train"))

References

[1] Speech and Language Processing. Daniel Jurafsky & James H. Martin. 2018