Term Extraction Transformation

The Term Extraction transformation extracts terms from text in a transformation input column, and then writes the terms to a transformation output column. The transformation works only with English text and it uses its own English dictionary and linguistic information about English.

You can use the Term Extraction transformation to discover the content of a data set. For example, text that contains e-mail messages may provide useful feedback about products, so that you could use the Term Extraction transformation to extract the topics of discussion in the messages, as a way of analyzing the feedback.

The Term Extraction transformation can extract nouns only, noun phrases only, or both nouns and noun phases. A noun is a single noun; a noun phrases is at least two words, of which one is a noun and the other is a noun or an adjective. For example, if the transformation uses the nouns-only option, it extracts terms like bicycle and landscape; if the transformation uses the noun phrase option, it extracts terms like new blue bicycle, bicycle helmet, and boxed bicycles.

Articles and pronouns are not extracted. For example, the Term Extraction transformation extracts the term bicycle from the text the bicycle, my bicycle, and that bicycle.

The Term Extraction transformation normalizes words so that the capitalized and noncapitalized versions of words are not treated as different terms. For example, in the text You see many bicycles in Seattle and Bicycles are blue, bicycles and Bicycles are recognized as the same term and the transformation keeps only bicycle. Proper nouns and words that are not listed in the internal dictionary are not normalized.

The Term Extraction transformation also stems nouns to extract only the singular form of a noun. For example, the transformation extracts man from men, mouse from mice, and bicycle from bicycles. The transformation uses its dictionary to stem nouns. Gerunds are treated as nouns if they are in the dictionary.

The Term Extraction transformation can work only with text in a column that has either the DT_WSTR or the DT_NTEXT data type. If a column contains text but does not have one of these data types, the Data Conversion transformation can be used to add a column with the DT_WSTR or DT_NTEXT data type to the data flow and copy the column values to the new column. The output from the Data Conversion transformation can then be used as the input to the Term Extraction transformation. For more information, see Data Conversion Transformation.

The Term Extraction transformation generates a score for each term that it extracts. The score can be either a TFIDF value or the raw frequency, meaning the number of times the normalized term appears in the input. In either case, the score is represented by a real number that is greater than 0. For example, the TFIDF score might have the value 0.5, and the frequency would be a value like 1.0 or 2.0.

Optionally, the Term Extraction transformation can reference a column in a table that contains exclusion terms, meaning terms that the transformation should skip when it extracts terms from a data set. This is useful when a set of terms has already been identified as inconsequential in a particular business and industry, typically because the term occurs with such high frequency that it becomes a noise word. For example, when extracting terms from a data set that contains customer support information about a particular brand of cars, the brand name itself might be excluded because it is mentioned too frequently to have significance. Therefore, the values in the exclusion list must be customized to the data set you are working with.

When you add a term to the exclusion list, all the terms—words or noun phrases—that contain the term are also excluded. For example, if the exclusion list includes the single word data, then all the terms that contain this word, such as data, data mining, data integrity, and data validation will also be excluded. If you want to exclude only compounds that contain the word data, you must explicitly add those compound terms to the exclusion list. For example, if you want to extract incidences of data, but exclude data validation, you would add data validation to the exclusion list, and make sure that data is removed from the exclusion list.

The reference table must be a table in a SQL Server 2000, a SQL Server, or an Access database. The Term Extraction transformation uses a separate OLE DB connection to connect to the reference table. For more information, see OLE DB Connection Manager.

The Term Extraction transformation works in a fully precached mode. At run time, the Term Extraction transformation reads the exclusion terms from the reference table and stores them in its private memory before it processes any transformation input rows.

If the extracted terms are written to a table, they can be used by other lookup transformation such as the Term Lookup, Fuzzy Lookup, and Lookup transformations.

The output of the Term Extraction transformation includes only two columns. One column contains the extracted terms and the other column contains the score. The default names of the columns are Term and Score. Because the text column in the input may contain multiple terms, the output of the Term Extraction transformation typically has more rows than the input.

The Text Extraction transformation uses internal algorithms and statistical models to generate its results. You may have to run the Term Extraction transformation several times and examine the results to configure the transformation to generate the type of results that works for your text mining solution.

The Term Extraction transformation has one regular input, one output, and one error output.

Extracting Terms from Text

To extract terms from text, the Term Extraction transformation performs the following tasks.

Tokenizing Text

First, the Term Extraction transformation identifies words by performing the following tasks:

  • Separating text into words by using spaces, line breaks, and other word terminators in the English language. For example, punctuation marks such as ? and : are word-breaking characters.

  • Preserving words that are connected by hyphens or underscores. For example, the words copy-protected and read-only remain one word.

  • Keeping intact acronyms that include periods. For example, the A.B.C Company would be tokenized as ABC and Company.

  • Splitting words on special characters. For example, the word date/time is extracted as date and time, (bicycle) as bicycle, and C# is treated as C. Special characters are discarded and cannot be lexicalized.

  • Recognizing when special characters such as the apostrophe should not split words. For example, the word bicycle's is not split into two words, and yields the single term bicycle (noun).

  • Splitting time expressions, monetary expressions, e-mail addresses, and postal addresses. For example, the date January 31, 2004 is separated into the three tokens January, 31, and 2004.

Tagging Words

Second, the Term Extraction transformation tags words as one of the following parts of speech:

  • A noun in the singular form. For example, bicycle and potato.

  • A noun in the plural form. For example, bicycles and potatoes. All plural nouns that are not lemmatized are subject to stemming.

  • A proper noun in the singular form. For example, April and Peter.

  • A proper noun in the plural form. For example Aprils and Peters. For a proper noun to be subject to stemming, it must be a part of the internal lexicon, which is limited to standard English words.

  • An adjective. For example, blue.

  • A comparative adjective that compares two things. For example, higher and taller.

  • A superlative adjective that identifies a thing that has a quality above or below the level of at least two others. For example, highest and tallest.

  • A number. For example, 62 and 2004.

Words that are not one of these parts of speech are discarded. For example, verbs and pronouns are discarded.

Note

The tagging of parts of speech is based on a statistical model and the tagging may not be completely accurate.

If the Term Extraction transformation is configured to extract only nouns, only the words that are tagged as singular or plural forms of nouns and proper nouns are extracted.

If the Term Extraction transformation is configured to extract only noun phrases, words that are tagged as nouns, proper nouns, adjectives, and numbers may be combined to make a noun phrase, but the phrase must include at least one word that is tagged as a singular or plural form of a noun or a proper noun. For example, the noun phrase highest mountain combines a word tagged as a superlative adjective (highest) and a word tagged as noun (mountain).

If the Term Extraction is configured to extract both nouns and noun phrases, both the rules for nouns and the rules for noun phrases apply. For example, the transformation extracts bicycle and beautiful blue bicycle from the text many beautiful blue bicycles.

Note

The extracted terms remain subject to the maximum term length and frequency threshold that the transformation uses.

Stemming Words

Third, the Term Extraction transformation stems words to their dictionary form as shown in these examples by using the dictionary internal to the Term Extraction transformation.

  • Removing s from nouns. For example, bicycles becomes bicycle.

  • Removing es from nouns. For example, stories becomes story.

  • Retrieving the singular form for irregular nouns from the dictionary. For example, geese becomes goose.

Normalizing Words

The Term Extraction transformation normalizes terms that are capitalized only because of their position in a sentence, and uses their non-capitalized form instead. For example, in the phrases Dogs chase cats and Mountain paths are steep, Dogs and Mountain would be normalized to dog and mountain.

Using Case-Sensitive Normalization

The Term Extraction transformation can be configured to consider lowercase and uppercase words as either distinct terms, or as different variants of the same term.

  • If the transformation is configured to recognize differences in case, terms like Method and method are extracted as two different terms. Capitalized words that are not the first word in a sentence are never normalized, and are tagged as proper nouns.

  • If the transformation is configured to be case-insensitive, terms like Method and method are recognized as variants of a single term. The list of extracted terms might include either Method or method, depending on which word occurs first in the input data set. If Method is capitalized only because it is the first word in a sentence, it is extracted in normalized form.

Sentence and Word Boundaries

The Term Extraction transformation separates text into sentences using the following characters as sentence boundaries:

  • ASCII line-break characters 0x0d (carriage return) and 0x0a (line feed). To use this character as a sentence boundary, there must be two or more line-break characters in a row.

  • Hyphens (–). To use this character as a sentence boundary, neither the character to the left nor to the right of the hyphen can be a letter.

  • Underscore (_). To use this character as a sentence boundary, neither the character to the left nor to the right of the hyphen can be a letter.

  • All Unicode characters that are less than or equal to 0x19, or greater than or equal to 0x7b.

  • Combinations of numbers, punctuation marks, and alphabetical characters. For example, A23B#99 returns the term A23B.

  • The characters, %, @, &, $, #, *, :, ;, ., , , !, ?, <, >, +, =, ^, ~, |, \, /, (, ), [, ], {, }, “, and ‘.

    Note

    Acronyms that include one or more periods (.) are not separated into multiple sentences.

The Term Extraction transformation then separates the sentence into words using the following word boundaries:

  • Space

  • Tab

  • ASCII 0x0d (carriage return)

  • ASCII 0x0a (line feed)

    Note

    If an apostrophe is in a word that is a contraction, such as we're or it's, the word is broken at the apostrophe; otherwise, the letters following the apostrophe are trimmed. For example, we're is split into we and 're, and bicycle's is trimmed to bicycle.

Configuring the Term Extraction Transformation

You can set properties through SSIS Designer or programmatically.

For more information about the properties that you can set in the Term Extraction Transformation Editor dialog box, click one of the following topics:

For more information about the properties that you can set in the Advanced Editor dialog box or programmatically, click one of the following topics:

For more information about how to set properties, see How to: Set the Properties of a Data Flow Component.

Integration Services icon (small) Stay Up to Date with Integration Services

For the latest downloads, articles, samples, and videos from Microsoft, as well as selected solutions from the community, visit the Integration Services page on MSDN or TechNet:

For automatic notification of these updates, subscribe to the RSS feeds available on the page.