The TikaPipelet converts various document formats (such as PDF, Microsoft Office, OpenOffice, etc.) to plain text using Tika technology: A record attachment containing the binary content can thus be converted to plain text and stored in an attribute. In addition to that, metadata properties of the document (like title, author, etc) can be extracted and written to record attibutes. To improve the Tika parsing process, it is possible to optionally pass the content-type and filename of the document stored in other record attributes via parameters contentTypeAttribute and fileNameAttribute.
The TikaPipelet supports the configurable error handling as described in SMILA/Development_Guidelines/How_to_write_a_Pipelet#Implementation. When used in JobManager workflows, records causing errors are dropped.
Supported document types
By default, SMILA contains only a subset of Tika. Therefore not all documents formats can be converted out-of-the-box by using the TikaPipelet. However it's easy to extend SMILA so that the TikaPipelet supports all document formats, see "Extending Tika" section below.
|Document format||supported out-of-the-box||supported by using||Hints|
|OpenOffice (OpenDocument formats)||yes||TikaPipelet||---|
|Plain text||yes||---||no conversion, given input text is used as "converted" text|
|HTML/XML||yes||HtmlToTextPipelet||BoilerpipePipelet can also be used for HTML text extraction|
|no||Tika extension||converted text will be empty with out-of-the-box SMILA, a warning will be written to the log|
As you see, SMILA (resp. its 'AddPipeline' which is the default indexing pipeline) per default uses the TikaPipelet only for converting binary document formats. When indexing text based documents another piplelet (HtmlToTextPipelet) is used. However after extending Tika this can be simplified by using TikaPipelet for all document formats.
|inputType||String : ATTACHMENT, ATTRIBUTE||runtime||yes||Selects if the input is found in an attachment or attribute of the record. Usually it doesn't make sense to use "ATTRIBUTE" here because the documents to convert are binary content.|
|outputType||String : ATTACHMENT, ATTRIBUTE||runtime||yes||Selects if output should be stored in an attachment or attribute of the record|
|inputName||String||runtime||yes||Name of input attachment or path to input attribute (process a String literal of attribute)|
|outputName||String||runtime||yes||Name of output attachment or path to output attribute for plain text (store result as String literal of attribute)|
|extractProperties||String||runtime||no||Specifies which metadata properties reported by Tika for the document should be written to which record attribute. See below for details.|
|contentTypeAttribute||String||runtime||no||Parameter referencing the attribute that contains the content-type of the document. If specified the content-type is used to better guide the Tika parsing process. Tika also performs a MimeType detection and the resulting value is stored in this attribute.|
|fileNameAttribute||String||runtime||no||Parameter referencing the attribute that contains the name of the file that was the source of the attachment content. If specified the filename is used to better guide the Tika parsing process.|
|exportAsHtml||Boolean||runtime||no||Flag that specifies if the output should be in HTML format (true) or not (false). Plain text output (false) is default.|
|pageBreak||Boolean||runtime||no||Flag that specifies if pageBreaks should be used to split the content into multiple output records (true) or not (false). The recordId of the output records is generated by concatenating the recordId of the input record with the pageNumber, seperated by #, e.g. (testdoc.pdf#1).This parameter is only interpreted if exportAsHtml is false. Default is (false).|
|pageNumberAttribute||String||runtime||no||Parameter that specifies the name of the attribute that should contain the extracted page number. This parameter is only interpreted if pageBreak is true. If not set, the page number is not set (default)|
|partsAttributeName||String||runtime||no||Setting that enables the output of multi-segment entities. This switches the behaviour of the pipelet to emit page-broken input objects not as multiple output records but rather as one ouput record with multiple parts. The parts are represented as a sequence of maps in the output record. The value of the partsAttributeName settings serves as key for this sequence. Default is (not set).|
|keepHyphens||Boolean||runtime||no||If set to "false", hyphens are removed from words at line breaks so that the separated syllables are contracted to one word (“charac-<newline>teristics” gets "characteristics"). If set to "true", this dehyphenation is disabled. Default is (false).|
|maxLength||Long||runtime||no||The maximum number of characters to extract. If a document contains more characters than specified all remaining characters are omitted. To get all available characters just omit this Parameter. This may lead to OutOfMemory Exceptions with big documents. Default is -1 (unlimited).|
Some notes on "maxLength" in combination with other parameters:
- If "exportAsHTML" is set to "true", the HTML tags will not be counted when checking the limit, so the actual output will be longer than maxLength characters: The output creation stops when the "real" text content of the HTML reaches maxLength characters. After this, also no additional tags will be appended.
- The extracted text is "trimmed" so the actual output can be shorter than maxLength characters cause leading and trailing whitespaces are removed.
- When "outputHyphen" and "exportAsHTML" are set to "false", the actual output can be shorter than maxLength characters, because the hyphens and linebreaks are removed from the limited output. With "exportAsHTML=true", this effect will probably not notable because usually the output will get longer because of the HTML tags.
Configuring the Property Mapping
In addition to the plain text content, Tika can extract metadata properties from documents like title, author, publisher, dates of publication etc, ... The names of these properties depend very much on the documents and what is actually extracted. Some well known names like Dublin Core (dc, dcterms) are used. For a complete list please refer to the Tika documentation. To check with your documents you can download Tika and use the Tika Application to see all extracted metadata.
To store such metadata properties in SMILA records, you must specify the names of the properties you want to store in the extractProperties parameter. Usually this parameter contains a sequence of maps. The map values have the following format:
|metadataName||String||runtime||yes||The name of the metadata property. This will be matched with the extracted metadata property names in a case-insensitive manner.|
|targetAttribute||String||runtime||no||The name of Record attribute to store metadata value(s) in. If not set the string provided in the metadataName will be used as attribute name.|
|singleResult||Boolean||runtime||no||Flag that specifies if only the first value (if multiple values exists) is used in the result (true) or if all values are used (false). Default is false.|
|storeMode||String||runtime||no||Specifies whether attributes already stored in the record target attribute will be left unchanged ("leave"), overwritten ("overwrite") or if the extracted properties will be added to potentially existing ones ("add"). Default is "add".|
The following example shows how to configure the pipelet to extract the text from the attachment called Content and stores the extracted text in the attribute Text. Additionally the eventually contained metadata properties Company, Creator and Title will be stored in properties.
E.g. if a word document with the value "ACME" as company and "John Doe" as creator, the resulting record would contain the plain text in the attribute Text, the value ACME in the attribute Company, as well as the value John Doe in an attribute Creator.
<proc:configuration> <rec:Val key="inputName">Content</rec:Val> <rec:Val key="inputType">ATTACHMENT</rec:Val> <rec:Val key="outputName">Text</rec:Val> <rec:Val key="outputType">ATTRIBUTE</rec:Val> <rec:Val key="contentTypeAttribute">MimeType</rec:Val> <rec:Val key="fileNameAttribute">FileName</rec:Val> <rec:Val key="exportAsHtml">false</rec:Val> <rec:Val key="pageBreak">falsec</rec:Val> <rec:Val key="keepHyphens">false</rec:Val> <rec:Val key="maxLength">100000</rec:Val> <rec:Seq key="extractProperties"> <rec:Map> <rec:Val key="metadataName">company</rec:Val> <rec:Val key="targetAttribute">Company</rec:Val> <rec:Val key="singleResult">false</rec:Val> </rec:Map> <rec:Map> <rec:Val key="metadataName">creator</rec:Val> <rec:Val key="targetAttribute">Creator</rec:Val> <rec:Val key="singleResult">false</rec:Val> </rec:Map> <rec:Map> <rec:Val key="metadataName">title</rec:Val> <rec:Val key="targetAttribute">Title</rec:Val> <rec:Val key="singleResult">true</rec:Val> </rec:Map> </rec:Seq> </proc:configuration>
- MS- Office
SMILA does not contain the complete Tika distribution, because some converters need third party libraries with licenses that we are not allowed to distribute. However, it is easy (and absolutely legal!) to include those parts of Tika into your SMILA installation yourself:
- Download org.eclipse.smila.tika.deps bundle from here
- Replace the appropriate bundle of your SMILA distribution with the downloaded bundle by just copying the downloaded bundle to <path-to-your-SMILA>/plugins folder.
That's it! After SMILA restart, all document formats supported by Tika will be also be supported by SMILA's TikaPipelet.