[jira] [Resolved] (SYSTEMML-452) JMLC API: Support for text analytics usecases

["Matthias Boehm (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/SYSTEMML-452?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Matthias Boehm resolved SYSTEMML-452.
-------------------------------------
       Resolution: Fixed
         Assignee: Matthias Boehm
    Fix Version/s: SystemML 0.10

> JMLC API: Support for text analytics usecases
> ---------------------------------------------
>
>                 Key: SYSTEMML-452
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-452
>             Project: SystemML
>          Issue Type: Improvement
>          Components: APIs
>            Reporter: Laura Chiticariu
>            Assignee: Matthias Boehm
>             Fix For: SystemML 0.10
>
>
> I am working on text analytics use case (e.g., document classification, entity extraction).
> I would like to use the JLMC interface at scoring time, but can't find the right method in org.apache.sysml.api.jmlc.PreparedScript.
> For entity extraction, I need features associated with every token in the document. In this case, the features are conceptually represented as a table with 3 columns: 
> - tokenID (Integer) - consecutive integer numbers, representing the position of the token in the document (entity extraction is essentially a problem of classifying every token as Begin_Entity, Inside_Entity, Outside_Entity, and hence the order of tokens in the document is important)
> - featureName (String): name of the feature, for example, whether the token is a capitalized word, or the surface form of the token, etc
> - featureValue (Integer): an integer, in this case always 1 since I do not include features that are absent.
> For document classification, the order of tokens in the document may or may not be important. In the simplest case, assume the order is not important. For each document, we just use the surface form of each token in the document as feature name, and the number of times that surface form appears in that document as feature value. So the features are: conceptually represented as a table with 2 columns: 
> - featureName (String): the surface form of the token
> - featureValue (Integer): the number of times the surface form appears within the document
> Essentially, for both use cases I would like to pass to JMLC a table with a schema, where each column has a known basic datatype (I can think of String, Integer, Float, Boolean). Is this possible ?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)