Supervised-machine-learning without coding
Submitted by Rajesh Gudikoti (@ragudiko) on Wednesday, 7 June 2017
Section: Full talk Technical level: Beginner
We can build the machine learning model which can understand the linguistic nuances and relationships specific to a industry. Once model is trained and evaluated, you can use it to extract domain specific entities from new documents.
Current natural language processing techniques cannot extract/interpret the data as required by domain/industry specific. The data(entities) represent different meaning in different domain. To overcome such business issue we have tool by which you can Seamlessly create and deploy industry specific models for building cognitive apps.
In this tool we build type system specific to industry. The type system consists of entities and relationship between entities. For e.g.
In case of Employee and Employeer, employee and employer are entities and employedBy can represnted as relationship between two entities.
SMEs of the industry have better understanding of business, so their inputs will be required to build type system.
Human Annotators guide the system to understand the semantics of the industry by annotating(mapping) the text/phrase with specific entities.
Based on human annotation the machine learning model is trained and evaluated.
The tool provides option to anlayze the performance of the model. If you are satisfied with model, you can go ahead and use to anlayze new sms text. The performance of the system are represented by scores which helps us understand how many and which entities were correctly identified.
Basic understanding of Natural Language Processing.
Rajesh is working as Architect at IBM, India Digital Business Group. I have been part of product development in various domains.