LinkedIn Releases “DeText”, an Open-Source Framework for NLP Tasks

Jul 30, 2020 | AS Team

LinkedIn Releases “DeText”, an Open-Source Framework for NLP Tasks title banner

LinkedIn launched "DeText", an open-source framework for various natural language processing tasks like estimation, classification, and language generation tasks. With the usage of deep neural networks, it leverages the semantic matching in order to recognize member purposes in the searching and recommendation system.


According to the statement over LinkedIn, it is able to be implemented to a wide range of tasks incorporating search and recommendation calculation, multi-dimension classification, and inquiry and conclusion. (Must read: What is LinkedIn Analytics? Explanation with LinkedIn Analytics Tools)


Senior engineer manager Weiwei Guo at LinkedIn said,“ DeText was devised with ample flexibility that fulfills the requirements of various rendering services. It’s obliged with “state-of-the-art” algorithms consolidated in an end-to-end model where the variables are concurrently modernized, but it endeavors for balancing its entire potency with high efficiency.”


Guo’s continued, “DeText can be considered as the cordless drill that enables users to interchange and optimizes NLP models on the basis of used cases.” 


Deep Learning-based neural language processing embraces the potential to expand how searching and recommendation system understand human intension, hence the leverages the power in industrial applications. 


Specification of DeText 


  1. It consists of numerous components that are customized through preload templates

  2. An embedding layer to transform a sequence of words into a matrix.

  3. Models for encoding,

  4. An interaction layer to produce specialties on the basis of text embeddings,

  5. Features processing that links conventional features with interactive features, and

  6. An MLP layer that connects broad and extensive features.


Tags #Artificial intelligence