Overview

The interactive website of our client empowers people to learn from each other’s experience. People ask questions and connect with others who contribute unique insights and quality answers. To encourage good answers, participants’ answers are occasionally featured on blogs as a part of the solution, and the atrocious answers are identified and flush out from the website for which we developed controversial event extraction, AI chatterbots, content recommendation, and sentiment analysis to identify and flag insincere questions. The complexity of the project was to perform on natural language processing and text mining to correspond using semantic objects. We execute hundreds of experiments with deep learning architectures to identify semantic word to handle this complexity.

The Requirement

  • Automatic detection and filtering of inappropriate language to improve the quality of conversations
  • Detection of inappropriate words such as spelling mistakes and variations, contextual, polysemy and semantic variations
  • Capture both local features as well as their inclusive semantics
  • Implementation of a novel deep learning-based technique to automatically identify inappropriate language

The Challenges

  • We were required to design a process to find potential awful messages and train the detector during the experiment period.
  • The problem for the websites was to handle toxic and disruptive content.
  • A key challenge was to weed out deceitful questions
  • To eliminate spaces between words was one of the toughest tasks

The Solution

  • We developed models that identify, detect and mark insincere questions using both machine learning and manual review to address the problems through which models can develop more scalable methods to detect toxic and misleading content
  • We offered all-inclusive training data that includes the questions that were asked identified as insincere
  • Implemented prediction module serverless environment

Automatic Anomaly Detection

acr
: • • • • • • • •

The Development

Training Phase

  • We used a dense layer, recurrent layers, and IO Layer. Training phase was performed on AWS deep learning AMI

Model Deployment

  • AWS Sagemaker hosting service

Achievement

  • By implementing this solution on cloud-based infrastructure. End-user get a quick response and reduce the cost for handling support
  • Useris able to get clear thoughts and visibility for their target product

Technology Stacks

  • Cloud InfrastructureAWS Pagemaker
    AWS Lamda
    Technologies
    Python
    Keras
    Flask Framework
    React
    Postgress SQL
Team
acr

1

React Developer

acr

1

Node Developer

acr

1

Python Developer

acr

1

Cloud Developer

The Result

  • We successfully developed a 100% toxic detection model to assist users and help them to improve online conversations
  • The client is indeed satisfied with our strict turnaround times as we always adhere to the deadline
  • The client had the great pleasure of working with us for building a desirable model using deep learning
  • Delivering the project in the mentioned deadline, the client signed further ongoing contracts with us
Contact us