Review Summary System

Project done as part of course work.

Transcript

1. Team – 22
   M Manoj Kumar – Srinath Ravichandran - Dharmesh kakadia – Sandhya S
   (201107502) - (201107625) - (201107616) - (201107617)
   REVIEW SUMMARY SYSTEM
   

   View Slide

2. OVERVIEW
   •  System to summarize reviews from various sources
   •  Users can view and compare products based on features
   •  Results exposed as RESTful web-service
   •  Ability to cater to different products
   

   View Slide

3. OVERALL WORK FLOW
   Feature
   Extraction
   Sentiment Analysis
   Sentiment
   Classification
   

   View Slide

4. DETAILED FLOW CHART
   Reviews Parse and Tag Feature Extraction
   Feature
   DB
   Opinion
   DB
   •  Once for a category
   of product
   •  Nouns #frequency
   •  Adjectives
   #frequency
   •  Classifier is
   designed based on
   this data.
   

   View Slide

5. Review •  Raw Review
   Sentence
   Pruning •  Preprocess data
   •  List of valid features
   Dependency
   relations
   •  Using Stanford
   Parser
   Opinions>
   Semantic
   Analyzer
   
   NoSQL
   (mongo)
   Feature
   DB
   •  Each sentence is
   passed through NLP
   logic.
   •  Features are
   extracted and rated
   according to the
   opinion of the
   setence.
   

   View Slide

6. PARALLELIZING WITH HADOOP
   Mapper
   Reducer
   Reducer
   Reducer
   mobile2
   Summary Data Base
   mobille1 mobille2 mobille3
   (Tag the Review)
   

   View Slide

7. DATABASE SCHEMA
   Trained Data
   •  Nouns #
   •  Modifiers #
   Tagged Reviews
   •  Features
   •  Ratings
   •  Review Text
   Review
   Summary
   •  Features
   •  Average
   Rating
   Product X
   

   View Slide

8. RESTFUL WEB SERVICES
   •  System exposes results as restful web services.
   Review
   System
   

   View Slide

9. EXPERIENCES & LEARNING
   •  NLP Dependency Relationships!!!
   •  REST is BEST
   •  SCHEMA defines EVERYTHING!!
   

   View Slide

10. FUTURE WORK
    •  Better feature Extraction.
    •  Synonym match can be extended with Wordnet::Similarity.
    •  Can be further optimized for blazing performance.
    •  Preprocess user query.
    

    View Slide

11. TOOLS USED
    •  NLP
    •  Stanford Parser
    •  Wordnet (Synonyms)
    •  Sentiwordnet
    •  Hadoop 20.2
    •  Mongo DB
    

    View Slide