boolean-query-retrieval

In this project we will be building boolean query retrieval model using 20_newsgroups dataset

Requirements

• python 3
• nltk
• pickle

Pre-processing

• Removal of meta data in each file
• Removal of symbols like [!@#$%^&*()] from the tokens
• Removal of non alpha numeric tokens
• Lemmatization of tokens
• Removal of words that are less than size 2 (useless words)

Methodology

• Download the 20_newsgroups dataset which is freely available and place in data folder
• For each file pre process the file and construct the index dictionary
• index dictionary will be: { token : list of files containing the token }
• Following precedence is assumed : NOT > AND > OR
• Sample User queries :
  • email AND version
  • email OR version
  • NOT email
  • email AND NOT version
  • email OR NOT version
  • email AND version OR print
  • email AND version AND ----OR --NOT--
• Retrieves all the files that satisfies the above conditions

Optimisations:

• When there are NOT followed by NOT, two NOT operations are cancelled each other without calculating again and again
  • NOT NOT boy = boy
• if query contains only AND operations then, operations are performed in the orderof tokens having minimum postings list.

References :

• Followed chris manning text book for making this model. link