shivam-bahedia / snlpminiproject

SNLP mini project 2019-20

SNLPminiproject - Team Blitz

Diagram

Module

FactChecker : 

• Tokenization of the input facts.
• Break the generated tokens in the form of Subject and Object by identifying the Predicate.
• Assigning true/false value to the fact by searching the predicate on wiki pages of the generated token.

LocalWiki :

• It fetches the result wiki pages of generated tokens.
• It also stores the previously fetched data in the formal on text files in Local Store folder.

Input: 
List of facts in natural language with their corresponding fact ids (train.tsv and test.tsv)

Output: 
Generates a triple file that maps the facts to its truth value (trainresult.ttl and testresult.ttl)

WorkFlow

• Tokenization - Generate tokens for the fact, based on words starting with a capital letter including the prepositions following a noun.
• Data fetch from Wiki - For each generated token check if the data exists in Local Store, if not fetch data from Wikipedia and store it in the Local Store.
• Get predicate for the fact - For each fact generate a generic predicate based on the predefined list of predicates.
• Search Data - Search the data of each token for pattern containing the predicate and the other token.
• Assign Truth Value - If the pattern matches, assign true to the truth value.

Positive Example

Fact: 3820514 Alfonso XIII of Spain's birth place is Madrid.

1. Tokenization:
   {"Alfonso XIII of Spain" ,"Madrid "}
2. Data fetched from Wiki:
   token1: {"Alfonso XIII of Spain"} 
   token2: {"Madrid"}
3. Get predicate for the fact:
   (B|b)orn.{0,150}(?i)
4. Search Data:
   String "Born (1886-05-17)17 May 1886 Royal Palace of Madrid" found in the wiki page of token1 ("Alfonso XIII of Spain") 
5. return truthValue =true for this fact.

Negative Example

Fact: 3885766 Lucille Ball's death place is Santa Monica, California.

1. Tokenization:
   {"Lucille Ball" ,"Santa Monica, California "}
2. Data fetched from Wiki:
   token1: {"Lucille Ball"} 
   token2: {"Santa Monica, California "}
3. Get predicate for the fact:
   (Died).{0,150}(?i)
4. Search Data:
   token2 not found in the wiki page of token1 with predicate
   token1 not found in the wiki page of token2 with predicate
5. return truthValue =false for this fact.

Execution Steps

Directly execute the Main Class.

Team Member 

Shivam Bahedia, Sourabh Poddar, Yamini Punetha