A Novel Model for Enhancing Fact-Checking

Warrant Generation through Deep Learning

“…given the claim, evidence and candidate warrant as input, the policy aimed to decide the warrant relevant words as delete, keep or generate. Afterword embeddings 𝑒 𝑖 is performed, we use Bi-GRU to get the vector representation of candidate warrant ℎ 𝑠 (1) + ℎ 𝑠 (1) + ℎ 𝑠 (2) + ⋯ + ℎ 𝑠 (𝑛) .…”

“…Following the acquisition of claim and evidence hidden state representations, we then pool the vectors on an average basis 𝑐𝑙𝑎𝑖𝑚 (𝑙) and 𝑒𝑣𝑖𝑑𝑒𝑛𝑐𝑒 (𝑙) through equations 9-13: ( 9 ) ℎ ⃗ 𝑖 (1) , ℎ ⃖⃗ 𝑖 (1) = 𝑏𝐺𝑅𝑈 (𝑒 𝑖 , ℎ ⃗ 𝑖−1 (1) , ℎ ⃖⃗ 𝑖+1 (1) ) ( 10 ) ℎ 𝑖 (1) = 𝑊 1 [ℎ ⃗ 𝑖 (1) , ℎ ⃖⃗ 𝑖 (1) ]…”

“…The Toulmin model components are necessary for fact-checking as Alkhawaldeh et al demonstrated in [1]. Argument mining is automatic recognition and extraction of the structure of inference and reasoning expressed in natural language arguments [2].…”

“…Despite the fact that utilising a warrant can aid in the performance of fact-checking tasks [1], to our knowledge, no previous work has proposed that a claim be connected to a piece of evidence via automated warrant creation rather than manual annotation. Additionally, no experiment was conducted using a labelled dataset, but rather through the use of case studies [6].…”