Combining Similarity Features and Deep Representation Learning for Stance Detection in the Context of Checking Fake News

Benchmark Model Rank Results
fake-news-detection-on-fnc-1Bi-LSTM (max-pooling, attention)#3Weighted Accuracy: 82.23Per-class Accuracy (Unrelated): 96.74
natural-language-inference-on-multinliStacked Bi-LSTMs (shortcut connections, max-pooling)#40Matched: 71.4Mismatched: 72.2
natural-language-inference-on-multinliBi-LSTM sentence encoder (max-pooling)#41Matched: 70.7Mismatched: 71.1
natural-language-inference-on-multinliStacked Bi-LSTMs (shortcut connections, max-pooling, attention)#42Matched: 70.7Mismatched: 70.5
natural-language-inference-on-snliStacked Bi-LSTMs (shortcut connections, max-pooling)#52% Test Accuracy: 84.8
natural-language-inference-on-snliBi-LSTM sentence encoder (max-pooling)#55% Test Accuracy: 84.5
natural-language-inference-on-snliStacked Bi-LSTMs (shortcut connections, max-pooling, attention)#56% Test Accuracy: 84.4