What does bidirectionality add in a Bi-LSTM text classifier?

What does bidirectionality add in a Bi-LSTM text classifier?medium

Type: conceptual
Topic: used-bi-lstm-in-sarcasm-detection-what-does-bidirectionali
Frequency: common
Tags: deep-learning, you, used, lstm, sarcasm, detection

Answer

A unidirectional LSTM only has past context at each timestep.

Explanation

A unidirectional LSTM only has past context at each timestep. Bi-LSTM runs two LSTMs — one forward, one backward — and concatenates hidden states. For sarcasm, the final word often recontextualizes earlier words ('Oh great, another Monday'). Bidirectionality lets the model use future context to reinterpret earlier tokens — key reason for 7%+ accuracy gain.

Follow-upCan you give a production example?