Skip to content

Latest commit

 

History

History
33 lines (18 loc) · 1.01 KB

readme.md

File metadata and controls

33 lines (18 loc) · 1.01 KB
Raw

Bi-Model-Intent-And-Slot

Pytorch implementation of paper "A Bi-model based RNN Semantic Frame Parsing Model for Intent Detection and Slot Filling".

We only build the model with decoder.

We do not tune the hyper-parameters carefully as it is so boring. Obtaining best result of intent accuracy is 0.9843 and f1 score of slot filling is 0.9563 when model runs a lot of epoch(need some lucky), but still lower than the claimed result of that paper(0.9899, 0.9689).

--- Intent acc Slot filling F1
paper 0.9899 0.9689
reproduce 0.9843 0.9600

Setup

Pytorch>=0.4.0, python3.

python train.py

Reference

  1. Dataset and codes calculator F1 score from here.

  2. Tricks of dealing with atis dataset from here.