Targeted Sentiment Analysis

This repository is the code for paper Implicit Syntactic Features for Targeted Sentiment Analysis, it contains the modified code used to train the targeted sentiment analysis. We mainly employ the tensorflow LSTM framework in the paper deep biaffine attention for neural dependency parsing. [CODE] All our modified codes mainly inlude /lib/models/ dirs, network.py, dataset.py, bucket.py and vocab.py parts of the original CODE

1. Pre-trainning Submodels

---

We pretrain the submodels on PTB data:

1.1 For POS Tagging model:

See the Readme.md in ./SUBMODELS/POS/ dirs

1.2 For Dependency Parser model:

See the Readme.md in ./SUBMODELS/Dependency Parsing/ dirs

2. Sentiment Analysis

---

Syntactic models:

2.1 Save pre-trained models:

We saves the best result models for both POS tagging model and Dependency Parsing model( include normal dependency model and non-postags dependency model) as the pre-trained models.

2.2 Re-Load the pre-trained models and train sentiment analysis model:

See the Readme.md in ./Sentiment/ dirs