Update README.md

README.md

@@ -40,7 +40,7 @@ bash run_emp_gpt3.sh
 
 ## 4. Evaluation
 
-Since our work evaluates the GPT-3's ability to express empathy in various aspects, we have to prepare a set of model checkpoints used in the machine-based automatic evaluation metric, such as Epitome, EmoAcc, IntentAcc, and diff-Epitome. We provide all model checkpoints used in our work through google drive. Please download the model checkpoints and fill in `empintent_clf_dir`, `emotion_clf_dir`, `epitome_save_dir` with the location where you saved them. Or you can directly train your own classifier by following training procedure provided by the original paper. (For IntentAcc and EmoAcc, refer this [code](https://github.com/anuradha1992/EmpatheticIntents). For Epitome, refer two codes, [code1](https://github.com/behavioral-data/Empathy-Mental-Health) and [code2](https://github.com/skywalker023/focused-empathy).)
+Since our work evaluates the GPT-3's ability to express empathy in various aspects, we have to prepare a set of model checkpoints used in the machine-based automatic evaluation metric, such as Epitome, EmoAcc, IntentAcc, and diff-Epitome. We provide all model checkpoints used in our work through [google drive](https://drive.google.com/drive/folders/1PXqmv-MZ1uphHvV81htuAhid2uKGGeGd?usp=share_link). Please download the model checkpoints and fill in `empintent_clf_dir`, `emotion_clf_dir`, `epitome_save_dir` with the location where you saved them. Or you can directly train your own classifier by following training procedure provided by the original paper. (For IntentAcc and EmoAcc, refer this [code](https://github.com/anuradha1992/EmpatheticIntents). For Epitome, refer two codes, [code1](https://github.com/behavioral-data/Empathy-Mental-Health) and [code2](https://github.com/skywalker023/focused-empathy).)
 
 `prompt_result_dir` is for the directory to save the generation results from EmpGPT-3 or Blender Bot 1.0 (using ParlAI framework).
 `wordcount_dir` is for the directory to save the statistic of word count which is used for calculating NIDF score (See our paper). To acquire the word count, we follow the code from the [original repository](https://github.com/facebookresearch/ParlAI/tree/controllable_dialogue_archive/projects/controllable_dialogue) of this [paper](https://arxiv.org/abs/1902.08654).