fatfishZhao / cross_nlp

cross_nlp

paper:A Cognition Based Attention Model for Sentiment Analysis

requirement

python 3
nltk
textstat
pytorch
spacy

dataframe data can be accessed in 链接：https://pan.baidu.com/s/1oCXJEjCzOc1HuOn8ONzdhA 提取码：3ecs 复制这段内容后打开百度网盘手机App，操作更方便哦
original Mishra data address
http://www.cfilt.iitb.ac.in/~cognitive-nlp/resources/Eye-tracking_and_SA-II_released_dataset.zip

experiments

experiments repair:
Download data frame data from Baidu Cloud, creat a "data" directory in root and put the data there.
Or If You Want to Generate Features Dataframe Files yourself:

1. Download raw data and run utils/convert_data_df.py and utils/convert_Mishra_data.py
2. Run cal_feature_thread.py and cal_feature_thread_Mishra.py
   PS: My code to calculate features are too slow and need about 4 hours on my server. I would be glad if some one can speed up my code.

regression1_3:
cross validation for all 14 people data
regression1_4:
cross validation for each person
regression1_5:
repeat 1_3, but use the percentage of reading time for each person for regression target
regression1_6:
repeat 1_3, but nomalize the features first regression1_7:
use RNN to regress regression1_8:
Split data beased on different sentences.
regression2_0:
cross validation for Mishra dataset.