u_fact = fact_graph_choose @ atten_tensor

[Elaine10-27]

Can you explain the u_fact = fact_graph_choose @ atten_tensor statement? I noticed that this statement appeared in many running files. But there was an error in this line when running. Thanks ~

[prometheusXN]

The u_fact = fact_graph_choose @ atten_tensor is used to choose the attention vector of the corresponding law article group to compute the representation of fact description in re-encoder. In fact, the fact_graph_choose is got by: fact_graph_choose = tf.where(fact_graph_choose == tf.reduce_max(fact_graph_choose, -1), tf.ones_like(fact_graph_choose), tf.zeros_like(fact_graph_choose)) , where only the position of the maximum value is one, the other position is zero. Then, we use the inner product operation to pick the corresponding attention vector in atten_tensor. I hope this answer can help you! xunuo0109@hotmail.com From: elaine Date: 2020-04-26 19:49 To: prometheusXN/LADAN CC: Subscribed Subject: [prometheusXN/LADAN] u_fact = fact_graph_choose @ atten_tensor (#1) Can you explain the u_fact = fact_graph_choose @ atten_tensor statement? I noticed that this song sentence appeared in many running files. But there was an error in this line when running. Thanks ~ — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[Elaine10-27]

Thank you for your patience and prompt response. I have another question about dataset. I processed the data according to tongji.py. But the training data and test data screened out are less than those described in the paper. I don't know which step is the problem. I directly use the new_law.txt and new_accu.txt given in your code as clearlawlist and cleararaculist. Looking forward to your reply ~

[prometheusXN]

Our dataset is from Wenmian Yang who is the author of paper "Legal Judgment Prediction via Multi-Perspective Bi-Feedback Network" (IJCAI 2019). There is actually some difference compared with the preliminary dataset of CAIL. Can you give me an e-mail address? I'll send the CAILsmall dataset to you. xunuo0109@hotmail.com From: elaine Date: 2020-04-28 15:48 To: prometheusXN/LADAN CC: prometheusXN; Comment Subject: Re: [prometheusXN/LADAN] u_fact = fact_graph_choose @ atten_tensor (#1) Thank you for your patience and prompt response. I have another question about dataset. I processed the data according to tongji.py. But the training data and test data screened out are less than those described in the paper. I don't know which step is the problem. I directly use the new_law.txt and new_accu.txt given in your code as clearlawlist and cleararaculist. Looking forward to your reply ~ — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[Elaine10-27]

Thank you so much. This is my mail box elaine1027zhl@hotmail.com. Looking forward to your mail. Thanks ~

[prometheusXN]

Ok, I have sent the dataset to you, please check it~~

[Elaine10-27]

I have received it. # ^_^ #