sent_len, sent_num

[savvyyy]

Hi @yafangy Great work. Thanks for sharing the code.
I was preparing datasets for laptop. I need to know how did you decide sent_len and sent_num in case of restaurant?
sent_len, sent_num = 83, 2000
and
sent_len, sent_num = 83, 676

While using the same values for laptop datasets, I got an error telling:-
IndexError: index 1900 is out of bounds for axis 0 with size 2000

Please guide me through this.

[yafangy]

I believe it is related to the dataset. The sent_len is the maximum words in each sentence, and sent_num is the number of sentences to be processed. For different dataset, the number of training sentences might be different. Maybe try to use a smaller sent_num if it exceeds the number of sentences in the dataset.

[savvyyy]

Hi @yafangy
Thanks for the help. It worked for my datasets.
I have one more query.
What changes I should make in order to get polarity of the aspects as mentioned in in this image
How did you came across getting the category and polarity.
Please guide.

[savvyyy]

Hi @yafangy Kindly help me with issue above. I am eagerly waiting to know the process.

[savvyyy]

Hello sir @yafangy
I have been waiting long for your reply. Kindly reply or give me any other input for my question. It will be really helpful sir. Please guide.

[yafangy]

This code is for aspect extraction and the result is shown in https://github.com/yafangy/Review_aspect_extraction/blob/master/image3_v2.png.

This polarity you found is in the training dataset but not in the scope of this current code. You can either add another sentiment analysis for the target word or the following. For the current approach I posted, you can try to simply increase the output from 0/1 (normal word vs. aspect word), to a 4-class multi-category (normal word, positive/neutral/negative aspect word) and add the polarity into consideration when you generate labels. Let me know if it works for you.