Dataset and replication package for the Automating App Review Response Generation (submitting to ASE 2019).
Run the code with
$ python model.py
You can change the configures in parameter.py, including the hidden_size, word_vec_size, num_epochs, etc. The important parameters are
use_sent_rate -- whether include sentence rating or not
use_sent_senti -- whether include sentence sentiment or not
use_sent_len -- define the sliced review length, i.e. the categorization interval, e.g., 20, else "False"
use_app_cate -- whether include app category or not
use_keyword -- whether include keyword information of one review or not
tie_ext_feature -- whether combine external features or not. "False" means that all the external features are not involved.
Some examples of generated reponses can be found in this link.
As the dataset is very large and also such data can only be used for academic purpose, you need to fill a requested form first before downloading the data.