Latent Keyphrase Inference (LAKI)
- Jialu Liu, Xiang Ren, Jingbo Shang, Taylor Cassidy, Clare Voss and Jiawei Han, "Representing Documents via Latent Keyphrase Inference”, Proc. of the 25th Int. Conf. on World Wide Web (WWW'16), Montreal, Canada, April 2016.
The current implementation requires SegPhrase to extract domain keyphrases. It has been added under this repository as a submodule.
We will take Ubuntu for example.
- g++ 4.8
$ sudo apt-get install g++-4.8
- python 2.7
$ sudo apt-get install python
$ sudo apt-get install pip $ sudo pip install sklearn
$ sudo pip install nltk
LAKI can be easily built by Makefile in the terminal.
$ ./train_dblp.sh #train a LAKI model using DBLP dataset. $ ./test/test_inference #receives a string query and returns top ranked document keyphrases
All the parameters are located in train_dblp.sh
INPUT refers to the input file of LAKI, can be downloaded from AMiner. For other datasets, please refer to the format of file indicated by RAW_TEXT (each single line indicates a document) and comment out line 25-28.
Number of threads.
Number of domain keyphrases extracted by SegPhrase
Number of occurrences for a valid domain keyphrase in the corpus.
####For other parameters regarding each individual module, please check the corresponding cpp files.