This is a github mirror of the link found from Illimine here.

Data is provided in this github repository, only because it does not consume too much space.

**** Illimine Copyright ****

University of Illinois at Urbana-Champaign, 2014

illimine.cs.illinois.edu

**** Additional Copyright ****

This package contains the source code and the dataset used in the following paper:

@inproceedings{conf/sdm/KERT14, author = {Marina Danilevsky and Chi Wang and Nihit Desai and Xiang Ren and Jingyi Guo and Jiawei Han}, title = {Automatic Construction and Ranking of Topical Keyphrases on Collections of Short Documents}, booktitle = {SDM}, year = {2014}, ee = {http://dx.doi.org/10.1137/1.9781611973440.46}, }

If you use any contents in this package, please cite the above paper as your reference.

**** Code explanation ****

(1). The input file is %1.txt under the folder %data. Every line is a string representing one document.

(2). Set up the parameters in flat_kert.bat

data: the input folder;

output: the output folder;

code: the code folder;

ntopic: number of topics;

minsup: the minimum frequency for a phrase;

gamma: the parameter for completness condition, default 0.5;

options.wp: the weight for phraseness;

options.maxn: the longest phrase length;

options.top: the number of phrases output for each topic;

(3). Execute

flat_kert.bat

This version of code uses LDA as the topic model. If you want to try BLDA, please use blda.py instead of mallet LDA. See readme_blda.txt in the code folder.

(4). Output:

%1.kert - %ntopic% lines, each containing top %options.top% phrases for one topic,
separated by \t.

**** Data explanation ****

This package contains a subset of DBLP paper titles in 20 conferences.

word_index.txt - the dictionary of terms, indexed from 1 pt.txt - the terms in each paper title. Every line has two columns: paper id and term id. Both are indexed from 1

**** For More Questions ****

Please contact illimine.cs.illinois.edu or Chi Wang ([email protected])