MultiKE-DMM is a topic model for short text leveraging multiple knowledge embeddings.
For more details, please refer to this paper.
If you use this package, please cite the paper: Jueying He, Jiayao Chen, and Mark Junjie Li. Multi-knowledge Embeddings Enhanced Topic Modeling for Short Texts. Neural Information Processing: 29th International Conference, ICONIP 2022, Part III page 521–532.
To quickly reproduce the results in the paper, follow these steps:
1.Download the data folder from (https://drive.google.com/drive/folders/1beiba3-9MusJnvqzGuklng1TApTmb7MR?usp=sharing) and replace ...\MultiKEDMM\data with this folder
2. Decompress all zip files in .../MultiKEDMM/data/ , including .../MultiKEDMM/data/shortTextCorpus/xxx/word_wiki.zip and .../MultiKEDMM/data/corpusKnowledgeEmbedding/MultiKESim.zip
3.Run .../MultiKEDMM/src/MultiKEDMM/resultMultiKEDMM.java
4.Evaluate the perfomance of MultiKE-DMM in .../MultiKEDMM/result/result.txt
To reproduce the results in a step-by-step manner, follow these steps:
Part 1 is to preprocess two kinds of external knowledge and four Wikipedia index files for four short text datasets.
1: Download the 4,776,093 wikipedia article from (http://pan.baidu.com/s/1slaTPoT) or (https://drive.google.com/open?id=1JTLu-AqNhf7xUWTgKKm8360wT0OQve7O)
2.Decompress the articles to _.../MultiKEDMM/data/wiki_full/ _
3.Run .../MultiKEDMM/src/Util/WordWikiIndex.java for each dataset to get the four wikipedia index files
4.Download the pre-trained entity embedding on WordNet trained by TransE from (https://drive.google.com/file/d/1DSK6M2U7GESYLpUqnp0BTJE1sSx6plMv/view?usp=drive_link)
5.Run .../MultiKEDMM/precessKnowledge/word2EntityEmbedding.py to get the entity embedding related to words according to Lesk algorithm
6.Download the pre-trained word embedding tools including Word2Vec(GoogleNews-vectors-negative300.bin), GloVe(gensim_glove.840B.300d.txt) and FastText(crawl-300d-2M.vec)
7.Run .../MultiKEDMM/precessKnowledge/word2Vector_xxx.py to get word vectors
8.Run .../MultiKEDMM/precessKnowledge/word2MultiKnowledgeEmbedding.py to construct the multi-knowledge embedding of words
9.Run .../MultiKEDMM/precessKnowledge/wordSimilarity.py to get word cosine similarity based on their word embeddings
Part 2 is to obtain the pdz, pzw, pd, and TopWords file for evaluating the performance of MultiKE-DMM.
1.Run .../MultiKEDMM/src/MultiKEDMM/resultMultiKEDMM.java after setting the inputs of MultiKE-DMM, including corpus, the similarity file of this corpus computed by wordSimilarity.py and a number of parameters
2.Get the evaluation of MultiKE-DMM in .../MultiKEDMM/result/result.txt
@inproceedings{DBLP:conf/iconip/HeCL22,
author = {Jueying He and
Jiayao Chen and
Mark Junjie Li},
title = {Multi-knowledge Embeddings Enhanced Topic Modeling for Short Texts},
booktitle = {Neural Information Processing - 29th International Conference, {ICONIP}
2022, Virtual Event, November 22-26, 2022, Proceedings, Part {III}},
series = {Lecture Notes in Computer Science},
volume = {13625},
pages = {521--532},
publisher = {Springer},
year = {2022},
url = {https://doi.org/10.1007/978-3-031-30111-7\_44},
doi = {10.1007/978-3-031-30111-7_44},
timestamp = {Thu, 20 Apr 2023 15:54:56 +0200},
biburl = {https://dblp.org/rec/conf/iconip/HeCL22.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}