Flsamodel

[ERijck]

I have added flsamodel, which includes the topic modeling algorithms FLSA, FLSA-W and FLSA-E. In experimental results, FLSA-W has outperformed other state of the art algorithms on various open datasets (e.g. LDA, LSI, NMF, ProdLDA).

Motivation:
Since Gensim features various state-of-the-art topic modeling algorithms, and my group's algorithms outperform these algorithms in terms of coherence-, diversity- and interpretability score, we believe our algorithms should be featured in Gensim too. Previously, I created a wrapper function that depended on FuzzyTM. In this PR, the algorithms are trained within Gensim.

People can use this code similarly to how LDAmodel is being used. For the 'corpus' the following datatypes are allowed:

• list of list of str.
• list of list of tuples (int, int) (bow).

The algorithms have been featured in various scientific publications. See the links below:

FLSA-W:
Rijcken, E., Scheepers, F., Mosteiro, P., Zervanou, K., Spruit, M., & Kaymak, U. (2021, December). A comparative study of fuzzy topic models and lda in terms of interpretability. In 2021 IEEE Symposium Series on Computational Intelligence (SSCI) (pp. 1-8). IEEE.

FLSA-E:
Rijcken, E., Zervanou, K., Spruit, M., Mosteiro, P., Scheepers, F., & Kaymak, U. (2022). Exploring Embedding Spaces for more Coherent Topic Modeling in Electronic Health Records. In IEEE International Conference on Systems, Man, and Cybernetics.

FLSA:
Karami, Amir, et al. "Fuzzy approach topic discovery in health and medical corpora." International Journal of Fuzzy Systems 20.4 (2018): 1334-1345.

These algorithms are featured in the FuzzyTM package:
Rijcken, E., Mosteiro, P., Zervanou, K., Spruit, M., Scheepers, F., & Kaymak, U. (2022, July). FuzzyTM: a software package for fuzzy topic modeling. In 2022 IEEE international conference on fuzzy systems (FUZZ-IEEE) (pp. 1-8). IEEE.

Experimental results:

• Topic quality (open datasets):
  https://research.tue.nl/en/publications/a-performance-evaluation-of-topic-models-based-on-fuzzy-latent-se

• Predictive power vs. topic quality (private dataset):
  Rijcken, E., Kaymak, U., Scheepers, F., Mosteiro, P., Zervanou, K., & Spruit, M. (2022). Topic Modeling for Interpretable Text Classification From EHRs. Frontiers in Big Data, 5.

[mpenkov]

Thank you for contribution @ERijck !