Latent Dirichlet Allocation (LDA) and its variant topic models have been widely used for performing text mining tasks. Topic models sometimes produce incoherent topics having noisy words with high probabilities. The reason is that topic model suffers from binary weighting of terms, sparsity and lack of semantic information. In this work, a fuzzy document representation is used within the framework of topic modeling that resolves these problems. Fuzzy document representation uses the concept of Fuzzy Bag of Word (FBoW) that maps each document to a fixed length fuzzy vector of basis terms. Each basis term in the fuzzy vector belongs to all documents in the dataset with some membership degree. Latent Dirichlet Allocation is tailored to use fuzzy document representation and results are compared with regular LDA and term weighted LDA over short and long text documents on document clustering and topic quality tasks. LDA with fuzzy representation generates more coherent topics than other two methods. It also outperforms the other methods on document clustering for short documents and produces comparable results with term weighted LDA for long documents.
CITATION STYLE
Akhtar, N., Sufyan Beg, M. M., & Javed, H. (2019). Topic Modelling with Fuzzy Document Representation. In Communications in Computer and Information Science (Vol. 1046, pp. 577–587). Springer Verlag. https://doi.org/10.1007/978-981-13-9942-8_54
Mendeley helps you to discover research relevant for your work.