Abstrakt
Topic modeling is a text mining technique that presents the theme of the corpus by identifying latent features of the language. It thus provides contextual information of the documents in the form of topics and their representative words, thereby reducing time, efforts, etc.
Topic modeling on English corpus is a common task, but topic modeling on regional languages like Marathi is not explored yet. The proposed approach implements a topic model on Marathi corpus containing more than 1200 documents.
Intrinsic evaluation of latent Dirichlet allocation (LDA) which is used to implement the topic model is carried out by coherence measure. Its value is maximum for 4 topics.
The retrieved topics are related to ‘Akbar–Birbal,’ ‘Animal stories,’ ‘Advise giving stories’ and ‘general stories.’ Dendrogram and word cloud are used for visualization. The dendrogram shows topic-wise documents and word cloud show sample informative words from different stories.
The proposed approach involves context while deriving the topics using synsets. Entropy value is 1.5 for varied datasets; entropy value ensures independence of topic and similarity between topics’ words.