I'm trying to build a suitable "code-free" Topic Model in "Orange Data Mining" software for the experimental part of my Thesis, in order to analyze the themes and trends of research articles' corpus dataset. The model then can be used by normal researchers who are not necessarily have programming & coding skills in order to apply their own Topic Modeling workflows.
The pre-dataset (corpus) were successfully collected as (.CSV/.TAB) format. The right selection of the optimal model, number of topics, and the visual results should be simple, applicable, and reliable for best understanding the relationship & coherence between the discovered topics.
In-terms-of text preprocessing steps, Orange provides a nice widget that contains all-steps-in-one interface. The issue with Orange's Topic Modeling approach is that I couldn't find a clear method (or widget) for measuring either the optimal-number of topics or the topic-coherence score in order to evaluate the topics extracted from the Latent Dirichlet Allocation "LDA" algorithm in Orange's Topic-Modeling widget.
ِAs in many literature-reviews, the usual solution to this issue is usually through through programming, which will be included in the code-script section of my Thesis, using Jupiter-Python code similar to the method described in the following article: LDA Topic Modeling using Python
The code-cells are assembled into a single file attached with the following link: Full Python Code
Here I got 2 questions:
Is it possible to use the same steps described in the previous article's coding-cells of Python, that are related with selecting the optimal-number of topics & topic-coherence score calculations into Orange, so that it can be transformed into one or mode connected widgets in Orange; leading to much easier use by normal researchers?.
What is the necessary widgets' sequence that better visualizing the extracted topics in order to better analyze the research articles' themes & trends?.
Thanks for your collaboration in advance..