How can I add the titles of newspaper articles to my topic modeling in a way that will make my model more accurate?
I've been trying to figure out what the topics are in a corpus of 1,000 articles that I've put together. The articles are all about a certain topic. The topics haven't been good enough. I tried LDA, BERTopic and NMF too and I got a idea about integrating the article titles for topic modelling but got no resources on the internet.Can anyone help me out?
Related
I am interested in looking at the entity or aspect level sentiment for various tweets. Are there any models that have already looked at this problem? I tried looking, but I couldn't really find anything. If not, how would I go about creating a custom model?
I am trying to use Gensim LDA modelling to topic model of dataset of food recipes. I wish to have topics based the key ingredients in the recipe. But the recipe text contains more words that are generic English and are not ingredient names. Hence my topic outcome is not as good as expected. I am trying to understand the impact of word frequency in the LDA topic outcome. Thanks.
Have you tried removing stop-words from the data on which you construct LDA model?
Also, please bear in mind that it is not really possible to influence the assignment of words among the topics. This has been discussed in the answer to this question: how to improve word assignement in different topics in lda
I am a beginner in NLP and I am recently considering the possibility of applying LDA Topic Modelling to Ancient Chinese Poems.
I am consider how many poems would I need to feed into the model to have a relatively good performance if possible. Can anyone gives some suggestion? I know it's a bit vague, but I just want to have a rough idea about it.
For topic modelling I use the method called nmf(Non-negative matrix factorisation). Now, I want to visualise it.So, can someone tell me visualisation techniques for topic modelling.
Check LDAvis if you're using R; pyLDAvis if Python. It was developed for LDA. But I guess it also works for NMF, by treating one matrix as topic_word_matrix and the other as topic proportion in each document.
http://nbviewer.jupyter.org/github/bmabey/pyLDAvis/blob/master/notebooks/pyLDAvis_overview.ipynb
I highly recommend topicwizard https://github.com/x-tabdeveloping/topic-wizard
(full disclosure: it was written by me)
It's a highly interactive dashboard for visualizing topic models, where you can also name topics and see relations between topics, documents and words.
Here are some example screenshots:
I'm looking to use Mallet to classify different documents by topics that I have defined. I know that Mallet will first determine the topics, then classify the documents but I want to skip the first step because I already have a list of topics with words associated with them. Is there any way to use pre-defined topic lists that I have created to classify documents with Mallet?
Any guidance is appreciated. Thanks!
If you're doing unsupervised learning (without training examples, i.e. docs for each topic), you cannot trivially just set the topics. The point is that the training algorithm does not know anything about the docs in advance. It just tries to separate/distribute them, based on the features you provide.
If you're doing supervised learning, topics are actually classes and you have documents for each class. Then the algorithm tries to learn which features are significant for each class. In mallet you should use the Classification module.
There are probably some fancy topic modelling ideas, which incorporate / skew the topic distributions according to specific keywords, but I don't think that's possible with Mallet.