I have a Sparse Ill formated SMS messages generated for the purpose of Spam Filtering . I am using it to extract the topic/label (s) of each but when I used Topic Modeling (LDA) the result was not effective specially as I've previously done it manually from my understanding of each message .Is there an automatic way to do it ?
Here is a sample of the messages
I was thinking since I have a manual labeling of each to use it as a classification problem (multiclass classification) .I want to get the possible topics with he probabilityof each to rank/arrange them . So shall I manually multilabel my training data and use the Naiive Bayes to
LDA works for large datasets.you may need to tune its parameters. LSA works good if you have your seed topics.you may use Henson library for this purpose.you mayc consider manipulating short texts before proceeding to supervised natura l language similarity methods.
https://radimrehurek.com/gensim/
Article Using Latent Semantic Analysis to Identify Quality in Use (Q...
Conference Paper Semi-supervised Max-margin Topic Model with Manifold Posteri...
I worked more in the preprocessing phase and uses POS by allowing nouns only to pass to the model which gave me better results but I had to do a trial and error and the best fir K was chosen based on my interpretation . Is this ok or I should use BTM instead ?
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification