|
I need to find some user intentions in a corpus of dialogues. I wonder if you recommend LDA or HTMM for intention recognition, and if so which of those may work better. |
|
The paper Finding topics in e-mails: is LDA enough? by Joty et al, asks a similar question, wether or not LDA can work in email conversations. Their conclusion is to use neither LDA nor HTMM but a model that explicitly incorporates quotation information. I'm also not sure that intentions would be necessarily well-modeled as topics or "hidden markov topics". I'd suggest you look for a way of guiding your model towards words that necessarily express intention. |
