Is there a classic process or model for multimodal fusion using text and tabular data as input to the model? Note that the text corpora can be directly used as input to the model and should not be preprocessed into structured data before being combined with tabular data.
Bram Janssens
There is no single way to do so, you should just make sure that the model is capable of combining and interpreting both the text and structured data. For example, in Article 360 Degrees Rumor Detection: When Explanations Got Some Expl...
we use neural network architectures and a concatenation layer to combine both inputs, with the model implicitely learning the representations. But for our more traditional learners (e.g., tree-based architectures), we use embedding representations to structure the text data before feeding it to the model. Both methods do fairly well.
2 votes
1 thanks
- Badges
- Science topic
- Similar topics
- Philosophy
- Ethics
More Futian Weng's questions See All
Similar questions and discussions