search

capstone496 / SpeechSentiments

1 stars 1 forks source link

Propose an overall structure for sentiment identification model (high level, machine learning method) #4

Open KatJHuang opened 6 years ago

rightnknow commented 5 years ago

Algorithm:One training model for text based data and one training model for audio-based data. These two models will taking the input from same source(Of course one will be transferred into text first, the other will be raw audio data)We will have a neural network at the end to do the decision.

To do: 1.Find proper dataset 2.Decide type of model for each training model

  1. Howe to tune the parameter

Deployment: Two way, need to decide:

  1. Cloud : Amazon Tensorflow https://aws.amazon.com/machine-learning/amis/
  2. Locla: Andriod NNAPI, Tensorflow Lite https://www.tensorflow.org/mobile/tflite/
rightnknow commented 5 years ago

Personally I prefer the cloud one, not only because tensorflow with cloud can be much powerful than local, but also we can collect data from user as well. We can also update our model with new data to improve our precision of the result.