Sequence Learning and sentiment analysis

Refer the screenshot from the class presentation below. Why is it called sequential when the prediction (in this case sentiment) is dependent on all the inputs (i.e. all the words in a sentence)? Off course, the sentiment depends on the bag of words and their sequence in the sentence (ex. to differentiate between a question and confirmed sentiment). But there are no intermittent prediction, rather at the end only. Also, are such algorithms/models (where there is no fixed inputs) have a characteristic inclination towards unsupervised training?

Any model that takes in variable-sequenced input, in which the items are temporally-related, can be called as a sequential model.

  • The sentences are ofcourse variable sized (number of words is not fixed)
  • Each word is linked to other words in the sentence (there are definitely not independent)

Hence we can call it a sequential model.

Types of sequence models: (Source)

The example (screenshot) you have shown is a type of many-to-one sequential model.


No, sequential models can be trained in supervised or unsupervised, however you want to model it.
The sentiment analysis model is an example of supervised models.
There are unsupervised sequential models too, like language models.