Sequence Model

In this blog we will explain a very important NLP concept that is commonly used in deep learning application such as speech recognition, machine translation and more which is sequent model. Important question:

What is Sequence model?

Sequence model is called sequence because input and output data are a function of time.

Example of sequence model application is:

- Turning written text to audio

- Music production where algorithms are applied in neural networks.

- Sentiment Classification of product

- DNA analysis: analyzing complicated coding of DNA into simple terms.

- Machine Translation: Translation from one Language to another through machine algorithms.

- Video Analysis. By analyzing every video frame,follow it and recognizing involved people

- Searching for names : in analyzing big texts and determining the proper names on it.

Sequence model has too many applications. They differ in shape and type, it is possible to find the input is a sequential value and the output is not like this or vice versa, the length of the input and output may be equal and may differ.

To explain sequence model, I will apply it on finding names on texts. This is called named- entity. This application teach the model how to search for specific names and apply it later on any list of names. The traditional method was to provide a Textual data, the output is list of zeros and ones for each word where zero means the name does not exists while one means word exists.

Let’s say I want to search for name of “Amazon” for example in text like:

Forbes listed Amazon the 2020 World’s Best Employers

The output array will be something like:

The idea of RNN depends on taking the first word in the text X <0> and entering it in the determination function, is it a name or not, and then the function outputs the number y <0> which determines whether it is zero or one, and then the network takes the second word X <1> It depends in some way on the result of the first word …
Then the command is repeated processing the third word X < 2 >, which depends on the result of processing the second word (and the first in some form) and so on until we reach the last word