Hello… may I question something?

I am a student studying at a foreign university lab.

I am studying stock price forecasting.

Adjusting the LSTM unit and layer is complete.

The topics(plan) of the next research were auto-keras or NLP.

​

But the professor said that the topic was too broad.

he wants Step by step to build up my capabilities.

So I looked for all the paper related to stock predict.

Recent research was based on the attention mechanism or the process of using natural language like news.

The following research themes should be decided, but the gap between basic research and the latest research is too large.

I do not know what to make the next research topic.

​

In the case of CNN, there are various networks such as LeNet, AlexNet, ZFNet, GoogLeNet, VGGNet, and ResNet

However, LSTM is SOTA(LSTM: A Search Space Odyssey https://arxiv.org/abs/1503.04069 ) in RNN

So, in the case of RNN is LSTM only

I think that methodology of improving accuracy of LSTM should be the subject of research.

do you have any course for forecasting time series data?

I do not know how to get directions

Your valuable opinions and thoughts will be very much appreciated.