Emotion: compare methods validate on new data

[audrism]

No description provided.

[gaohsiung]

In order to compare the methods on our dataset reasonably, it is important to understand the current state of the art in the sentiment or emotion analysis. I think enough literature survey on the related paper is necessary. I will start from the paper archived in our google drive. I will summarize all the potential methods we could use on our dataset in one file, probably also including some extra interesting points or conclusions (Just based on my view, quite subjective, though).

A brief look at my summary:

[gaohsiung]

Continue the literature survey on the text analysis. This week I did it in my own way, and I draw a schematic map about it and just give a very broad review of what kind of methods it contains and also the current state of the art. The map will be continuously updated by me based on what kinds of ideas could be helpful for our analysis.

https://coggle.it/diagram/XHV_WRSzVvZ13Y8D/t/text-sentiment-analysis/93cfaf2c3732e25fd565634b47f86651962bdf0055fd078a18c5df6759ab5277

[gaohsiung]

I will use the self-attention model to do the word embedding, comparing with doc2vec model, and I will make some slides briefly talking about the attention mechanism, including its unique features and advantages compared with traditional word embedding, such as word2vec or doc2vec.