Perhaps a noob question. But how would somebody go around doing sentiment analysis not on document level but on a topic level? Imagine I have a corpus of speeches and I want to know speaker's sentiment on a specific topic because I know that speech is 'super' long and is filled with economic development, migration, nuclear weapons, etc. I imagine I could create the topics myself and create 'tags' that would identify those topics in speeches or use a computer to do it unsupervised something like lsa. But how would I go about combining these two features for instance?
Thank you so much for this. I have one question (a basic one, sorry I am new to this) - In the Lexicon based approach I see the Sentiment Score column for each document (something like 0.151), how do I get the overall score, do I take an average of all these scores.
Do you mean the overall score by author (or movie)? You can use a GroupBy node, group by author (or movie) and use for example an average aggregation function.
thank you sooo much. nice presentation, clear and concise! thank you also for sharing the models, datasets and demonstrating the whole thing. Also thanks for the gift code.