Basic analysis pt 1

[MansMeg]

Below are issues we discussed at the last concept-history meeting.

• [x] Add seperate files with most distinct words (in file name) and top words (not distinct) so it is easy to compare the different type of top words
• [x] Produce one graph per year (or 5-year period) of token frequencies and one for proportions - one graph per sense from the model. We want to follow the senses development over time. X-axis is time and y-axis is frequency and proportions of word sesnses.
• [ ] produce sense per keyword and lemma (ie what senses are specific to propaganda, what are specific to information).
• [x] run similar models for Johans media words as well (and also for my migration words if easy).

[rbbby]

More on sense development over time: "de bolla et al, in their article on "liberty", has some interesting ideas we might want to follow up on. they count how many - and which - words that remain on toplists as time goes by. that way they can trace "core" senses, and those that are lost or added."

[MansMeg]

Cool! Do they use the same type of multinomial cluster model or just context words? It was probably 2 years since I read the deBolla article.

[rbbby]

Cool! Do they use the same type of multinomial cluster model or just context words? It was probably 2 years since I read the deBolla article.

They use context windows and word cooccurrences.

[MansMeg]

Ah, yes. Then we could use similar ideas but for the individual clusters.