There are two ways to use the topic model diagnostics included topicdoc. You can calculate all the topic diagnostics at once using topic_diagnostics or use the other functions to calculate the diagnostics individually.
The only prerequisite for using topicdoc is that your topic model is fit using the topicmodels package and that your document-term matrix (DTM) is slam coercible. This includes DTMs created through popular text mining packages like tm and quanteda.
For this example, the Associated Press Dataset from topicmodels is used. It contains a DTM created a series of AP articles from 1988.
library(topicdoc)
library(topicmodels)
data("AssociatedPress")
lda_ap4 <- LDA(AssociatedPress,
control = list(seed = 33), k = 4)
# See the top 10 terms associated with each of the topics
terms(lda_ap4, 10)
#> Topic 1 Topic 2 Topic 3 Topic 4
#> [1,] "i" "percent" "bush" "soviet"
#> [2,] "people" "million" "i" "government"
#> [3,] "two" "year" "president" "united"
#> [4,] "police" "billion" "court" "president"
#> [5,] "years" "new" "federal" "people"
#> [6,] "new" "market" "new" "police"
#> [7,] "city" "company" "house" "military"
#> [8,] "time" "prices" "state" "states"
#> [9,] "three" "stock" "dukakis" "party"
#> [10,] "like" "last" "campaign" "two"Here’s how you would run all the diagnostics at once.
topic_diagnostics(lda_ap4, AssociatedPress)
#> topic_num topic_size mean_token_length dist_from_corpus tf_df_dist
#> 1 1 3476.377 4.1 0.3899012 24.08191
#> 2 2 1910.153 5.6 0.5044673 26.67523
#> 3 3 2504.622 5.4 0.3830014 26.46131
#> 4 4 2581.848 6.5 0.3988826 25.52163
#> doc_prominence topic_coherence topic_exclusivity
#> 1 1053 -81.83339 7.813034
#> 2 598 -79.50691 9.560433
#> 3 783 -106.40062 9.162590
#> 4 775 -84.46149 9.058854Here’s how you would run a few of them individually.
topic_size(lda_ap4)
#> [1] 3476.377 1910.153 2504.622 2581.848
mean_token_length(lda_ap4)
#> [1] 4.1 5.6 5.4 6.5A full list of the diagnostics included are provided below.
| Diagnostic/Metric | Function | Description |
|---|---|---|
| topic size | topic_size |
Total (weighted) number of tokens per topic |
| mean token length | mean_token_length |
Average number of characters for the top tokens per topic |
| distance from corpus distribution | dist_from_corpus |
Distance of a topic’s token distribution from the overall corpus token distribution |
| distance between token and document frequencies | tf_df_dist |
Distance between a topic’s token and document distributions |
| document prominence | doc_prominence |
Number of unique documents where a topic appears |
| topic coherence | topic_coherence |
Measure of how often the top tokens in each topic appear together in the same document |
| topic exclusivity | topic_exclusivity |
Measure of how unique the top tokens in each topic are compared to the other topics |