Dropping posts by % or proportion of words recognized by sentiment model

I have a dataset of 40K reddit posts and I am trying to estimate sentiment per post using a dictionary-based machine learning model. I am using a dictionary that contains 8K unique words and phrases to predict the sentiment.

One challenge I am facing is that for some posts, the dictionary only recognizes 1 positive and/or negative word, so I decided to exclude such posts from my dataset as I think it would be misleading to predict sentiment if the dictionary only captures one or two words at most. I coded this as follows:

#Loading packages
library(tidyverse)
require(readxl)
require(writexl)
library(quanteda)
library(stm)
library(stmCorrViz)
library(stringi)

#Filtering out posts where ONLY 1 positive and 1 negative words are recognized ##

valences_by_post_oneword <- valences_by_post  # subset data for posts where only one word was recognized 

one_pn<- valences_by_post_oneword %>% 
  filter(positive==1 & negative==1)

valences_by_post_oneword <- valences_by_post_oneword %>% 
  filter(!(positive==1 & negative==1)) 
  
#2011 & 2012
valence_oneowrd<-valences_by_post_oneword %>%
  filter(year == 2011 | year ==2012)%>%
  group_by(month_year) %>%
  summarize(mean_valence= mean(valence), n=n())

However, it makes more sense to use a % to exclude posts, rather than a specific number of words as the number of words varies across posts, so I would like to exclude posts where the dictionary only recognizes 5% or less of the total words in a given post, but I am not sure how to code this

Here is how the data currently looks like:

post   negative_words    positive_words   total_words valence_score
xyz.     2                1                  10            -0.66

The positive and negative word columns refer to the number of recognized words by the dictionary per reddit post,
while "total_words" refers to all words in a given post regardless of whether they were recognized, and "valence_score" is the estimate sentiment score by reddit post, which is measured as follows:

negative words + positive words/ total words in the given post

Since you already have a column for total words in the post, you can simply keep using filter(). For a dummy dataset like this:

library(tidyverse)

df <- tribble(
  ~post, ~negative_words, ~positive_words, ~total_words,
  "xyz", 2, 1, 10,
  "abc", 5, 3, 12,
  "def", 2, 1, 100,
)

You can do:

df %>% filter((negative_words + positive_words) > 0.05* total_words)

# OUTPUT
# A tibble: 2 × 4
  post  negative_words positive_words total_words
  <chr>          <dbl>          <dbl>       <dbl>
1 xyz                2              1          10
2 abc                5              3          12

which keeps the "xyz" and "abc" posts but filters out the "def" post since its total recognised words are lower than 5% of all.