Comparing Slogan Generation Methods
Genetic
-
Enjoy the part like water for sentinel. (0.958)
- Enjoy a track of routine on sentinel. (0.958)
- Make all safety in safe for a sentinel. (0.958)
- Demand and enjoy the ease in sentinel (0.958)
-
Write a base for demand on sentinel. (0.948)
-
Demand the of potential as sentinel. (0.945)
-
Enjoy a sentinel performance show. (0.922)
-
Themes for head on sentinel. (0.913)
- Contents with application of sentinel. (0.913)
- Make the sentinel performance plays. (0.897)
BrainSup
What level of creativity? Why?
Genetic
BrainSup
Comparing Evaluation
Both
-
Length - More points for being shorter
-
Entity - Mentioning the thing that is being advertised
-
Frequent Words - Try to only use simple words
- Bigram Semantic Relatedness - More frequent 2-grams, more semantically coherent
-
Bigram - Grammar
-
Domain Relatedness / Semantic Similarity - Does it match the prompt?
-
Diversity / Variety - Are words repeated?
-
Target Word / Keyword - Is a specific word mentioned?
- Emotion Color / Polarity Subjective
-
Semantic Cohesion - Measure similarity between the words in the sentence
-
Phonetic Features - Rhymes, Alliteration, Plosives
-
Unusual Word - more points for being unusual
- Dependency Likelihood - Sentence makes sense
Do you prefer complex evaluation functions or ensembles of simple evaluation functions?
Are there places in your project where you could evaluate smaller parts and then make a weighted combination of those evaluations?
Where do your evaluation functions fall?
Where do you want them to be?
What are the places in your project where you enforce the rules?
What are the places in your project where you are using reward?
Would it make sense to invert the method on some of these?
Dictatorship vs Capitalism
What do you think about the automatic fixes in the genetic paper?
BrainSup: External Evaluation
-
Catchiness
-
Humor
-
Relatedness
-
Correctness
- Success
Do we always need to validate with external evaluation by humans?
