- Joined
- Apr 18, 2024
- Messages
- 24
- Likes
- 8
- Degree
- 0
Hi @eliquid ...
I'm still learning ALL kinds of "high value" topics on this journey... You helped me a lot with my understanding of these topics. I was re-reading this thread... its sooo good. Like reading a book a third time... you pickup new items you missed or didn't quite understand correctly the first 2 times.
I'm assuming the answer to the question below is, "it depends", but wanted to ask.
When you add negatives, do you have a general rule when you use ngrams vs actual terms (as the source to start with)? I know there are several ways and sources to generate a negative list.
1) I see the problems with using ngrams (as you stated it above). Funny, I've used 2 different tools and actually coded my own ngram tool - the issue is real
2) assuming there isn't a real problem with using actual search terms?
Yes, you might have more of negative terms when using search terms and/or it takes longer to create the "accurate" list... BUT that isn't a reason NOT to use them.
Before I wrote this post/question:
I believe my answer is this:
==> I was trying to do too much, too fast using ngrams for an account / campaign / a group that was too big. Thus I spent time doing it and making sure it was good and accurate. Before going live with it, I spot checked a few via google's data and found problems
I'm thinking now:
a) IF only using ngrams, it needs to be a small / concise reason or group.
b) if not small, always use actual search terms (more exact match negatives) to prevent mistakes
c) if time is limited ( use ngram as the very early starting point, but NEVER only use ngrams)
d) always make more time
thanks for your input, feedback
I'm still learning ALL kinds of "high value" topics on this journey... You helped me a lot with my understanding of these topics. I was re-reading this thread... its sooo good. Like reading a book a third time... you pickup new items you missed or didn't quite understand correctly the first 2 times.
I'm assuming the answer to the question below is, "it depends", but wanted to ask.
When you add negatives, do you have a general rule when you use ngrams vs actual terms (as the source to start with)? I know there are several ways and sources to generate a negative list.
1) I see the problems with using ngrams (as you stated it above). Funny, I've used 2 different tools and actually coded my own ngram tool - the issue is real
2) assuming there isn't a real problem with using actual search terms?
Yes, you might have more of negative terms when using search terms and/or it takes longer to create the "accurate" list... BUT that isn't a reason NOT to use them.
Before I wrote this post/question:
I believe my answer is this:
==> I was trying to do too much, too fast using ngrams for an account / campaign / a group that was too big. Thus I spent time doing it and making sure it was good and accurate. Before going live with it, I spot checked a few via google's data and found problems
I'm thinking now:
a) IF only using ngrams, it needs to be a small / concise reason or group.
b) if not small, always use actual search terms (more exact match negatives) to prevent mistakes
c) if time is limited ( use ngram as the very early starting point, but NEVER only use ngrams)
d) always make more time
thanks for your input, feedback