Mon, 14 Jun 2021

Using word2vec for locating negative keywords

27 May 2021, 00:24 GMT+10

Looking within the suitable means at your campaigns search queries offers a high potential for optimizing your performance. We tend to have approaches to analyze n-grams of the users' questions and map all relevant KPIs like value Per Orders, Conversion Rates, worth Per Click, etc. Wait a moment! n-Grams, what's that? Let's say we've the subsequent search query:

Hugo boss Onlineshop

1-Grams would be Hugo boss Onlineshop

2-Grams: Hugo boss Hugo Onlineshop boss Onlineshop.

When you are doing this over thousands of queries, you'll find some attention-grabbing patterns that perform entirely differently. Let'sLet's say the performance is incredibly unhealthy. The action would be to feature some negative keywords for that pattern.

This approach works quite well if you have enough sample size (e.g., Clicks > 100) on an n-Gram pattern - to drawback is that there is still a high quantity of sporadic words wherever you waste loads of cash on however it might want long to review all of them. However, will we tend to modify that?

Build a Word2Vec model for similarity searches

Word2Vec was fabricated at google and is victimization neural networks to make up models that perceive the context of words. There are pre-trained models out there for many languages - in our case, we tend to build our model with all offered search queries we tend to purchase within the past.

Adwords improvement Use Case: notice negative search patterns and add them as negative keywords

Let's say we've one Adwords account that contains keywords for all product brands, for instance, +hugo +boss. We complete by staring at the n-Gram analysis that some folks are looking out for the pattern brand + location. The search intent is to shop local and not online. That'sThat's the rationale that cities like "berlin" and "Hamburg" are doping up with relatively high CPOs and unhealthy conversion rates if they seem in the search query. Ok, time to question the model - let's use "berlin" as input and find some similar words:

Wow, pretty spectacular results! The output shows similar words at the side of their similarity value. For that case, I restricted the development to the highest 20s. What will it mean for my negative keyword list now? Based on simply 1-Gram ('berlin'), wherever we've enough click data, the model suggests an inventory of terribly similar words that presently don't have enough samples to boost our attention by just observing the n-Gram list. We tend to use the output to feature a large quantity of the latest negative keywords that forestall to acquire future search queries that are unlikely to convert.

So the total method sounds like this now:

  • Classify negative search patterns supported their n-Gram knowledge wherever we've enough data
  • Use this classification as input for querying our word2vec model and obtain thousands of comparable words

This ballroom dance approach works nicely for Google AdWords accounts

that are using:

- Phrase, Modified Broad, or Broad match keywords

- Google Shopping

- Dynamic Search Ads

If you're interested in inducing a demo for our Querylyzer SaaS module with an Associate in Nursing web-based interface for that method (with Google AdWords API interface for straightforward action rollout), please write to

More Oregon News

Access More

Sign up for Oregon State News

a daily newsletter full of things to discuss over drinks.and the great thing is that it's on the house!