As SEO and the field of data science draw closer, Google is taking a step forward by introducing an algorithm based on a convolutional neural network to better understand user intent.
What is neural matching?
Covering 30% of search queries, the neural matching algorithm aims to match search queries and web pages. This method connects words to concepts. As voice search seems to be the next way of using search in the years to come, being able to better understand concepts is important for Google’s user experience. There’s something important here: this method does not rely on link signals. Still, it uses pages that have already been ranked.
How does it work?
So, how might this neural matching algorithm perform? How can results still be relevant?
This method is inspired by the ad hoc retrieval one. I’m sure it is going to ring a bell… as document relevant ranking (another name used for this field) is using the TF-IDF score and the cosine similarity to match phrases in documents and needs expressed in search queries.
As the neural matching algorithm is only using words in search queries to match concepts, the ad hoc retrieval method is not sufficient. This is where the artificial intelligence comes into play. Here the technique of artificial intelligence unsupervised learning is used to move forward in the understanding from words to concepts.
Helped by the Deep Relevance Matching Model (DRMM) for ad-hoc retrieval, the neural matching algorithm is based on relevance. As indicated in the patent, this method employs a joint deep architecture at the query term level over the local interactions between query and document terms for relevance matching.
What impact will neural matching have for SEOs?
Here, beyond all the artificial intelligence issues, the question is: will this new algorithm have an impact on content writers and publishers?
What kind of queries are being included as part of the 30% that are covered by this new algorithm? We know the RankBrain algorithm covers about 15% of the search queries that have never been searched for before.
We have already begun to see the impact: We talk a lot about artificial intelligence and we now use some machine learning algorithms to better understand intricacies between SEO and business data and then to better drive SEO decisions.
How are we going to integrate this relevance matching technique (decorrelated from other ranking signals at this point) in our everyday workflow? From now on, SEOs who include Natural Language Processing techniques in their workflow should aim for semantic matching. However, relevance from the point of view of a super synonym (so named by Danny Sullivan) can still appear to be a spamming technique if it means including synonyms within the content.
This is a look back at a big change in search but which continues to be important: understanding synonyms. How people search is often different from information that people write solutions about. pic.twitter.com/sBcR4tR4eT
— Danny Sullivan (@dannysullivan) 24 septembre 2018
At the moment, we don’t yet know enough about the queries that are targeted or how content should be optimized. The current persistence of link-based connections between websites and penalties for synonym stuffing may hinder uninformed attempts to optimize for neural matching. However, it is clear that we will need to rethink how we create content in order to adapt.