A newly published research paper by Google details a novel approach to enhancing the ranking of web pages. This algorithm promises significant advancements in deep neural network algorithms responsible for assessing relevance.
The algorithm introduces a ranking method known as Groupwise Scoring Functions.
While there’s no official confirmation from Google regarding its application, the researchers’ assertions of notable improvements suggest it might be in use. It’s not unreasonable to speculate that Google has adopted this algorithm.
Does Google Use Published Algorithms?
Google has previously indicated that research papers shouldn’t necessarily be seen as a reflection of current search engine practices.
Google infrequently confirms whether the algorithms mentioned in their patents or papers are implemented. This uncertainty also applies to the discussed algorithm.
Is this Algorithm Part of the March 2019 Core Update?
The research highlights Google’s commitment to understanding search queries and the nature of web pages. This aligns with the trend in Google’s recent research.
Google recently launched a broad core update, considered one of the most substantial in years. The connection between this algorithm and the update remains speculative, as Google rarely discloses details about specific algorithms.
In my opinion, this method could be one element of a comprehensive update to Google’s search ranking system, although it likely isn’t the sole component. The March 2019 Core Ranking Algorithm probably consists of a suite of enhancements.
Why this Algorithm is Important
The research paper begins with an observation that traditional machine learning algorithms evaluate web pages independently. They then compare these pages to determine relevance.
Here’s how the research describes existing algorithms:
"While in a classification or a regression setting a label or a value is assigned to each individual document, in a ranking setting we determine the relevance ordering of the entire input document list."
The new proposal suggests that considering the age of relevant web pages can offer insights into user intent. Instead of ranking pages solely against each other, analyzing their age first allows for a better understanding of user needs, leading to more accurate rankings.
The new algorithm is described as follows:
"The majority of the existing learning-to-rank algorithms model such relativity at the loss level using pairwise or listwise loss functions. However, they are restricted to pointwise scoring functions, i.e., the relevance score of a document is computed based on the document itself, regardless of the other documents in the list.
…the relevance score of a document to a query is computed independently of the other documents in the list. This setting could be less optimal for ranking problems for multiple reasons."
Cross-document Comparison
The paper outlines how current ranking strategies fail to fully realize the potential for improving search relevance.
An example is used to highlight this issue:
"Consider a search scenario where a user is searching for a name of a musical artist. If all the results returned by the query (e.g., calvin harris) are recent, the user may be interested in the latest news or tour information.
If, on the other hand, most of the query results are older (e.g., frank sinatra), it is more likely that the user wants to learn about artist discography or biography. Thus, the relevance of each document depends on the distribution of the whole list."
The age of relevant pages can refine search answers, enhancing the precision of result selection.
Modeling Human Behavior for Better Accuracy
The research further notes that users often compare search results in relation to other pages. The suggestion is that a ranking model utilizing similar comparisons would yield more accurate results.
"…user interaction with search results shows strong comparison patterns. Prior research suggests that preference judgments by comparing a pair of documents are faster to obtain, and are more consistent than the absolute ratings."
A greater predictive capability is achieved by modeling user interactions relatively. This approach indicates that users evaluate clicked documents against surrounding ones before selecting, suggesting that a model mimicking such behavior is more effective.
The New Algorithm Works
When evaluating algorithm research, it’s crucial to assess whether researchers claim it advances beyond the current methods.
Research indicating minimal improvements that require significant resources generally suggests it might not be ideal for Google’s algorithms. Conversely, algorithms demonstrating substantial advancements with minimal cost are more likely candidates for inclusion.
The researchers concluded that this innovative method enhances both Deep Neural Network and tree-based models, highlighting its practical value. While Google doesn’t reveal specific algorithm usage, the potential of this algorithm to yield significant improvements increases the likelihood of its future implementation.
Understanding information retrieval research allows insight into potential developments. Recognizing areas previously unstudied offers clues regarding Google’s activities.
For instance, SEO communities once assumed Facebook likes influenced rankings, yet research shows such assumptions were unlikely. This method’s proclaimed success underscores the importance of evidence-based algorithms.
In summary:
"Experimental results show that GSFs significantly benefit several state-of-the-art DNN and tree-based models…”
How this Can Help Your SEO
Ranking is shifting away from traditional factors. Long-standing metrics such as anchor text, headings, and links are diminishing in importance.
The concept of analyzing commonalities between relevant pages provides insights into user intent. Even if Google isn’t utilizing this specific algorithm, the approach remains valuable.
Understanding user needs enhances the capacity to craft pages that address these needs, potentially improving ranking abilities.
Focus on understanding user intent and creating relevant content.
For more details:
- Learning Groupwise Scoring Functions Using Deep Neural Networks (PDF)
More Resources
Images by Shutterstock, modified by the author.
