Author: Olaf Kopp
Reading time: 4 Minutes

Modifying search result ranking based on implicit user feedback

Topics: , , , ,

Rate this post

The patent describes a system that improves search result rankings based on how long users view each result. It calculates relevance by comparing the number of longer views to shorter views, with higher relevance boosting the ranking of a result in future searches.

General Information

  • Patent ID: US10229166B1
  • Assignee: Google LLC
  • Countries: United States
  • Last Publishing Date: March 12, 2019
  • Inventors:
    • Hyung-Jin Kim
    • Simon Tong
    • Noam M. Shazeer
    • Michelangelo Diligenti
  • Expiration Date: November 2, 2026


The background of the patent explains that search engines aim to provide the most relevant documents to users based on their queries. To do this, they analyze various clues about user needs, such as device type and location, and use techniques like PageRank, which ranks documents based on the number and quality of links to them. Additionally, the system can use user interactions, like clicks and viewing time, to determine and rank the relevance of search results, assuming that user behavior is a strong indicator of relevance.

Figure 3 illustrates another example information retrieval system. It showcases a user (1002) interacting with a client device (1004) to submit a query. This query is sent to a server system (1014) hosting a search engine (1030). The search engine processes the query and retrieves relevant documents, which are then ranked and presented to the user via the client device. The system includes components such as an indexing engine (2010), scoring engine (2020), ranking engine (2030), and rank modifier engine (2070) to enhance the search results’ relevance.


  • Measure of Relevance:
    • The system determines the relevance of a document based on the ratio of longer views to shorter views.
    • This relevance measure is used to rank search results.
  • Tracking User Behavior:
    • Tracks individual user selections (clicks) on document results within the context of a search query.
  • Weighted Views:
    • Weighs document views based on viewing length to produce weighted views.
    • Combines weighted views to determine the relevance measure.
  • Weighting Methods:
    • Applies continuous functions to document views for weighting.
    • Applies discontinuous functions to classify views into categories and assign weights.
  • Viewing Length Differentiators:
    • Weights views based on the category of the search query.
    • Weights views based on the type of user generating the selections.
  • Independent Relevance:
    • The measure of relevance is independent of other document results returned in response to the query.
  • Rank Modifier Engine:
    • The system includes a rank modifier engine that uses user feedback to adjust the ranking of search results.
  • Per-Query Data Collection:
    • Collects data on a per-query basis to determine user preferences for document results.
  • Click Fraction:
    • Transforms user click data into a click fraction to re-rank future search results.
  • Presentation Bias Reduction:
    • The relevance measure aims to reduce presentation bias in the search results shown to users.
FIGS. 4B and 4C:
Illustrate example weighting functions used to adjust the relevance of document views.
FIG. 4B shows a continuous weighting function.
FIG. 4C shows a discontinuous weighting function, categorizing viewing times and assigning weights accordingly.

Ranking Factors or Scoring Criteria

The document does not mention specific, concrete scoring criteria or ranking factors in detail. However, it does outline general methods and factors used in determining and adjusting the relevance of search results. Here are the main points:

  • User Interaction Data:
    • Tracks user clicks on search results.
    • Measures the duration of time users spend viewing each document (viewing length).
  • Viewing Time:
    • Longer views are given higher relevance.
    • Shorter views are given lower relevance.
  • Weighting Functions:
    • Applies continuous and discontinuous weighting functions to adjust relevance scores based on viewing time.
    • Classifies viewing times into categories and assigns weights accordingly.
  • User Context:
    • Differentiates weights based on the type of user and the category of the search query.
  • Implicit User Feedback:
    • Uses user behavior data (clicks and viewing times) to inform and adjust the ranking of search results.
    • Transforms user clicks into a “click fraction” to influence future rankings.

SEO Implications Based on the Patent Information

  1. Prioritize User Engagement:
    • Ensure content is engaging enough to encourage longer viewing times. The patent emphasizes that longer views correlate with higher relevance scores. Therefore, creating high-quality, engaging content that keeps users on the page longer can positively influence search rankings.
  2. Optimize for User Intent:
    • Align content closely with user intent and query categories. Since the system adjusts relevance based on the category of the search query and the type of user, understanding and addressing the specific needs and intentions behind user queries can improve relevance scores and rankings.
  3. Improve Click-Through Rates (CTR):
    • Focus on improving CTR by optimizing titles and meta descriptions. User clicks are tracked and factored into the ranking process. Compelling and accurate titles and descriptions that attract clicks can boost your relevance scores and, subsequently, your search rankings.
  4. Reduce Bounce Rates:
    • Ensure a seamless and relevant user experience to minimize short views and bounces. Since shorter views are given lower relevance scores, reducing bounce rates by providing valuable and easily accessible information can help maintain longer user engagement and improve rankings.


Content from the blog

LLMO: How do you optimize for the answers of generative AI systems?

As more and more people prefer to ask ChatGPT rather than Google when searching for read more

What is the Google Knowledge Vault? How it works?

The Google Knowledge Vault was a project by Google that aimed to create an extensive read more

What is BM25?

BM25 is a popular ranking function used in information retrieval systems to estimate the relevance read more

The dimensions of the Google ranking

The ranking factors at Google have become more and more multidimensional and diverse over the read more

Interesting Google patents for search and SEO in 2024

In this article I would like to contribute to archiving well-founded knowledge from Google patents read more

What is the Google Shopping Graph and how does it work?

The Google Shopping Graph is an advanced, dynamic data structure developed by Google to enhance read more