Author: Olaf Kopp
Reading time: 5 Minutes

Refining search results

Topics: ,

Rate this post

The patent describes a computer-implemented method for enhancing the process of refining search results based on the analysis of user queries and search sessions. It focuses on receiving data that represents a search query, identifying relevant search results, and adjusting these results based on predefined criteria related to user behavior and characteristics.

  • Patent ID: US9418104
  • Countries Published For: United States
  • Last Publishing Date: August 16, 2016
  • Assignee: Google Inc., Mountain View, CA (US)
  • Inventors: Hyung-Jin Kim (Sunnyvale, CA, US), Oleksandr Grushetskyy (Cupertino, CA, US), Andrei Lopatenko (Cupertino, CA, US)

This patent belongs to the field of ranking, reranking and scoring.


The background of patent US9418104 discusses the challenge of identifying and presenting documents or items relevant to a user’s needs in a useful manner within internet search engines. It touches upon the complexity of essentially “mind-reading”—inferring what the user wants based on various clues, some of which might be specific to the user (like the device they’re using or their location) and others that are more general (such as the assumption that if a web page is linked by many other pages, it might be more relevant).

The background highlights traditional techniques for determining the relevance of documents, including analyzing the level of backlinks to a document and user interactions with search results, like click rates on certain results, to gauge their relevance. These approaches are based on the idea that external references to a document and user selection behaviors are good indicators of a document’s relevance to a search query. It also briefly mentions the importance of identifying and eliminating attempts to artificially inflate the relevance of a page through manipulative practices.

This section sets the stage for the patent by outlining the existing challenges and methods in refining search results and the need for more sophisticated techniques to improve search relevance and user satisfaction.


The claims of patent US9418104 focus on a computer-implemented method for refining search results based on analyzing user query information and adjusting the search results according to a set of predefined criteria related to user characteristics and behaviors. Key aspects covered in the claims include:

  • Receiving Query Information: The method begins by receiving data representative of a user’s search query from a user search session.
  • Identifying Search Results: Based on the search query, a plurality of search results is identified. Each search result is associated with user characteristics and data representing requester behavior relative to previously submitted queries.
    • Upon receiving the search query, the system identifies a plurality of search results. This identification process likely involves searching a database or index of web pages, documents, or other content types that have been previously crawled and indexed by the search engine.
    • The identification process is highly dependent on the algorithms used by the search engine, which may consider factors such as keyword matches, semantic analysis of the query, and possibly the user’s search history or profile.
  • Ordering Based on Requestor Behavior: The method involves ordering the user characteristics for each search result based on the data representing requester behavior, which is associated with previously submitted queries and the respective search result.
    1. Computing the Metrics: For each search result, the system computes metrics based on the collected data that quantitatively represent the behavior of past requestors. These metrics could include click-through rates, average time spent on the page, bounce rates, follow-on query rates, and more.
    2. Establishing Order Based on Behavior: The search results are then ordered based on these metrics. Results that historically resulted in more engagement or were deemed more relevant based on user interactions are prioritized. This ordering is a dynamic and continuous process, adapting to new data and patterns in user behavior.
    3. Adjustment for Contextual Relevance: The order of search results is further refined by taking into account the specific context of the current search query and the characteristics of the current user. This means that the historical data is filtered and weighted to highlight interactions from users with similar characteristics or under similar contexts to the current search.
  • Adjusting User Characteristics: Adjustments are made to the ordered user characteristics based upon predefined compatibilities related to the user characteristics, such as language or location compatibilities. Initially, each search result is associated with a set of user characteristics, which may include information like the user’s language preference, geographic location, device type, search history, and behavioral patterns.
  • Ranking Search Results: The search results are then ranked based upon the adjusted user characteristics, aiming to enhance the relevance of the search results to the user’s query and preferences.
  • Compatibility Definitions and Adjustments: The claims detail how compatibilities might define relationships between user characteristics (e.g., language or location), and how these definitions can guide the adjustment of the order of user characteristics and the ranking of search results.
  • Application of Weights and Removal of Characteristics: Some claims specify the application of weights to the data representing requester behavior and the removal or adjustment of data associated with certain user characteristics based on their compatibility or relevance.
  • Combining Data for Ranking: In some claims, the method includes combining data related to user behavior for multiple languages or locations associated with the search requester, thereby providing a more nuanced approach to ranking the search results.

The Rank Refiner Engine and the Rank Modifier Engine work closely together within the search engine framework to refine and personalize search results for users. However, they serve distinct roles in the process of adjusting and ranking search results based on user behaviors and characteristics. Here’s a comparative analysis of their differences:

Rank Refiner Engine

  • Functionality: The Rank Refiner Engine primarily focuses on the detailed analysis and processing of user characteristics and requestor behavior data. It examines historical interactions and preferences to determine how these factors should influence the ranking of search results.
  • Data Analysis: This engine is heavily involved in analyzing data to identify patterns, preferences, and the relevance of search results based on user interactions. It looks at the compatibility of user characteristics (e.g., language, location) and adjusts the significance or weight of these characteristics in the context of search queries.
  • Adjustment Criteria: The Rank Refiner Engine utilizes predefined compatibility criteria to adjust the ordering of user characteristics and their influence on search result rankings. It is responsible for refining the raw data into actionable insights that can inform how search results should be personalized.

Rank Modifier Engine

  • Functionality: The Rank Modifier Engine takes the insights and adjusted data from the Rank Refiner Engine (or similar inputs) and applies these adjustments to the actual ranking of search results. Its primary function is to modify the search result rankings based on the refined data.
  • Ranking Adjustments: This engine is directly involved in altering the order of search results before they are presented to the user. It applies the adjustments recommended by the Rank Refiner Engine, ensuring that the search results are ordered in a way that aligns with the user’s inferred preferences and needs.
  • Dynamic Interaction: While the Rank Refiner Engine is more static, focusing on analyzing and adjusting data based on historical behavior, the Rank Modifier Engine is dynamic, actively changing search result rankings in real-time based on the latest inputs and adjustments.


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