Author: Olaf Kopp
Reading time: 6 Minutes

Reordering search query results in accordance with search context specific predicted performance functions

Topics: , , ,

Rate this post

The patent titled “Reordering Search Query Results in Accordance with Search Context Specific Predicted Performance Functions” revolves around improving search result relevance by customizing the order of results based on the search context and predicted performance functions. The invention is particularly focused on tailoring search engine outputs to better meet individual user needs or preferences by taking into account various contexts like user behavior, query nature, and historical data.

The patent is critical for the processes of personalization an reranking.

Patent ID: US8645390B1
Countries: United States
Last Publishing Date: Feb. 4, 2014
Inventors: Bilgehan Uygar Oztekin, Daria Antonova, Kedar Dhamdhere, Finnegan Southey, Zhenyu Mao
Assignee: Google


The background of the patent outlines the challenge of providing relevant search results to users due to the generic nature of search queries. Traditional search engines generate the same set of results for all users, regardless of their individual interests or contexts. For instance, a query like “apple” could refer to the technology company or the fruit, but traditional engines do not distinguish between these differing user intents.

This one-size-fits-all approach often leads to search results that include items of little or no interest to the user, highlighting the need for a system that can personalize search results based on the searcher’s history or preferences without requiring explicit input from the user. The patent aims to address this need by introducing a method and system for customizing search results in a way that is more likely to match the searcher’s intent, thus improving the overall search experience.


The claims of the patent focus on methods and systems for processing search queries by reordering search results according to search context-specific predicted performance functions. These are the key points derived from the claims:

  • Search Context Determination: For each of a variety of search contexts and scoring primitives, the method involves determining correlations between the scoring primitives and actual user selections from previously executed search queries across multiple users.

Scoring primitives are defined components or metrics used to evaluate and score search results in relation to their perceived relevance or usefulness to a user’s query within a specific search context. These scoring primitives are critical to the process of reordering search results based on context-specific predicted performance functions. Here’s a closer look at what scoring primitives could involve:

    • Query-Relevance Score: A metric assessing how closely the content of a search result matches the query terms. This can be based on keyword matching, semantic analysis, or other text analysis techniques.
    • User Interaction Metrics: Data derived from historical interactions of users with search results, such as click-through rates, dwell time on a page, or the frequency of a result being chosen relative to its position in the search results.
    • Content Quality Indicators: Measures of the quality or trustworthiness of the content, which could include domain authority, content freshness, or the presence of verified information.
    • Contextual Relevance: A score that evaluates how well a search result fits the context of the search, which could include the user’s location, the time of day, the device used for the search, or the user’s search history.
    • Social Signals: Metrics that reflect the popularity or endorsement of content within social networks or communities, such as shares, likes, or mentions.
  • Machine Learning Application: Utilizing machine learning on these correlations to identify a predicted performance function, which consists of a weighted subset of the scoring primitives that meet predefined predictive quality criteria. Each search context is associated with its respective predicted performance function.

  • Execution and Reordering of Search Queries: When a user submits a search query, the method includes associating this query with a specific search context and ordering the search results based on the predicted performance function relevant to the associated search context. This aims to enhance the relevance of the search results to the user’s needs or preferences.

The patent describes using scoring primitives in a machine learning framework to determine their correlation with actual user selections of search results. By analyzing historical data on how users interact with search results within various search contexts, the system can identify patterns and preferences that are not immediately obvious.

For each search context, a predicted performance function is derived through machine learning. This function comprises a weighted subset of the scoring primitives, where the weights are determined based on how strongly each primitive correlates with desirable user outcomes (e.g., selecting a search result). The predicted performance function is then used to reorder search results for queries that match the context, with the goal of improving the relevance of the results presented to the user.

The reordering of search query results involves both context-specific adjustments that can be applied during a user’s search session and potentially broader applications that might influence the ranking of search results for other users in general.

During the User’s Session

    • Personalization: The method uses historical user interaction data to personalize search results during an individual’s search session. By analyzing past queries and selections within a specific context, the system dynamically adjusts the ordering of search results for that session. This personalization aims to better align with the user’s immediate needs and preferences based on their historical interactions.
    • Context-Specific Adjustments: The adjustments are made according to the search context identified for the current query. Context can include factors like the user’s location, time of search, device used, and the nature of the query itself (e.g., commercial intent, informational need). The system utilizes predicted performance functions tailored to these contexts, ensuring that the reordering is relevant to the specific circumstances of the search.

General Reordering for Other Users

    • Learning from Collective User Behavior: While the method prioritizes session-specific adjustments, it also learns from the collective behavior of all users. This learning process involves analyzing broad patterns in how users interact with search results across various contexts. The insights gained can lead to adjustments in the predicted performance functions that are applied not just to individual sessions but potentially across similar search contexts for other users.
    • Updating Predictive Models: Over time, the machine learning algorithms update the predictive models based on new user interaction data. These updates can refine the system’s understanding of which search results are most valuable to users in different contexts. As a result, even users not directly involved in the learning data can benefit from improved search result ordering when their queries fall into contexts that have been optimized through collective user data.

Implications for SEO

1. Emphasize User Experience and Engagement:

Given that user interaction metrics (like click-through rates, dwell time) are considered as scoring primitives, SEO strategies should prioritize content that engages users effectively. This means creating content that not only attracts clicks but also retains user attention, encouraging them to interact with the page longer.

2. Contextual Relevance is Crucial:

SEO efforts need to consider the search context more deeply, including the user’s location, device, time of search, and potentially their search history. Tailoring content and SEO strategies to fit different contexts can increase the likelihood of content being ranked higher for relevant queries.

3. Quality Over Keywords:

While keywords remain important, the patent underscores the importance of content quality and relevance. SEO strategies should focus on creating high-quality, authoritative content that genuinely meets users’ needs, rather than overly focusing on keyword optimization.

4. Monitor and Adapt to User Behavior:

SEO strategies should be dynamic, adapting based on real-time data on user behavior and preferences. This includes adjusting content and strategies based on how users interact with both your site and your competitors’ sites within the search results.

5. Leverage Social Signals:

The importance of social signals as scoring primitives suggests that content with higher social engagement may be deemed more relevant. SEO strategies should, therefore, include efforts to promote content sharing and engagement across social platforms.

6. Diversify SEO Tactics:

Given the variety of scoring primitives that could influence search result rankings, SEO strategies should be multifaceted, incorporating a wide range of tactics from on-page SEO to social media engagement and beyond.

7. Personalization is Key:

As search engines move towards more personalized results based on historical user interaction data, understanding and targeting audience segments becomes increasingly important. SEO efforts should aim to understand and cater to the specific needs and search behaviors of different user groups.

9. Be Prepared for AI and Machine Learning Influence:

The use of machine learning to identify and adjust scoring primitives for reordering search results suggests that SEO strategies need to be prepared for shifts in how search engines evaluate and rank content. Staying informed about advancements in AI and machine learning within search engines can help in adjusting SEO strategies accordingly.


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