Search Ranking Bandit Algorithm for New Web Resources

Solution Overview

Problem

Existing search engine systems struggle to adequately rank web resources lacking prior-history features, leading to inaccurate search results and increased user searches, as these resources are less likely to receive user interaction-based feedback.

Innovation Solution

Implementing a Bandit Algorithm that combines machine learned algorithms to predict relevancy parameters and exploration scores, allowing for the inclusion of lower-ranked web resources in search engine results, thereby collecting user feedback and improving their ranking over time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a ranking algorithm uses only pre-feedback information to rank web resources, then web resources with high pre-feedback scores are displayed, but web resources lacking prior-history features cannot be adequately ranked even if they are highly relevant

Engineering Contradiction:

Improveranking accuracyVSAvoidability to rank new web resources

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary actions by proactively displaying web resources lacking prior-history features in higher positions before sufficient user feedback is available. This allows the system to collect feedback data in advance, enabling future accurate ranking of these resources without waiting for natural feedback accumulation.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system introduces a feedback mechanism where user interactions with web resources lacking prior-history features are actively collected and stored. This feedback data is then used to update the ranking algorithm, improving the system's ability to rank previously unranked resources accurately in future search queries.

Inventive Principle:

Principle #23Feedback

2Measurement precision

If web resources lacking prior-history features are displayed in lower positions, then accurate search results are provided based on available data, but user feedback is not collected and these resources remain unrecommended

Engineering Contradiction:

Improvesearch result accuracyVSAvoiduser feedback data

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The system performs preliminary actions by proactively displaying web resources lacking prior-history features in higher positions before sufficient user feedback is available. This allows the system to collect feedback data in advance, enabling future accurate ranking of these resources without waiting for natural feedback accumulation.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system introduces a feedback mechanism where user interactions with web resources lacking prior-history features are actively collected and stored. This feedback data is then used to update the ranking algorithm, improving the system's ability to rank previously unranked resources accurately in future search queries.

Inventive Principle:

Principle #23Feedback

3Stability of the object's composition

If the ranking algorithm only reinforces top-ranked web resources with positive feedback, then stable ranking is maintained, but new potentially relevant web resources are not discovered

Engineering Contradiction:

Improveranking stabilityVSAvoiddiscovery of new web resources

Core Design Contradiction:

Stability of the object's compositionVSAdaptability or versatility

Solution Approach 1:

The system introduces dynamic adjustment to the ranking algorithm by varying the display positions of web resources lacking prior-history features across different search queries. This dynamic approach allows the system to explore new resources while maintaining overall ranking stability through controlled experimentation rather than rigid reinforcement.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system performs preliminary actions by proactively displaying web resources lacking prior-history features in higher positions before sufficient user feedback is available. This allows the system to collect feedback data in advance, enabling future accurate ranking of these resources without waiting for natural feedback accumulation.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS10102292B2Method and system of processing a search query

Publication Date: 2018.10.16 Y E HUB ARMENIA LLC

AI summary

There is disclosed a computer implemented method for processing a search query, the method executable by a search engine server, search engine server coupled to a crawled web resource database and a communication network. The computer implemented method aims at placing lower ranked web resources (for example, due to lack of prior user interaction with these web resources, as they may be new, etc) on the upper positions of the SERP to attract more user feedback to gather information about the lower ranked web resources. In this way, the search engine provider may improve the search results mix by giving a chance to get user feedback (and, hence, improve their scores) for more potentially highly relevant web resources (which may yet lack user interaction data to allow proper high scoring).