Enterprise Search Ranking via User Action Tags
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Enterprise search systems face challenges in improving the relevance and quality of search results due to the lack of interlinking between documents, making traditional page rank methods like Google's ineffective.
Innovation Solution
A system that calculates search-independent object ranks based on user actions and connections between objects, using a damped, positive, column-stochastic matrix to determine eigenvector solutions, and combines these ranks with text matching scores for improved search result ordering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional page rank methods like Google's are used, then search results can be ranked based on interlinking, but this approach is ineffective for enterprise document systems since documents in such systems tend not to be highly interlinked
Solution Approach 1:
The patent changes the ranking parameters from link-based metrics to user-action-based metrics. Instead of relying on interdocument links, the system uses coefficients derived from user actions (viewing, tagging, creating) to calculate object ranks. This parameter change makes the system adaptable to enterprise environments where documents lack extensive interlinking but do have user interaction patterns.
Solution Approach 2:
The patent introduces user actions as an intermediary between documents and ranking. Rather than directly linking documents, user actions (viewing, tagging, creating) serve as mediators that establish relationships and influence rankings. This intermediary layer enables effective ranking in enterprise systems where direct document interlinking is sparse.
2Reliability
If search-independent object ranks are calculated based on user actions, then the ranking process can be isolated from real-time production systems, but this requires complex matrix calculations and data processing
Solution Approach 1:
The patent segments the search system into distinct components: a ranking calculation system that operates independently from the real-time production system. The ranking calculations are performed separately using matrix operations on coefficients derived from user actions, while the production system handles document storage and retrieval. This segmentation isolates the complex calculations from the production environment, maintaining data integrity while managing complexity.
Solution Approach 2:
The patent performs ranking calculations in advance, before actual search queries are executed. User action coefficients are collected and processed to compute object ranks beforehand using matrix operations. These pre-calculated ranks are then combined with text matching scores during search operations. This preliminary action separates the computationally intensive ranking process from real-time search operations.
3Measurement precision
If a flexible ranking schema that adapts to user interactions is implemented, then search results can be improved, but the system requires collecting and processing extensive user action data
Solution Approach 1:
The patent extracts only the essential user actions needed for ranking (viewing, tagging, creating) rather than processing all possible user interactions. These specific actions are used to compute coefficients that feed into the ranking matrix. By extracting only the relevant user actions, the system achieves precise ranking without being overwhelmed by the full volume of user data.
Solution Approach 2:
The patent applies different weights and coefficients to different types of user actions based on their local quality or importance. Not all user actions are treated equally; instead, the system assigns different significance levels to viewing, tagging, and creating actions. This local quality approach allows the system to process extensive user data efficiently by focusing computational resources on the most informative actions.
Data Source
AI summary
An interface can allow for associating documents with tags. A search system can use connections between the tags and documents to determine search-independent rank values for the documents. The search-independent rank values can be combined with term matching indications to get a total relevance of the document.