Collaborative Filtering Engine Using Action Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In an enterprise context, it is challenging to provide effective recommendations to users regarding objects they may want or need to interact with due to the scarcity of links between such objects, making it difficult to identify similar objects for recommendation.
Innovation Solution
A method for collaborative filtering of enterprise objects is implemented, where vectors representing actions performed on objects are created and compared using similarity functions like Tanimoto similarity, allowing for the calculation of similarity scores between objects, which are then used to recommend similar objects to users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional recommendation methods are used in enterprise context, then implementation is simple, but recommendation quality is poor due to scarcity of links between objects
Solution Approach 1:
The patent transforms the recommendation approach by changing the parameters used for comparison from direct object links to action-based vectors. Instead of relying on sparse object-to-object links, the system creates vectors representing actions performed on objects and computes similarity based on these action patterns, fundamentally changing the parameter space for recommendation quality
Solution Approach 2:
The patent introduces action vectors as an intermediary between objects and recommendation logic. Rather than directly comparing objects or using simple link analysis, the system uses action vectors as a mediating representation that captures behavioral patterns, enabling more reliable recommendations while maintaining manageable system complexity
2Measurement precision
If action-based vectors are created and compared for all object pairs, then recommendation accuracy improves, but computational complexity increases
Solution Approach 1:
The patent improves measurement precision by changing from binary link existence to continuous action-based vector similarity. The Tanimoto similarity computation on action vectors provides a nuanced, continuous measure of object similarity based on action patterns, significantly improving measurement accuracy compared to traditional binary or count-based methods
Solution Approach 2:
The patent applies partial action by computing similarity scores for object pairs selectively rather than exhaustively. The system calculates action vector similarities only when needed for recommendations, and uses thresholding to filter results, avoiding the excessive computational burden of computing all possible pairwise similarities in the enterprise
Data Source
AI summary
Various systems and methods for collaborative filtering of enterprise objects are described. A first object and a second object, which may be enterprise objects, may be compared. Vectors associated with each object may be created. These vectors may contain elements that indicate one or more actions performed involving other objects. Using these vectors, a similarity comparison may be performed to determine the amount of similarity between the objects. Based on a similarity value determined from the similarity comparison between the vectors, the second object may be recommended to a user.