Ranking-Based Network Embedding for Clustering Significance Manipulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network-based clustering methods fail to accurately measure the significance of documents due to the 'abusing' problem, where intentional manipulation of links or irrelevant words increase document significance, and they are unable to perform multi-labeling or non-labeling effectively, especially in large datasets like mobile apps.
Innovation Solution
A ranking-based network embedding method using a neural network that outputs vectors representing neighboring objects, trained to prioritize objects with high significance, employing the Net2Vec algorithm with an authority ranking loss function to prevent manipulation and enable multi-labeling and non-labeling by using external indicators independent of content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If network-based clustering methods use link analysis to measure document significance, then clustering speed and effectiveness are improved, but the system becomes vulnerable to abusing where intentional manipulation of links increases document significance falsely
Solution Approach 1:
The patent introduces an intermediary verification mechanism that checks whether neighboring objects of a document are semantically related to the document's content. This intermediary layer prevents direct manipulation of significance scores through arbitrary link structures, as the semantic relationship verification acts as a filter between the link analysis and significance calculation.
Solution Approach 2:
The patent implements a feedback mechanism where the significance of a document is validated by examining whether its neighboring objects (linked documents) are semantically relevant to the document's content. This feedback loop ensures that manipulated links without semantic meaning do not artificially inflate document significance, thereby maintaining measurement accuracy while preserving clustering speed.
2Device complexity
If traditional clustering methods assign documents to single clusters, then computational complexity is reduced, but the ability to perform multi-labeling and non-labeling is lost
Solution Approach 1:
The patent applies multi-functionality by enabling the clustering system to perform multiple operations: single-label clustering, multi-labeling (assigning documents to multiple clusters), and non-labeling (assigning no cluster). This universal approach allows the same system to adapt to different clustering needs without increasing computational complexity, as the underlying algorithm remains consistent while the output flexibility varies.
Data Source
AI summary
A clustering method includes configuring a network with clustering target objects; collecting significance of the clustering target objects; performing network embedding for outputting a set of vectors representing neighboring objects of the clustering target objects constituting the network using a neural network; and performing clustering on the clustering target objects using the set of vectors and information on each of the clustering target objects, wherein the neural network is trained so that neighboring objects having high significance are output with higher probability.