Nonnegative Matrix Factorization based Heterogeneous Graph Embedding Method for Trigger-Action Programming in IoT
ABSTARCT :
Nowadays, users can personalize IoT devices/Web services via Trigger-Action Programming (TAP). As the number of connected entities grows, the relations of triggers and actions become progressively complex (i.e., the heterogeneity of TAP), which becomes a challenge for existing models to completely preserve the heterogeneous data and semantic information in trigger and action. To address this issue, we propose IoT-NMF, a Nonnegative-Matrix-Factorization-based heterogeneous graph embedding method for TAP. Prior to using IoT-NMF, we map triggers and actions to an IoT Heterogeneous Information Network (IoT-HIN), from which we can extract three structures that preserve heterogeneous relations in triggers and actions. IoT-NMF can factorize the structures simultaneously for getting low-dimensional representation vectors of the triggers and actions, which can be further utilized in Artificial Intelligence of Things (AIoT) applications (e.g., TAP rule recommendation). Lastly, we demonstrate the proposed approach using an IFTTT dataset. The result shows that IoT-NMF outperforms the state-of-the-art approaches.
EXISTING SYSTEM :
? Network analysis has attracted considerable attention as networks exist in various complex systems, such as biological and social systems.
? Network analysis heavily relies on the network representation, which is traditionally represented as discrete adjacency matrix.
? However, this straightforward representation usually cannot well reflect the underlying distinct structural characteristics of networks and suffers from the data sparsity issue.
? The representations of nodes within a community should be more similar than those belonging to different communities.
? Therefore, whether the learned embedding space can well reflect the community structures in the original network is a critical requirement for network embedding methods.
DISADVANTAGE :
? These algorithms usually result in an eigenvalue decomposition problem and the communities can be identified from the resultant eigenvectors.
? However, it is generally hard to tell the exact physical meanings of those eigenvectors, which is important for explaining the final results when associated with real world applications.
? In this paper, we propose how to apply nonnegative matrix factorization based methods to solve the community discovery problem.
? In this paper, we will investigate another important issue, community discovery, in network analysis.
PROPOSED SYSTEM :
• In this paper, we propose a novel Modularized Nonnegative Matrix Factorization (M-NMF) model to incorporate the community structure into network embedding.
• We proposed a novel Modularized Nonnegative Matrix Factorization (M-NMF) model for network embedding, which preserves both the microscopic structure (firstand second-order proximities) and mesoscopic community structure.
• A number of community detection methods have been proposed from different perspectives.
• M-NMF0 is our proposed M-NMF model which only preserves the first- and second-order proximities.
ADVANTAGE :
? We further compare the performance of the proposed Symmetric Nonnegative Matrix Factorization (SNMF) method with other competitive algorithms.
? The data set we used here is the WebKB data set4, which consists of about 6000 web pages from computer science departments of four schools.
? Efficiently identifying those communities can help us to know the nature of those networks better and facilitate the analysis on those large networks.
? The users that are familiar with many users (via user network) while they do not watch similar movies or give similar ratings as their friends can be thought as boundary spanners.
|