Systems and methods for unsupervised streaming feature selection in social media
Systems and methods for exploiting link information in streaming feature selection, resulting in a novel unsupervised streaming feature selection framework are disclosed.
1. A method for managing high-dimensional data in streaming social media, the method comprising:
generating, by a processor, a model for unsupervised streaming feature selection from high-dimensional data defining a set of linked data instances, by:
modeling link information associated with the set of linked data instances via extracting social latent factors, by, for each of the set of linked data instances, assigning a k-dimensional social latent vector at a time step t of a plurality of time steps, and
modeling feature information of the high-dimensional data by generating a data matrix for the high-dimensional data, the data matrix representing similarities in features between different ones of the set of data instances, wherein the feature information is modeled to be consistent with the social latent factors, each of the set of features dynamically generated and arriving at the time step t from a social media stream; and
selecting, by the processor and utilizing both of the feature information and the link information of the model, a subset of features at each of the plurality of time steps to facilitate clustering including, including
utilizing the social latent factors as a constraint to perform feature selection through a regression model as the social latent factors are extracted for each of the set of linked data instances.
2. The method of claim 1 , wherein the social latent factors are extracted based on a mixed membership stochastic blockmodel.
3. The method of claim 1 , wherein each of the plurality of data instances is associated with a latent factor vector, the latent factor vectors of the plurality of data instances interacting with each other to form one or more social relationships.
4. The method of claim 3 , wherein the social latent factors are extracted from the one or more social relationships using a scalable interference algorithm.
5. The method of claim 1 , wherein the social latent factors are extracted using a scalable interference algorithm.
6. The method of claim 1 , wherein the link information is a directed graph or an undirected graph.
7. The method of claim 1 , wherein a subset of the most relevant features is identified at the time step given each social latent factor for the plurality of data instances.
8. The method of claim 1 , wherein selecting one or more relevant features in the social media stream from the one or more features through a regression model includes modeling feature information based on feature similarity between the plurality of data instances.
9. The method of claim 1 , further comprising:
identifying a new feature.
10. The method of claim 9 , wherein identifying the new feature includes determining whether to select the new feature from the one or more relevant features.
11. The method of claim 1 , further comprising:
determining whether to discard one or more existing features.