Sony Corporation

Sony Electronics Inc

Chiranjit Acharya

Mark Earl Plutowski

Tsunayuki Ohwa

Takashi Usuki

Khemdut Purang

A computer-implemented method includes receiving a dataset representing a plurality of users, a plurality of items, and a plurality of ratings given to items by users; clustering the plurality of users into a plurality of user-groups such that at least one user belongs to more than one user-group; clustering the plurality of items into a plurality of item-groups such that at least one item belongs to more than one item-group; inducing a model describing a probabilistic relationship between the plurality of users, items, ratings, user-groups, and item-groups, the induced model defined by a plurality of model parameters; and predicting a rating of a user for an item using the induced model.

User's preference prediction from collective rating data

A computerized method of representing a dataset includes obtaining a dataset, the dataset defining an attribute space; decomposing the attribute space into a plurality of attribute subspaces; generating a parent taxonomy of the obtained dataset with respect to one of the plurality of attribute subspaces, the parent taxonomy organizing the obtained dataset into a plurality of data subsets; generating a child taxonomy with respect to another one of the plurality of attribute subspaces, the child taxonomy organizing each of the plurality of data subsets within the parent taxonomy into at least one data subset; iteratively repeating generating the child taxonomy until a predetermined termination condition is satisfied, wherein the child taxonomy of a preceding iteration is the parent taxonomy of the current iteration; and assigning category labels to the data subsets.

Method and system for subspace bounded recursive clustering of categorical data

A computer implemented method including determining whether a predetermined condition is satisfied with respect to a case-based dataset stored within memory accessible by at least one processor; generating, for each of a plurality of rule-based classifiers, rule-based classification data identifying class boundaries between the records upon determining that the predetermined condition has been satisfied; computing a structural risk of the rule-based classification data for each of the rule-based classifiers with respect to the records identified within the record data; selecting rule-based classifiers having generated rule-based classification data identifying class boundaries between the records with a structural risk below a predetermined threshold; identifying selected rule-based classifiers having rule-based classification data that is within a predetermined degree of similarity with the case-based classification data; and replacing the case-based dataset stored within the memory with rule-based classification data of at least one of the identified rule-based classifiers, wherein determining whether the predetermined condition is satisfied comprises determining whether the number of records identified within the record data exceeds a predetermined record threshold.

Method and system for transitioning from a case-based classifier system to a rule-based classifier system

A computerized method of representing a dataset with a taxonomy includes obtaining a dataset containing a plurality of records; initializing a folder-set containing a plurality of folders; assigning labels to folders within the folder set; and classifying the plurality of records into the plurality of folders according to a predetermined entropic similarity condition.

Computer program product and method for folder classification based on folder content similarity and dissimilarity

A computerized method of representing a dataset with a taxonomy includes representing a plurality of records within a dataset as a plurality of clusters in an initial taxonomy layer; generating a hierarchy of cluster sets based on clusters within the initial taxonomy layer, wherein the structure of the hierarchy is independent of the order of the records within the dataset; and labeling clusters within the hierarchy.

Method and system for order invariant clustering of categorical data

A computer implemented method including determining whether a predetermined condition is satisfied with respect to a case-based dataset stored within memory accessible by at least one processor; generating, for each of a plurality of rule-based classifiers, rule-based classification data identifying class boundaries between the records upon determining that the predetermined condition has been satisfied; computing a structural risk of the rule-based classification data for each of the rule-based classifiers with respect to the records identified within the record data; selecting rule-based classifiers having generated rule-based classification data identifying class boundaries between the records with a structural risk below a predetermined threshold; identifying selected rule-based classifiers having rule-based classification data that is within a predetermined degree of similarity with the case-wise classification data; and replacing the case-based dataset stored within the memory with rule-based classification data of at least one of the identified rule-based classifiers.

Records including category data is clustered by representing the data as a plurality of clusters, and generating a hierarchy of clusters based on the clusters. Records including category data are classified into folders according to a predetermined entropic similarity condition.

Clustering and classification of multimedia data

A computerized method of representing a dataset with a taxonomy includes obtaining a dataset comprising a plurality of records, the dataset being characterized by a vocabulary and each of the plurality of records being characterized by at least one term within the vocabulary; identifying nearest neighbors for each term within the vocabulary; imputing a degree of membership for each nearest neighbor identified for each term within the vocabulary; augmenting the obtained dataset with the imputed degree of membership; and generating a taxonomy of the augmented dataset.

Augmented dataset representation using a taxonomy which accounts for similarity and dissimilarity between each record in the dataset and a user's similarity-biased intuition

A computerized method of representing a dataset with a taxonomy includes augmenting a dataset containing a plurality of records with a plurality of predetermined exemplars; representing the plurality of records and predetermined exemplars within the augmented dataset as a plurality of clusters in an initial taxonomy layer; generating a truncated hierarchy of cluster sets based on clusters within the initial taxonomy layer, wherein clusters within the truncated hierarchy contain no more than a predetermined number of exemplars; and labeling clusters within the truncated hierarchy.

Method and system for seed based clustering of categorical data using hierarchies

A computerized method of representing a dataset with an optimal layer of a taxonomy includes obtaining a taxonomy including a hierarchical arrangement of layers, wherein each layer represents a cluster set containing at least one cluster, wherein each cluster represents at least one record within a dataset; identifying a range of taxonomy layers based on a measure of intra-cluster homogeneity of each cluster within the taxonomy; selecting a taxonomy layer within the identified range as an optimal layer of the taxonomy, the selecting being based on a measure of inter-cluster heterogeneity between clusters of a taxonomy layer within the identified range; and labeling clusters within the optimal layer of the taxonomy.

Optimal taxonomy layer selection method

System for folder classification based on folder content similarity and dissimilarity

A model selection process for selecting a finite mixture model that is suitable for a given task is described. A set of finite mixture models (FMMs) is produced from a set of parameter values and training data for a given task using an Expectation Maximization (EM) process. An FMM from the set of FMMs is selected based on a minimum description length (MDL) value calculated for each of the FMMs.

Growing community of inventors

San Jose, CA, United States of America

Chiranjit Acharya