Method and apparatus for labeling data

A device, comprising:

• a processing system including a processor;
• and
• a memory that stores executable instructions that, when executed by the processing system, facilitate performance of operations, the operations comprising:
• generating a plurality of textual documents describing content of a media catalog, wherein the generating of the plurality of textual documents comprises applying image pattern recognition to other documents that include text and images;
• applying a group of Latent Dirichlet Allocation (LDA) topic models to a corpus associated with the plurality of textual documents, wherein the LDA topic models have different numbers of topics;
• comparing, via a coherence model, the group of LDA topic models to select a number of labels for classes of the corpus;
• applying, to the corpus, K-means clustering and an unsupervised learning process that generates vector representations of words to cluster the words into similar meaning groups;
• determining feature sets from the clusters of words for each of the classes using a cosine similarity function, wherein the feature sets represent the labels;
• generating trained models by assigning each of the plurality of textual documents to a selected one of the labels via a voting mechanism that counts matches of words in each document of the plurality of textual documents to the labels;
• receiving and processing a raw textual document to generate a processed document by applying tokenizing, removing stop words, one of bigram and trigram modeling, and name entity recognition analysis to the raw textual document, wherein the applying of the removing of stop words is based on a type of media content corresponding to other content described by the raw textual document and comprises an identification of characters associated with a type of the raw textual document;
• selecting a corresponding model from among the trained models according to a class of the classes that is applicable to subject matter of the raw textual document;
• automatically assigning the processed document to a target label of the labels of the corresponding model via the voting mechanism that counts matches of words in the processed document to the labels of the corresponding model;
• and
• generating engagement modeling, cancellation modeling, or both for one or more subscribers according to consumed content that includes the other content and according to the target label of the other content.

A non-transitory machine-readable medium, comprising executable instructions that, when executed by a processing system including a processor, facilitate performance of operations, the operations comprising:

• generating a plurality of textual documents describing content of a media catalog, wherein the generating of the plurality of textual documents comprises applying image pattern recognition to other documents that include text and images;
• determining classes from a corpus associated with the plurality of textual documents based on machine learning associated with topic modeling, data clustering analysis and an unsupervised learning process that generates vector representations of words resulting in vectors;
• generating trained models for each of the classes by determining labels for each of the classes based on the vectors and by assigning each of the plurality of textual documents to a selected one of the labels based on a highest number of matches of words in each document of the plurality of textual documents to the labels;
• processing a raw textual document to generate a processed document by applying Natural Language Processing to the raw textual document, the applying of the Natural Language Processing to the raw textual document comprising applying tokenizing, removing stop words, one of bigram and trigram modeling, and name entity recognition analysis to the raw textual document, wherein the applying of the removing of stop words is based on a type of media content corresponding to other content described by the raw textual document and comprises an identification of characters associated with a type of the raw textual document;
• selecting a corresponding model from among the trained models according to a class of the classes that is applicable to subject matter of the raw textual document;
• automatically assigning the processed document to a target label of the labels of the corresponding model based on a highest number of matches of words in the raw textual document to the target label;
• and
• generating engagement modeling, cancellation modeling, or both for one or more subscribers according to consumed content that includes the other content and according to the target label of the other content.