Items of interest within digital information may be detected and associated with a label that provides context to the item of interest. The label may describe an item category of the item of interest. The knowledge base of item categories may be limited. Additional item categories may be learned by accessing sets of vocabulary that may relate to the known item categories.
FIELD OF THE DISCLOSURE(1) This disclosure relates to vocabulary-informed categorization of items of interest included within digital information.BACKGROUND(2) Machine learning techniques may be used to identify and/or contextually categorize items of interest depicted in digital media content and/or other digital information. For example, digital media content may comprise digital images. Items of interest may comprise objects detected in the images. Some techniques may include one-shot learning, zero-shot learning, open set recognition, visual-semantic embedding, and other techniques. One-shot learning techniques may be configured to learn object categories from one, or only few examples. The examples may be referred to as “training data.” To compensate for the lack of training data and to enable one-shot learning, knowledge may be obtained from other sources, for example, by similarity of features, semantic attributes, and/or other information. Zero-shot learning may be configured to recognize novel categories of detected object with no training data by obtaining knowledge from auxiliary categories. For example, zero-shot learning may explore the use of attribute-based semantic representations. An attribute vector prototype of each category must be pre-defined which may be very computationally expensive for a large-scale dataset. In some instances, semantic word vectors may be used to embed a given class name without human efforts; they can therefore serve as an alternativ
1. A system configured for vocabulary-informed categorization of items of interest included within digital information, the system comprising: one or more physical processors configured by machine-readable instructions to: store categorization information that includes associations of a set of semantic labels with a set of items of interest included in a set of instances of digital information, individual semantic labels describing individual item categories of individual items of interest; determine additional item categories, the additional item categories being determined based on vocabulary that is related to the item categories described by the set of semantic labels; update associations of one or more of the semantic labels included the set of semantic labels with one or more of the items of interest within the set of items of interest with one or more other semantic labels that describe one or more of the additional item categories; and store the updated associations within the categorization information; wherein determining the additional item categories comprises: accessing a web document comprising text; searching the text within the web document, and identifying phrases and/or sentences which include descriptions of the individual item categories included in the categorization information; identifying, within the text, a set of vocabulary including contextually similar vocabulary and/or synonymous vocabulary related to the phrases and/or sentences; and defining the additional item categories based on the set of vocabulary. |
10. A method of vocabulary-informed categorization of items of interest included within digital information, the method being implemented in a computer system comprising one or more physical processors and storage medium storing machine-readable instructions, the method comprising: storing categorization information that includes associations of a set of semantic labels with a set of items of interest included in a set of instances of digital information, individual semantic labels describing individual item categories of individual items of interest; determining additional item categories, the additional item categories being determined based on vocabulary that is related to the item categories described by the set of semantic labels; updating associations of one or more of the semantic labels included the set of semantic labels with one or more of the items of interest within the set of items of interest with one or more other semantic labels that describe one or more of the additional item categories; and storing the updated associations within the categorization information; wherein determining the additional item categories comprises: accessing a web document comprising text; searching the text within the web document, and identifying phrases and/or sentences which include descriptions of the individual item categories included in the categorization information; identifying, within the text, a set of vocabulary including contextually similar vocabulary and/or synonymous vocabulary related to the phrases and/or sentences; and defining the additional item categories based on the set of vocabulary.