Contextually generated perceptions
Embodiments of the present invention train multiple Perception models to predict contextual metadata (tags) with respect to target content items. By extracting context from content items, and generating associations among the Perception models, individual Perceptions trigger one another based on the extracted context to generate a more robust set of contextual metadata. A Perception Identifier predicts core tags that make coarse distinctions among content items at relatively higher levels of abstraction, while also triggering other Perception models to predict additional perception tags at lower levels of abstraction. A Dense Classifier identifies sub-content items at various levels of abstraction, and facilitates the iterative generation of additional dense tags across integrated Perceptions. Class-specific thresholds are generated with respect to individual classes of each Perception to address the inherent sampling bias that results from the varying number and quality of training samples (across different classes of content items) available to train each Perception.
1. A system that generates contextual metadata relating to one or more target content items, the system comprising:
(a) a plurality of perception models (embodied in non-transitory computer memory and processed by a physical computer processing unit), each of which, during a first process for training perception models, is trained to predict a plurality of tags, wherein each tag corresponds to a class of training sample content items used to train the perception model; and
(b) an association between a first class of a first trained perception model and a second class of a second trained perception model of a subset of the tags predicted by a first trained perception model with a class of a second trained perception model, wherein the association is stored in the non-transitory computer memory and, during a second process for using integrated trained perception models, is employed by the first trained perception model, with respect to a target content item, to trigger the second class of the second trained perception model and generate a tag corresponding to the triggered class.
2. The system of claim 1 , further comprising:
(a) a plurality of models, including a first model and a second model (each of which is stored in non-transitory computer memory and processed by a physical processing unit), wherein, during the first process for training perception models:
(i) the first model is trained to predict a first tag by submitting to the first model a first set of training sample content items pre-tagged with the first tag;
(ii) the first model is trained to predict a second tag by submitting to the first model a second set of training sample content items pre-tagged with the second tag;
(iii) the second model is trained to predict a third tag by submitting to the second model a third set of training sample content items pre-tagged with the third tag; and
(iv) the second model is trained to predict a fourth tag by submitting to the second model a fourth set of training sample content items pre-tagged with the fourth tag; and
(b) an association (stored in the non-transitory computer memory) between (i) a subset of tags generated by the first model when presented with the third set of training sample content items during the second process for using integrated trained perception models and (ii) the third tag;
(c) wherein the first model, when presented with a target content item during the second process for using integrated trained perception models (i) generates a set of predicted tags and, if the set of predicted tags matches the subset of tags in the association, (ii) triggers the second model to predict the third tag.