Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content

Abstract

A property vector representing extractable measurable properties, such as musical properties, of a file is mapped to semantic properties for the file. This is achieved by using artificial neural networks “ANNs” in which weights and biases are trained to align a distance dissimilarity measure in property space for pairwise comparative files back towards a corresponding semantic distance dissimilarity measure in semantic space for those same files. The result is that, once optimised, the ANNs can process any file, parsed with those properties, to identify other files sharing common traits reflective of emotional-perception, thereby rendering a more liable and true-to-life result of similarity/dissimilarity. This contrasts with simply training a neural network to consider extractable measurable properties that, in isolation, do not provide a reliable contextual relationship into the real-world.

1 . A digital file containing media content and an embedded multi-dimensional property vector comprising a plurality of assembled property vector components each relating to a measurable property derived from a plurality of sets of quantized representations of signal qualities of the digital file, the multi-dimensional property vector realising a quantification of similarity/dissimilarity in media content between media content in the digital file relative to another different data file sharing media content modality with said digital file. 
     
     
         2 . The digital file of  claim 1 , wherein the digital file is an audio file and the embedded multi-dimensional property vector is a vector correlated to musical properties defined by measurable signal qualities indicative of rhythm, tonality, timbre and musical texture. 
     
     
         3 . The digital file of  claim 2 , wherein each of the assembled property vector components reflect a measure in vector dissimilarity space for the musical property assessed relative to reference vectors in semantic distance separation space. 
     
     
         4 . The digital file of  claim 2 , wherein the digital file is a product of a process that includes the steps of:
 comparing, in response to processing in an artificial neural network “ANN”, a subjectively-derived semantic vector against a property space vector, the subjectively-derived semantic vector generated externally from the ANN and independently of the property space vector generated from processing within the ANN, wherein
 i) the subjectively-derived semantic vector for each of a first data file and also a different second data file describes respective content in semantic space, and the subjectively-derived semantic vector is derived using natural language processing (NLP) of a text description for respective content in each of the first data file and the second data file, and 
 ii) the property space vector is derived from processing in the ANN of measurable signal quality extracted for respective content of both the first data file and the second data file as applied to the input of the ANN; 
   correlating a semantic dissimilarity measure quantified between subjectively-derived semantic vectors of said first and second data files with a property separation distance for the respective property space vectors of said first and second data files; and   through adaptation of said weights, configuring the ANN to map pairwise similarity/dissimilarity in property space for the first data file and the second data file towards corresponding pairwise semantic similarity/dissimilarity in semantic space for the first data file and the second data file to value semantic dissimilarity measures over property separation distances to produce, at the output of the AI system, property separation distances that converge to corresponding semantic dissimilarity measures.   
     
     
         5 . The digital file of  claim 1 , wherein the digital file is an image and the embedded multi-dimensional property vector comprising a plurality of assembled property vector components each relating to a visual property derived from a plurality of sets of quantized representations of measurable signal qualities of the digital image file and wherein the measurable signal qualities are measurable properties indicative of at least some properties selected from the group consisting of:
 brightness,   contrast,   color,   intensity,   shape,   relative size of a feature,   relative feature position, and   rate of change one or more of brightness, contrast, color, intensity, shape, relative size of a feature and relative feature position.   
     
     
         6 . The digital file of  claim 5 , wherein the digital file is a product of a process that includes the steps of:
 comparing, in response to processing in an artificial neural network “ANN”, a subjectively-derived semantic vector against a property space vector, the subjectively-derived semantic vector generated externally from the ANN and independently of the property space vector generated from processing within the ANN, wherein
 i) the subjectively-derived semantic vector for each of a first data file and also a different second data file describes respective content in semantic space, and the subjectively-derived semantic vector is derived using natural language processing (NLP) of a text description for respective content in each of the first data file and the second data file, and 
 ii) the property space vector is derived from processing in the ANN of measurable signal quality extracted for respective content of both the first data file and the second data file as applied to the input of the ANN; 
   correlating a semantic dissimilarity measure quantified between subjectively-derived semantic vectors of said first and second data files with a property separation distance for the respective property space vectors of said first and second data files; and through adaptation of said weights, configuring the ANN to map pairwise similarity/dissimilarity in property space for the first data file and the second data file towards corresponding pairwise semantic similarity/dissimilarity in semantic space for the first data file and the second data file to value semantic dissimilarity measures over property separation distances to produce, at the output of the AI system, property separation distances that converge to corresponding semantic dissimilarity measures.   
     
     
         7 . The digital file of  claim 1 , wherein digital file is a text file. 
     
     
         8 . The digital file of  claim 1 , wherein the digital file is a product of a process that includes the steps of:
 comparing, in response to processing in an artificial neural network “ANN”, a subjectively-derived semantic vector against a property space vector, the subjectively-derived semantic vector generated externally from the ANN and independently of the property space vector generated from processing within the ANN, wherein
 i) the subjectively-derived semantic vector for each of a first data file and also a different second data file describes respective content in semantic space, and the subjectively-derived semantic vector is derived using natural language processing (NLP) of a text description for respective content in each of the first data file and the second data file, and 
 ii) the property space vector is derived from processing in the ANN of measurable signal quality extracted for respective content of both the first data file and the second data file as applied to the input of the ANN; 
   correlating a semantic dissimilarity measure quantified between subjectively-derived semantic vectors of said first and second data files with a property separation distance for the respective property space vectors of said first and second data files; and   through adaptation of said weights, configuring the ANN to map pairwise similarity/dissimilarity in property space for the first data file and the second data file towards corresponding pairwise semantic similarity/dissimilarity in semantic space for the first data file and the second data file to value semantic dissimilarity measures over property separation distances to produce, at the output of the AI system, property separation distances that converge to corresponding semantic dissimilarity measures.   
     
     
         9 . A database containing a multiplicity of the digital file of  claim 1 . 
     
     
         10 . A digital file embedded with a multi-dimensional property vector realising a quantization of commonality or dissimilarity in contextual properties reflecting human-perceived qualities of the digital file, the digital file generated from a process trained by creating a multiplicity of pairs of independent vectors representing human-perceived qualities and measurable quantities of electronic source files, the process of creating the multiplicity of pairs further including:
 generating a first vector in semantic space based on measured dissimilarities in human-generated descriptions between pairs of source files such that the first vector provides a user-centric perception of pairwise closeness; and   generating a second vector from pairwise comparison of measurable properties extracted from content of the same pair of source files; and   adapting a process by which the second vector is generated such that pairwise distances for the first vector approach the pairwise distance for the second vector.   
     
     
         11 . A database containing a multiplicity of the digital file of  claim 10 .