A method of training an artificial neural network (ANN) having a first branch and at least one other different second branch where each branch has a plurality of inputs and a plurality of outputs and each branch is tasked to process a specific derivable property derived from measurable signal qualities of content of a data file containing at least one of audio content, image content and text content, the method comprising:
- for a selected pair of different data files, extracting measurable signal qualities from content of each of a first data file and a different data file of said pair;
- identifying a plurality of different extracted properties from those measurable signal qualities for each data file, where each different extracted property is represented by a processable vector;
- selectively inputting processable vectors for specific derivable properties as input to a corresponding one of said first branch and said at least one other different second branch respectively tasked with processing those different specific derivable properties;
- at an output of said first and said at least one other different second branch, generating a corresponding property space vector in property space for the respective specific derivable property of the first data file and the different data file of the selected pair, each property space vector describing a measurable signal quality extracted for respective content of respectively both the first data file and the different second data file;
- assembling a first multi-dimensional vector for the first data file and a distinct second multi-dimensional vector for the different data file, wherein each multi-dimensional vector is obtained from assembling respective property space vectors from outputs of said first branch and said at least one other different second branch;
- determining a distance measure between the first multi-dimensional vector and the second multi-dimensional vector;
- deriving a subjectively-derived semantic vector using natural language processing (NLP) of a text description of content for each of the first data file and the different second data file, wherein each subjectively-derived semantic vector describes content in semantic space respectively for the first data file and also the different second data file and each subjectively-derived semantic vector is generated independently of the property space vectors;
- assessing the subjectively-derived semantic vectors against corresponding property space vectors by correlating, in the ANN, quantified semantic dissimilarity measures for the subjectively-derived semantic vectors with related property separation distances for the property space vectors to provide an output that is adapted, over time, to align a result in property space to a result in semantic space;
- in response to said determined distance measure, adjusting weights of the ANN by a backpropagation process that takes into account identified discrepancies arising between said determined distance measure in property space and quantified semantic dissimilarity distance measures assessed in semantic space for the first data file relative to the different data file, thereby training the system to value semantic perception reflected in the quantified semantic dissimilarity distance measures over property assessment reflected by the distance measure between the first multi-dimensional vector and the second multi-dimensional vector and such that the ANN maps 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 train the ANN to identify and quantify similarity or dissimilarity in audio content, image content and text content, and
- to output, once trained, a measure of similarity between said content of said first data file relative to content in said second data file.