I absolutely recognise the usefulness of algorithmic approaches in uncovering new insights but, given the all-embracing use of the term 'arts', I'm a little concerned about the sample here? It would be very interesting to know how the visual arts sector is stratified by gender and one might do this in the fine arts market via auction and gallery data (if it was available). However, the scope here seems to be limited to commercial media arts?
This reflects a problem across the piste in using algorithmic approaches in the creative sector. Algorithmic methods need access to substantial datasets in order to do their magic and this can marginalise the outputs and make-up of the relatively fragmentary and micro-businesses which make up most of the creative sector? If NESTA does have the data, however, we'd be very interested to see whether the relatively high level of women's participation across the visual arts is reflected in the most prestigious and lucrative sectors of the visual arts, for example?
Gender bias in visual representation and language has long been extensively documented via processes of close-reading visual arts and media and, whilst this kind of algorithmic textual analysis adds evidence, it doesn't, in itself, advance the general understanding of the way gender works in visual culture.
There are equally well-documented issues with using algorithmic text analysis alone in understanding popular culture. Indeed, it's a very blunt instrument to suggest that gender balance in media remains unchanged since WWII - representation of women is heavily inflected by historical context and whilst, as this data indicates, there may be a 'baseline' in representing gender, these tropes are mobilised very differently in different contexts in response either to critial motivations, changes outside the domains of gender, or in response to agitation by women themselves.
There is also a substantial body of experimental work in the visual arts engaging critically with the gendered 'representational regime' in popular and canonical culture which has driven a great deal of the innovation in gender representation in mainstream media - so, whilst taking media output as a whole may indicate an unchanged baseline, this fails to take account of substantial shifts in the way women are represented in key genres or authorial works.
A body of work exists also demonstrating that audiences are variously complicit and/or dissident in reading texts. Whilst these insights are interesting - and make great headlines - it remains important that algorithmic approaches are utilised as additional evidence informing human reading and analytic techniques and not regarded as sufficient in themselves.