First published on LinkedIn on June 9, 2018.
Before IBM’s Watson there was the Practical Intelligent Question Answering Technology; referred to by IBM as 'Piquant'. Piquant was developed by the then Research Manager Charles Lickel, after the idea appeared in his head whilst at a lunch-outing with his colleagues at a nearby restaurant. He noticed that the entire restaurant was glued to the TV screens within the restaurant premises, witnessing the 74 game winning streak of Ken Jennings in the game of 'Jeopardy!'. Lickel’s idea was to create a system which could play and answer 'Jeopardy!' clues using what is now known as Predictive Annotation.
Piquant took life shortly after with the backing of the then Research Executive, Paul Horn; the idea was to explore how to best integrate and balance various technologies such as a knowledge base, NLP and planning and traditional text-based IR in order to build an efficient, modular, multi-agent Question Answering environment; which took 6 years to develop with a team of only four people, however this made it one of the top three Text Retrieval Conference QA Systems at the time. Piquant would usually take several minutes before it could respond with a success rate of 35% to 'Jeopardy!' clues, the idea was to build a system which could compete at the expert level of humans in real-time on 'Jeopardy!'.
To achieve this, Piquant would use local resources, as for the requirement was that to contest in the 'Jeopardy!' challenge the system had to be self-contained and not link to a live web-feed. This involved the development of a complementary approach of fetching data – the system would have to do a text search, using terms in the question presented to it, as queries and search engine scores as confidences for candidate answers generated from retrieved document titles; it would attempt to look for the right answer in the database by relating the queries from the clue. This failed to produce promising results and the team ended up overhauling nearly everything; from architecture metrics to evaluation protocols and in 2008 Piquant became the predecessor to Watson.
Watson would use a wide range of encyclopaedias, dictionaries, thesauri, newswire articles and literary articles, in combination with a refined design of the DeepQA architecture. With an overhauled system in place, Watson was able to play a series of sparring games against players that had previously appeared in 'Jeopardy! Tournament of Champions' matches, winning 64% of the games played.
Fast track to today and Watson is now being released to developers as a bundle of cognitive computing systems understanding natural language and having the ability to learn. With a collection of systems for all types of components, the idea of next generation advertising sprung into mind. Could IBM’s Watson be used to take advertising to the next level?
From several demo applications released earlier this year using Insights components to process real time Twitter and Facebook data streams in-combination with Personality Insights and Relationship Extraction modules, you are able to extract meaningful dataviews – allowing you to see users’ sentiments, make assumptions based on the content in their feeds for example on whether they are in a relationship and whether they have children; in-combination with other modules you can even allocate data to specific custom metrics such as brand keywords, language, followers, and area which then quickly allows you to build a detailed 360* view of the user; see their openness, conscientiousness, extraversion, agreeableness, emotional rage, needs, values and their brand specific awareness. The challenge is to create meaningful data reports from raw data and to communicate them with specific API’s such as Bing, Yahoo! and Google to use such data to better target and engage users, by displaying them with relevant content that matches their feeds. No longer will you need to crawl through numerous reports to find the data you need, to then push your creatives manually to your target markets.
Digital retailers such as roztayger.com are already implementing IBM’s Watson technology into their services; using Watson they were able to create a “Designer Match” to analyse your personality through your Facebook and/or Twitter feed to match you to items that may be of deep interest to you, based on your personality extracted from your social media feeds. This is just one Watson module, imagine the possibilities of combining other modules such as IBM’s Visual Recognition to analyse visual content of images and videos to understand content directly without the need for textual description – allow the application to see what makes you, you, not only through text but through visual media; combined with Cognitive Insights you have the possibility to get contextual and relevant observations and predictions about that user – see the differences in their fashion choices throughout the years and predict their future inspirations; invoking such technology to large clothing chains, outlets and retailers could increase not only positive brand outlook, but give retailers more bang for their buck when it comes to advertising. It may be as simple as asking do the users sentiments show a dislike for a certain brand? What trends have they followed throughout the years? Can their interests, personality, and other brand sentiments be used to better analyse your brand targeting and predict their future purchases based on the gathered data?
If yes, let your IBM application communicate with an advertisers API to promote and increase bidding for relevant ads using your tailored dataviews – does the user show negative sentiments towards a certain brand, are they millennials or gen x? Promote your creatives with relevancy to the user and target them through meaningful channels with meaningful content relatable to them and their past, present and predictable future sentiments.
Could IBM lead the future of digital marketing? And will you, as a potential advertiser adapt and accept the way we run and see marketing?