Content discovery engine vendor Jinni believes that social TV should be exactly what it says on the tin and be about getting people together around the TV again, watching something they all enjoy. TV’s history is rooted in family viewing of sitcoms and popular movies at a time when there was often little choice, so there were few arguments over what to watch. The fragmentation of content has had something to do with the decline in shared viewing but cultural factors compounded by the growth of single device one to one viewing have probably played a bigger role.
So when people talk about social TV today they are more likely to mean lone viewing laced with social interaction via companion devices. In this context Jinni’s new ‘Watch Together’ API could well find greater use among dispersed groups of friends wondering what they all might enjoy watching rather than to re-establish traditional family viewing by identifying the lowest common denominator.
In fact Jinni is leaning that way itself, pitching the app as much at identifying those friends on Facebook that happen to share your tastes and get together in a virtual living room sharing tittle tattle over the network. But it can also apply semantic search in an attempt to identify content from a given source, which could be all programs currently showing or all movies on at a cinema nearby, that a specified group of people would all enjoy.
Jinni claims it has established a lead in contextual search and recommendation and in being able to drill down via natural language processing to match individual tastes to content more accurately than with basic metadata techniques reliant on keywords. At one time it looked like Google agreed, having initially licensed Jinni for Google TV before deciding to plow its own furrow with its own semantic search project. This has already led to Google’s traditional keyword searches being enhanced with semantics to combine the best of the two.
However Google is not yet changing the way it ranks web pages, which is still determined by various purely objective measures associated with the words it contains, and how many other sites link to it. The aim is more to exploit semantics to refine the search process itself and create new opportunities for advertising related to more clearly defined user preferences and tastes.
Then in September 2011 Microsoft licensed Jinni for use in Xbox Live, with Spain’s leading pay TV service Prisa TV also becoming a customer. Microsoft was attracted by the ability to drill down via a number of different semantic categories. Broadly speaking there are two types of search, navigational aimed at locating a given object such as a book or particular piece of information, and semantic search where the objective is more blurry, to find a number of things people may be interested in to varying degrees. Microsoft is also incorporating Jinni into its Bing search engine in an attempt to make up ground on Google by applying what it believes is more nuanced and mature semantic search. Any lead is likely to be short lived given Google now has its eye on the ball as far as semantic search is concerned and is bringing great resources to bear upon it.
Semantic search really comes into its own for multiple keywords since it can exploit their combined meaning rather than merely their co-location. For example a search on Apple plus vice president would yield the names of the company’s management, just as a film title combined with actors would elicit the members of the caste.
Jinni’s engine goes deeper by attempting to capture moods and nuances expressed by search terms used and matching these to the assets available for viewing. The company employs the biological analogy of genetic matching, just as DNA profiling can be used to identify members of a family. It breaks content down into multiple ‘genes’ each coding for varying levels of mood, style, plot, setting and other parameters, with users being categorized in a similar way. Then matching is achieved by comparing the relevant genetic profiles, using techniques similar those that have evolved for DNA sequence matching in the life sciences.
What we do not have yet, unlike in the life sciences, are meaningful studies comparing the effectiveness, accuracy or popularity of different content discovery engines or methods. Such studies would be difficult to do, but would be interesting and there are at least proven statistical techniques available.