#semantic processing

In computer science, a stack is an abstract data type that serves as a collection of elements, with two principal operations:

push, which adds an element to the collection, and

pop, which removes the most recently added element that was not yet removed.

Usos: 

When encoding information, we are more likely to retain it if we deeply process even a simple word list by focusing on the semantics (meaning) of the words.

Types of processing:

Deep: semantically (the meaning of the word), "type of..."; about 90% of people can recognize words this way

Shallower: (the sound of the word), "rhymes with..."; about 60% of people can recognize words this way

Most shallow: structural (the appearance of the letters), "Written in capitals?"; about 12% of people can recognize words this way

We're proud to join a group of top notch organisations from both research and industry that are working to build and implement a full-fledged cross-media analysis platform.

We all know from our daily online experience that media contents are on the rise. Digital photography and DIY videos are part of our information diet: more than 71% of online adults use video-sharing sites on a regular basis and 46% of Internet users post original photos and videos.   

While media contents increase in volume technologies for media analysis across different channels remain a tough challenge for many organisations around the World.

We as Insideout10 with our experience on semantic analysis developed in the past two years around our WordPress plugin WordLift and in partnership with our sister company Redlink (focused on developing a Platform-as-a-Service for Semantic Processing) are much aware that information extraction from multimedia documents is considerably harder than from textual documents. Text provides an underlying structure at both syntactical and semantic level; yes, multi-language semantic analysis is a huge topic and disambiguation remains a tough challenge for computer scientists but technologies like free-text search have been around for many years and do provide nowadays valuable means of information extraction. Media (being a photo, a video or an audio clip) on the other hand doesn't have a common structure and it is fragmented: deriving a context out of photos representing the same "thing" (a person, a place, a brand) across different channels is not as easy as detecting the same "thing" in various textual documents (where you can use part-of-speech tagging and combine natural-language-processing algorithms with information available in the Linked Data Cloud). 

MICO is a European Union Framework Programme 7 research and development project and it will bring a common model for combining different analysis techniques for media contents making these challenges far more easy than they are today. We're planning to use the same semantic processing framework we use for text analysis (being Apache Stanbol, Apache Marmotta and Apache Solr - see here for more information) and combining it with state-of-the-art media extractors developed by the Fraunhofer Institute (a leading research institution in Europe well known for its invention of the MP3 audio format). 

Key benefits for our clients and partners

Insideout10 (along with Interact Egypt and RealNetworks) will validate these technologies in our Online Video Platform Helix Cloud: a solution enterprises and telcos use for providing multi-screen video experiences across their networks (more info available here: Hybrid and OTT for Telco 3.0).

This will enable our clients the creation of value added services (content discovery, personalisation using the social graph, content augmentation, semantic advertising and more) that take advantage of the cross-media analysis MICO will provide. We are also planning to bring these enhancements to the blogging community using WordLift and our TV Plugin WeeoTV (used for the development of http://enel.tv).

Next steps

We're currently gathering user-stories and technical requirements and we will keep pushing out updates and experiences in this area in the next coming months: our goal is to bring cross-media analysis to a next level while making these technologies as accessible as possible for organisations and developers around the World.

As usual we're more than happy to hear your media-related stories and challenges you're facing in today’s multimedia Internet.

We share links, videos, text or images without a pattern on various social networks. If there is a Tennis grand slam and you are a fan you enter in to the buzz, then there are elections and you become a concerned citizen.

There is a marked change in behaviour across time and someone who just falls on your Facebook timeline or Tweets may not appreciate the temporal shift from being a techie to Tennis fan to an aware citizen.

Its important to highlight yourself aggregated these social networks and keep track of our trends across time as it draws more focus on you and your interests. For instance, if you want to identify if there is burst of posts of some topic by an individual or he/she goes uniformly across weeks.

Its about giving credibility backed by your behaviour.

Actwitty creates a timeline of trends, the way you have distributed your shares across topics.

Above screen from Actwitty shows how a sample user was talking a lot about technology, business, non profits around last week.

Actwitty offers users access to data under each topic on timeline. Clicking each topic opens the data on the Feeds section.

We mention lots of important stuff whenever we share a link, video, image or just a simple text. Lets consider following examples:

Christmas celebrations were great at University

I think Messi is best footballer of all times.

We must really put serious thought to move to green energy

I have been to travel to India, Malaysia, Hongkong and Macau

This pretty much tells a lot about ones life, interests and behaviour.

Such data must just not be kept as a part of timeline, it must be highlighted for enriching the dynamic profile of a user.

A profile which can highlight these mentions and organise them under topics.

Actwitty pulls the data from social networking services, semantic engine of Actwitty categorises the data into topics. Actwitty steps ahead and finds key terms or mentions in every data element and associate them with mentions.

Above sample profile hase some mentions like McDonalds, Angry Birds, Steve Balmer, Facebook and Twitter.

"What we see is what we believe" - Actwitty offers mining of data under each mention. Clicking any mention opens the data under that mention on the Feeds section.

Online public profiles of people are looked for knowing them better and making interactions more worthy in real world. Its very important to create that right impression. Problem with existing online profiles is that they are scattered and cater to specific audience.

There are few steps needed to gain maximum potential from your efforts on Internet. Few things which are needed in social media

Aggregate all your public profiles.

Dynamic Profiles beat Static Profiles.

Reputation "Always" backed by data.

Aggregate all your public profiles You may be an enthusiast creating technology and business vibes by micro-blogging on services like Twitter and Tumblr. At the same time you may have aspects of being a Hollywood fan from Facebook. Its important to integrate profiles.

Dynamic Profiles beat Static Profiles We create public profiles and fill in our details but in majority cases such public profiles go stale. They reflect education, work, interests which are manually added. There is a lot more which can be extracted from data we share. Like when I say "Roger Fedrer is one of the finest Tennis player" definitely means I am a Roger Fedrer fan. This must be associated with me.

Reputation "Always" backed by data We need to highlight what are our interest/expertise domain. And this is not just by showing some influence score number, everything has to be backed by your data. If you are seen as a techie, what in your data from Twitter or Facebook give you that reputation.

Actwitty as its USP offers all 3 with a zero effort from user and quick navigation to data.

Above profile highlights how an individual has spread his Internet engagement across 2 services Facebook and Twitter. Technology and Entertainment are his dominant topics.

Actwitty allows users to activate multiple services to be aggregated.

Actwitty at its core pulls user data from social network sites, its semantic engine processes the data for various things like

Topic of discussion (Technology, Entertainment, Health, Business etc.)

Key terms mentioned in the shared data

Data shared has video, image, blog link etc.

Extracts topical distribution across services.

Actwitty then creates a user impression as shown above which highlights the way user shares his data on his social networking platform.

Above profile shows how we can see number of links, videos, key term mentions are made by the user.A click on videos, links, images shows the data on the Feeds section.

Actwitty has spiralled through various ideas with one primary focus being organise user's impression for others to see. We have chosen the services judiciously which gives the maximum benefit to the users.

Facebook, Twitter are social network, micro-blogging platforms respectively. A lot of data is originated on these two services and lots of traffic is driven to blogs via a link exchange on Facebook or Twitter. This means majority of bloggers do share their links on Facebook and Twitter.

A huge user base to target, amount of data which can be tapped on twitter and Facebook, a variety of user base and last but not the least the denseness of network by gaining access to friends and followers through viral propagation.