Classifications help us find the material we are looking for.

Here is the part 1 of this article.

By now, the web has such a great amount of content that it has become impossible to apply homogeneous classification schemes to organize knowledge and make it available; unless only a  specific domain is considered (more than 2,5 million new articles are published  each and every day).

Classification schemes are structures that use  results and relations as information to be added to  content. The following four types can be identified: hierarchical, tree, faceted, according to reference models (or paradigms).

Structured information storage is ultimately aimed at improving human knowledge.

Our goal with WordLift consists in developing an application that will structure content so as to simultaneously represent various classification methods for machines, enabling the latter to organize the content that is published on digital networks so as to  make it usable in different ways.

Due to the impasse met by semantic technologies, introduced in part 1, in the first phase of our analysis we excluded the digital world as the mandatory recipient of our solution.

Therefore, during the first phase we looked to the classification systems that mankind has used to organize its knowledge before the computing era; then we considered the evolution of faceted interfaces; the technologies that put the different web environments into reference with each other; and what is the consolidated on the web regarding the considered topics (interlinking with dbpedia, freebase, geonames and methodologies required by the search engines to classify and publish content).

It’s not easy to identify the answers; especially because the essential technological component is increasingly and continually evolving. In the book “Organizzare la Conoscenza“ (in Italian)… already mentioned in the previous post, at a certain point in Chapter 2  the essential categories – those having various facets in common and valid for all disciplines – are introduced.

They are introduced by the Indian mathematician  Shiyali Ramamrita Ranganathan, who was the first – around 1930 – to talk about this analysis, consisting in breaking down a topic into components and then building it up again based on a code. He chose five essential categories: space and time, on which everyone agrees; energy, referring to activities or dynamism and indicating the ‘action’ in semantics; matter, for example of a material and its property; personas to indicate the main subject of that context, even when it’s not a human being.

These categories are considered abstract, but nevertheless we used them to design the back-end interface for the editors, and mapped them to the corresponding types in the schema.org vocabulary.

WordLift is indeed an editor built on top of the universally recognised vocabulary of concepts published by http://schema.org/, consisting so far of more than 1,200 items divided into nine essential categories: Action, CreativeWork, Event, Intangible, Medical Entity, Organization, Person, Place, Product.

In this November 2015 the schema.org vocabulary has over 217 million pages (URLs) containing a total of more than six billion triples.

WordLift 3.0 is a semantic editor that analyses content and automatically suggests metadata according to schema.org vocabulary categories that we have somewhat simplified for users, dividing them in this first experimental phase into four essential categories: Who (Person, Organization), Where (Place), When (Event), What (CreativeWork, Product, Intangible). However, users can add any amount of results to those suggested by the application, thus creating a personal vocabulary within the application.

The next release, which will complete the experimental phase in January 2016, will allow to assign different levels of importance to the results, creating a hierarchical and tree classification (by using the mainEntity that schema.org has created to mark articles).

For the future we are considering the Dewey (Dewey Decimal Classification) hierarchical classification that is used in all libraries across the world.

This is the general process that has led us to design a solution in which semantic technologies work jointly with relational technologies to automatically associate a set of metadata, or a semantic graph, to a specific content.

Identifying the technological development and services for users was not simple, but on the other hand the maturation and affirmation of the Open Data Linked cloud and of dbpedia (freebase, geonames) was essential to enable the WordLift 3.0 editor to generate reusable datasets.

[the first blog post of this brief Semantic Story is here] 

In the world of digital networks, the term knowledge is generically used to identify and justify all activities aimed at improving data collection and organization. Of all sorts.

Knowledge can be improved when information is made available for a variety of readings and reports aimed to interpret reality, fantasize on trends, evolution, a possible future, in order to somehow control or dominate it. 

Project processes have a necessary, preparatory activity in a project program, called identification of the reference scenario. In short, it consists in discovering and assimilating background contexts, or those that prepare the scene in which the subject of the study, as if it was an actor, inserts itself to explain the reasons for the first plan.

In computing knowledge is part of artificial intelligence. In this field the aim is (was) to achieve automation through strategies by making attempts and mistakes. This way of sketching a scenario is called  Knowledge Representation. This symbolic representation was limited by the difficulty to relate various scenario. The usual Tim Berners-Lee, still a WWW leader, is the one responsible for its evolution. Through the W3C he launched in 1996 the XML standard allowing to add semantic information to contents, so they could be related. It’s the beginning of the Semantic Web which made it possible to publish, alongside documents, information and data in a format allowing machines to automatically process them.

“Most of the information content in today’s web is designed to be read only by human beings …” (Tim Berners-Lee again) “computers cannot process the language in web pages”.

Semantic web means a web whose content is structured so that software can read it: read it, answer questions and interact with users.

Introduction freely adapted from .. and for whoever wants to know the whole story.

Having introduced the value of any operation aimed to develop what will automatically set or suggest the metadata to be attached to the content in order to make it readable by machines, one still has  to understand and define the following: what are the components of this structure or metadata?   How can the significant elements be extracted uniformly disregarding the language? Which types of ontological categorisation and which relations must be activated in a content in order for it to become part of a  semantic web for all? And especially: how can all this be done simultaneously?

And this is where the whole research and development area that revolves around the semantic technologies got stuck. We believe that this impasse was also caused by the lack of agreement among the various scientific paths necessary to achieve any kind of standardization. And also because of language and lexical differences, which are pushed towards a kind of ‘local’ multi-language system by the web itself and by the technologies that are distributed.

Considering the topic and the context of this post, we should leap from 1986, when the first markup languages were born, to 1998, when the standard XML was defined, and finally today, November 2015. We have performed this leap, at least partially, by means of  a query (described here below) on Wikidata.

The path we have followed (considering that our group lacks scientific skills distributed among all the included fields of knowledge) involves:

• accepting that semantic technologies as they had been conceived and applied could not fully meet our need to make the machines understand and order content;
• redefining the context after the cultural and economic affirmation of the open data world and the data structure of the Linked Open Data.

Therefore, remembering what was dictated by the Austrian logician, mathematician and philosopher  Gödel (also loved by the computing world), who stated:  a world cannot be understood from inside the system itself; in order to understand any of it, we have to go out and observe it from the outside; we have initially deconstructed it by enclosing  in sets all that would have necessarily been part of the final solution and then we turned to the  world that preceded the current one: the  analogical world and how it had tackled and replied to problems arising from the organization and classification of large amounts of “knowledge”.

A study/guide was very useful to us (and we therefore thank its authors): Organizzare la conoscenza: dalle biblioteche all’architettura dell’informazione per il web (Claudio Gnoli, Vittorio Marino and Luca Rosati).

The query on Wikidata to reconstruct the story of markup languages

Here below is the query you can make with a click (result were incomplete because we only entered languages whose creation date has a value in Wikidata – this value is expressed by Property:P571).

PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
PREFIX wikibase: <http://wikiba.se/ontology#>
PREFIX p: <http://www.wikidata.org/prop/>
PREFIX v: <http://www.wikidata.org/prop/statement/>
PREFIX q: <http://www.wikidata.org/prop/qualifier/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT DISTINCT ?entity ?ml ?sl WHERE {
 ?entity wdt:P31 wd:Q37045 . # ?entity is a markup language
 ?entity wdt:P571 ?sl . # ?sl is the inception date of ?entity
 ?entity rdfs:label ?ml . # ?entity name is ?ml
 FILTER(LANG(?ml) = "it") # ?ml is in Italian
 }
 ORDER by ?sl
 LIMIT 100

…. continues and here is the part 2 of this article.

When working in technology, sometimes you find yourself day-dreaming about Artificial Intelligence, Space Ships, Aliens. The trip can be intensified by a legacy of Science Fiction movies, that inspire and give motivation to work harder on the actual project. The bad part about being a science fiction junkie is that you always search for new movies worth watching. Along the years you become pickier and pickier, it’s never enough.

After re-reading The Foundations Series by Isaac Asimov, you crave for more and have the urge to find available movies with a solid background in literature. That seems to be a good filter for quality:

You want to watch all sci-fi movies inspired by books.

Before watching them of course you need to list them. There are many resources on the web to accomplish the task:

1 – imdb, rotten tomatoes: they offer detailed information about the movie, e.g. what the movie is about, what are the actors, some reviews. There are interesting user curated lists that partially satisfy your requirements, for example a list of the best sci-fi movies from 2000 to now. These websites are good resources to get started, but they don’t offer a search for the details you care about.

2 – individual blogs: you may be lucky if a search engine indexed an article that exactly replies to your questions. The web is huge and someone might have been so brave to do the research himself and so generous to put it online. Not always the case, and absolutely not reliable.

3 – Linked Data Cloud: the web of data comes as a powerful resource to query the web at atomic detail. Both dbPedia and Wikidata, the LOD versions of Wikipedia, contain thousands of movies and plenty of details for each. Since the LOD cloud is a graph database hosted by the public web, you can freely ask very specific and domain crossing questions obtaining as a result pure, diamond data. Technically this is more challenging, some would say “developer only”, but at InsideOut we are working to democratize this opportunity.

From the title of the post you may already know what option we like most, so let’s get into the “how to”.
We need to send a query to the Wikidata public SPARQL endpoint. Let’s start from a visual depiction of our query, expressing concepts (circles) and relation between them (arrows).

Let’s color in white what we want in output from the query and in blue what we already know.

– Why is it necessary to specify that m is a Movie?
Writings can inspire many things, for example a song or a political movement, but we only want Movies.

– Why it is not specified also that w is a Writing and p is a Person?
Movies come out of both books, short stories and sometimes science essays. We want our movies to be inspired by something that was written, and this relation is implied by the “author” relation. The fact that p is a person is implied from the fact that only persons write science fiction (at least until year 2015).

Let’s reframe the picture in a set of triples (subject-predicate-object statements), the kind of language a graph database can understand. We call m (movie), w (writing) and p (person) our incognitas, then we define the properties and relations they must match.

• m is a Movie
• m is based on w
• w is written by p
• p is a science fiction writer

Since the graph we are querying is the LOD cloud, the components of our triples are internet addresses and the query language we use is SPARQL. See below how we translate the triples above in actual Wikidata classes and properties. Keep in mind that movies, persons and writings are incognitas, so they are expressed with the ?x syntax. Everything after # is a comment.

?m <http://www.wikidata.org/prop/direct/P31> <http://www.wikidata.org/entity/Q11424> .
# ?m is a Movie

?m <http://www.wikidata.org/prop/direct/P144> ?w .
# ?m is based on ?w

?w <http://www.wikidata.org/prop/direct/P50> ?p .
# ?w written by ?p

?p <http://www.wikidata.org/prop/direct/P106> <http://www.wikidata.org/entity/Q18844224> .
# ?p is a science fiction writer

As you can see the triples’ components are links, and if you point your browser there you can fetch triples in which the link itself is the subject. That’s the major innovation of the semantic web in relation to any other kind of graph database: it is as huge and distributed as the web. Take a few moments to appreciate the idea and send your love to Tim Berners-Lee.

Similarly to SQL, we can express in SPARQL that we are selecting data with the SELECT…WHERE keywords. The PREFIX syntax makes our query more readable by making the URIs shorter:

PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>

SELECT ?w ?m ?p WHERE {
?m wdt:P31 wd:Q11424 . # ?m is a Movie
?m wdt:P144 ?w . # ?m is based on ?w
?w wdt:P50 ?p . # ?w written by ?p
?p wdt:P106 wd:Q18844224 . # ?p is a science fiction writer
}

If you run the query above you will get as result a set of addresses, being the URI of the movies, writings and persons we searched for. We should query directly for the name, so let’s introduce ml (a label for the movie), wl (a label for the writing) and pl (a label for the person). We also impose the label language to be in english, via the FILTER command.

PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

SELECT ?pl ?wl ?ml WHERE {
?m wdt:P31 wd:Q11424 . # ?m is a Movie
?m wdt:P144 ?w . # ?m is based on ?w
?w wdt:P50 >?p . # ?w written by ?p
?p wdt:P106 wd:Q18844224 . # ?p is a science fiction writer
?p rdfs:label ?pl . # ?p name is ?pl
?w rdfs:label ?wl . # ?w name is ?wl
?m rdfs:label ?ml . # ?m name is ?ml
FILTER(LANG(?pl) = "en") # ?pl is in english
FILTER(LANG(?wl) = "en") # ?wl is in english
FILTER(LANG(?ml) = "en") # ?ml is in english
}

Let’s run the query on the dedicated Wikidata service. You can image this process as SPARQL trying to match the pattern in our picture on all its data, and giving back as result only the values of our incognitas that can satisfy the constraints. The results are:

You got new quality movies to buy and watch, to satisfy the sci-fi addiction. Our query is just an hint of the immense power unleashed by linked data. Stay tuned to get more tutorials, and check out WordLift, the plugin we are launching to manage and produce linked data directly from WordPress.

Some fun exercises:

• EASY: get the movies inspired to writings of Isaac Asimov
• MEDIUM: get all the movies inspired by women writers
• HARD: get all music artists whose songs were featured in a TV series

We’ve finally reached an important milestone in our validation work in the MICO project…we can begin testing and integrating our toolset with the first release of the platform to evaluate the initial set of media extractors. 

This blog post is more or less a diary of our first attempts in using MICO in conjunction with our toolset that includes:

• HelixWare – the Video Hosting Platform (our online video platform that allows publishers and content providers to ingest, encode and distribute videos across multiple screens)
• WordLift – the Semantic Editor for WordPress (assisting the editors writing a blog post and organising the website’s contents using semantic fingerprints)
• Shoof – a UGC video recording application (this is an Android native app providing instant video-recording for people living in Cairo)

The workflow we’re planning to implement aims at improving content creation, content management and content delivery phases. 

The diagram describes the various steps involved in the implementation of the scenarios we will use to run the tests. At this stage the main goal is to:

• a) ingest videos in HelixWare,
• b) process these videos with MICO and
• c) add relevant metadata that will be further used by the client applications WordLift and Shoof.  

While we’re working to see MICO in action in real-world environments the tests we’ve designed aims at providing valuable feedback for the developers of each specific module in the platform.

These low-level components (called Technology Enablers or simply TE) include the extractors to analyse and annotate media files as well as modules for data querying and content recommendation. We’re planning to evaluate the TEs that are significant for our user stories and we have designed the tests around three core objectives:

1. output accuracy​­ how accurate, detailed and meaningful each single response is when compared to other available tools;
2. technical performance ​­ how much time each task requires and how scalable the solution is when we increase in volume the amount of contents being analysed;
3. usability ​evaluated both in terms of integration, ​modularity ​and usefulness. ​

As of today being, everything still extremely experimental, we’re using a dedicated MICO platform running in a protected and centralised cloud environment. This machine has been installed directly by the technology partners of the project: this makes it easier for us to test and simpler for them to keep on developing, hot-fixing and stabilising the platform.    

Let’s start

By accessing the MICO Admin UI (this is accessible from the `/mico-configuration` directory), we’ve been able to select the analysis pipeline. MICO orchestrates different extractors and combines them in pipelines. At this stage the developer shall choose one pipeline at the time.  

Upon startup we can see the status of the platform by reading the command output window; while not standardised this already provides an overview on the startup of each media extractor in the pipeline.

For installing and configuring the MICO platform you can read the end-user documentation: at this stage I would recommend you to wait until everything becomes more stable (here is a link to the MICO end-user documentation)!

After starting up the system using the platform’s REST APIs we’ve been able to successfully send the first video files and request the processing of it. This is done mainly in three steps:

1. Create a Content Item

2. Create a Content Part

3. Submit for processing

In the next blog posts we will see how to consume the data coming from MICO and how this data will be integrated in our application workflows.

In the meantime, if you’re interested in knowing more about MICO and how it could benefit your existing applications you can read:

• the combined use cases prototypes: the application scenarios where MICO will be used
• the combined use cases: the tests we’re planning to run on MICO

Stay tuned for the next blog post!