NELL: Never-Ending Language Learning

Can computers learn to read? We think so. “Read the Web” is a research project that attempts to create a computer system that learns over time to read the web. Since January 2010, a computer system called NELL (Never-Ending Language Learner) has been running continuously, attempting to perform two tasks each day:

• First, it attempts to “read,” or extract facts from text found in hundreds of millions of web pages (e.g., playsInstrument(George_Harrison, guitar)).
• Second, it attempts to improve its reading competence, so that tomorrow it can extract more facts from the web, more accurately.

At present, NELL has accumulated a knowledge base of 644,836 beliefs that it has read from various web pages. It is not perfect, but NELL is learning. You can track NELL’s progress on @cmunell on Twitter, browse and download its knowledge base, read more about our technical approach, or join the discussion group.

Source

Read Full Post »

IBM’s Watson heads to medical school

There is a new chapter in the life of IBM’s super computer, named Watson: IBM’s Watson goes to medical school. In colaboration with Dr. Herbert Chase of Columbia University, Watson is learning how to treat and diagnose ailments in the human body. In fact, Watson is doing well enough that this might be the harbinger of medical diagnostic tools to come.

That means a patient could walk into a hospital and tell the computer about what is bothering them — whether it’s leg pain or a cough or a sore throat — and Watson can quickly process that information and spit out a diagnosis that has the highest probability of being correct. For most cases, that would save hospitals a lot of time because the computer could plow through the large number of cases hospitals regularly contend with that require simple treatments.

Watson was able to diagnose an eye problem with a fictional victim that had a 73 percent chance of being correct, according to a report by the Associated Press.

Source

Note: The OpenEphyra father , Nico Schlaefer, is also working on this project.

Read Full Post »

The Importance of Knowledge Management

Most companies are focused on producing a product or service for customers. However, one of the most significant keys to value-creation comes from placing emphasis on producing knowledge. The production of knowledge needs to be a major part of the overall business strategy.

One of the biggest challenges behind knowledge management is the dissemination of knowledge. People with the highest knowledge have the potential for high levels of value creation. But this knowledge can only create value if it’s placed in the hands of those who must execute on it. Knowledge is usually difficult to access – it leaves when the knowledge professional resigns.

“The only irreplaceable capital an organization possesses is the knowledge and ability of its people. The productivity of that capital depends on how effectively people share their competence with those who can use it.” – Andrew Carnegie

Therefore, knowledge management is often about managing relationships within the organization. Collaborative tools (intranets, balanced scorecards, data warehouses, customer relations management, expert systems, etc.) are often used to establish these relationships. Some companies have developed knowledge maps, identifying what must be shared, where can we find it, what information is needed to support an activity, etc. Knowledge maps codify information so that it becomes real knowledge; i.e. from data to intelligence.

For example, AT&T’s knowledge management system provides instant access for customer service representatives, allowing them to solve a customer’s problem in a matter of minutes. Monsanto uses a network of experts to spread the knowledge around. Employees can lookup a knowledge expert from the Yellow Page Directory of knowledge experts.

In the book Value Based Knowledge Management, the authors advocate that every organization should strive to have six capabilities working together:

1. Produce : Apply the right combination of knowledge and systems so that you produce a knowledge based environment.
2. Respond : Constantly monitor and respond to the marketplace through an empowered workforce within a decentralized structure.
3. Anticipate : Become pro-active by anticipating events and issues based on this new decentralized knowledge based system.
4. Attract : Attract people who have a thirst for knowledge, people who clearly demonstrate that they love to learn and share their knowledge opening with others. These so-called knowledge professionals are one of the most significant components of your intellectual capital.
5. Create : Provide a strong learning environment for the thirsty knowledge worker. Allow everyone to learn through experiences with customers, competition, etc.
6. Last : Secure long-term commitments from knowledge professionals. These people are key drivers behind your organization. If they leave, there goes the knowledge.

Knowledge professionals will become the dominant force behind the new economy, not unlike the farmer was once the key player behind the agricultural age. By the year 2010, one-third of the workforce in the United States will be comprised of knowledge professionals. It is incumbent upon all organizations to embrace this need for managing knowledge. Just take a look at those organizations that seem to create value against the competition. You will invariably find a strong emphasis on knowledge management.

Source: exinfm

Read Full Post »

3 Knowledge Management Technology Posts

Here is an interesting view over Knowledge Management Technology, wrote by Arjun Thomas.

Main points:

• “Now while it’s a great thing that Microsoft have finally awoken to the wonder that is the wiki, an in-depth look at the functionality would give an average wiki user the shudders.”
• “If you ask any KM expert about whether there is a standard method of implementing KM chances are he/she would probably say there isn’t one. It entirely depends on what your goals are and how these processes are received by your audience.”
• “While we all agree KM is more of a cultural initiative, there is no doubting that without a solid technology backbone chances are you’re heading down a dark road.  I’ve reviewed a large number of applications that claim to provide the perfect KM solution, and guess what? they don’t.”

Read Full Post »

Swiss Army Tools: Aperture framework

Aperture is an open source library for crawling and indexing unstructured information sources. Aperture supports a number of common source types and document formats and provides easy ways to extend it with custom implementations.

Aperture consists of a number of related but independently usable components:

• Crawling of information sources: file systems, websites, mail boxes
• MIME type identification
• Full-text and metadata extraction of various file formats
• Opening of crawled resources

Aperture has a strong focus on semantics. Besides extracting as much metadata contained in the file formats as possible (e.g., titles, authors, comments, …) , this is also combined with source-specific metadata (e.g., location, last modification date, …). All metadata is mapped to properties from the NIE namespace to allow uniform processing of the crawled and extracted information. NIE is an ontology developed within the Nepomuk Project and now maintained by an open source community.

Features

• Crawl information systems such as file systems, websites, mail boxes and mail servers
• Extract full-text and metadata from many common file formats
• View files in their native applications
• Ease of use: easy to learn, easy to code, easy to deploy in industrial projects
• Flexible architecture: can be extended with custom file formats, data sources, etc., with support for deployment on OSGi platforms
• Aperture consists of many parts that can be used as plugins, so it can be extended and limited very easily.
• Data exchange based on Semantic Web standards (e.g. RDF, SPARQL, …)

Supported File Formats

• Plain text
• HTML, XHTML
• XML
• PDF (Portable Document Format)
• RTF (Rich Text Format)
• Microsoft Office: Word, Excel, Powerpoint, Visio, Publisher
• Microsoft Works
• OpenOffice 1.x: Writer, Calc, Impress, Draw
• StarOffice 6.x – 7.x+: Writer, Calc, Impress, Draw
• OpenDocument (OpenOffice 2.x, StarOffice 8.x)
• Corel WordPerfect, Quattro, Presentations
• Emails (.eml files)
• ical files
• VCARD files (.vcf)
• archives (zip,tar,gz,bz2)

Crawlers

Crawlers support the extraction of information from heterogeneous data sources. Here are the source types that currently are supported:

• File Systems (local, remote, removeable media)
• Websites and intranets
• IMAP e-mail servers
• Microsoft Outlook (alpha)
• Internet Calendar (ical) files
• mbox mailboxes
• thunderbird addressbooks
• apple addressbook

Support

At this moment the project is still in alpha stage, thus limited support is available.

License

The Aperture code is published under a permissive BSD-like open source license that allows the use of Aperture in proprietary applications.

Read Full Post »

Semantic Web

“The Semantic Web is an extension of the current web in which information is given well-defined meaning, better enabling computers and people to work in cooperation.” - Tim Berners-Lee, James Hendler, Ora Lassila, The Semantic Web, Scientific American, May 2001

The World Wide Web is the biggest repository of information ever created and according to the W3C consortium, the Web can only reach its full potential only if data can be shared, processed, and understood by automated tools as well as by people .

• “Stairway to Heaven” is a song by the English rock band Led Zeppelin.

• Led Zeppelin were an English rock band formed in 1968 by Jimmy Page.

The statements can be easily understood by people with basic knowledge of English, but they are not trivial to computers. They follow a syntax (grammatical rules for specifying correct word order and inflectional structure in a sentence), but how can convey semantic (grammatical rules for assigning meaning to a sentence)?

The core process is centered in describing properties of things (shape, colour, size), and relationships between things (x is member of y).

While XML (Extensible Markup Language) is accepted within the NLP (Natural language programming) community as the main format to describe data and relationships between data items, especially when emphasis is in simplicity, interoperability and usability, the use of Semantic Web Stack [Figure 1] technologies and tools like RDF (Resource Definition Framework), RDFS (RDF Schema), and OWL (Ontology Web Language) in NLP applications, is still limited.

Figure 1 – Semantic Web stack.

The Web Ontology Language Overview is supposed to be used when information is processed by a machine, and only then presented to humans. It represents the meaning of terms in vocabularies by describing functions and relationship and it’s known as ontology.

OWL is created to meet Web Ontology Language requirements:

• XML provides an essential syntax for content structure within documents, yet associates no semantics meaning or constrains to the document.
• XML Schema and DTD is a language. They provide a means for defining the structure, content and semantics of XML documents.
• RDF is a language for describing data models for objects(“resources”) and relations, providing basic semantics. It can be expressed in XML syntax.
• RDF Schema like XML schemas provide means of defining the structure and content of properties and classes in RDF-based resources. It has specific semantics for hierarchies of properties and classes.
• OWL extends RDFS by adding more advanced constructs to describe semantics of RDF statements, like relations between classes (e.g. disjointness), cardinality, equality, restrictions of values, richer typing of properties and characteristics of properties (e.g. symmetry and transitivity), and enumerated classes. It’s based on description logic and brings reasoning power to the semantic web.
• SPARQL is a RDF query language, and queries RDFS, OWL and any RDF-based data. Used to retrieve semantic data.

Currently RIF (Rule Interchange Format) is evolved in an ongoing process of standardizations, designed to enable interoperability among rule languages in general. The layers of “Unifying Logic” and “Proof” are undergoing active research and have not been fully implemented.

The original vision of semantic web hasn’t been achieved, but we are progressing towards the desired result. It already helps in the production of more connectible, interoperable, and adaptable software, while keeping maintenance cheap and easy.

As researched by Leo Sauermann, positive results have been obtained while using the semantic web, namely increased profit (better customer satisfaction, shareholder value, user work support), data integration (a consistent data model to build upon. integrate content from different organizations, providers and departments, disparate data sources, legacy data), better querying systems, and taxonomies(categorization, or classification, of things based on a predetermined system).

The main problem of semantic is how to avoid and manage:

• Vastness – information increases at a exponential rate. Automated reasoning system are require to manage so vast volumes of inputs.
• Vagueness – undefinition of certain concepts, generate uncertainty in queries. By using fuzzy logic, some uncertainty can be removed.
• Uncertainty – uncertain values for of variables, different results are obtained (each synthoms in a medical report, can have different probabilities attributed, which can lead to different diagnoses). The can be mitigated with the user of probabilistic models and reasoning methods.
• Inconsistency – when large ontologies are used or diverse documentation corpus, contradiction between documents or inside ontologies are bound to happen. Defensible reasoning and consistent reasoning are two techniques which can be employed to deal with inconsistency.
• Deceit – With the rise of data sources, some producers will produce information intentionally, that is misleading and damaging to consumer.

The implementation of “unifying logic” and “proof” layers is still an ongoing process.

Read Full Post »

Gold Rush

Usually X marks the spot, but the path for conversion of unstructured knowledge into a reliable and efficient knowledge database of facts isn’t straightforward. Despite the knowledge being already assembled in a machine-optimal-representation, information recovery into a English natural language answer isn’t trivial.

Nowadays, the amount of information that companies deal with is overwhelming. Being able to convert unstructured information, into a fact representation, which can be stored and manipulated through data warehouse techniques, is striking gold, as we indeed generate knowledge.

With this is mind, the key concept is the development of a knowledge management system, like a semantic network, embedded within a ontology. Setting up an ontology hierarchy will allow to change the scope of the project easily and increase the flexibility. For instance, we could  construct a general ontology for mapping the global state of the world, e.g. to map generic relations such as “a son has a mother and a father”, and within this generic ontology we could have more specific ontologies, mapping the business logic, such as e.g. “cars have 4 tires”.

After mapping the semantic network, the representation allows an  easily and efficiently creation, inference, exploration, and retrieval of knowledge. Besides, the network also allows to detect contradictions. This feature is very important, as the source documents are written in natural language, and detection of such occurrences is essential.

But, what’s the easiest and efficient way to inquire the knowledge database? Perhaps, using English Natural Language Questions, such as: “When was Eiffel tower constructed?”. These questions grant users, the ability to inquire it, requiring only a small learning curve of the system. This is a valuable feature, as we want to have a system that can be used not only by technicians, but also for a wide range of users without technical background.

Yes, but how to deal with the knowledge retrieval?

If we annotate the source unstructured text (Tokenization , Part-of-Speech tagging, Entity Recognition, Stemming), using frameworks like GATE or UIMA, we can get relevant actions from the phrase. Then, we can attach for each action, the passage in the text where the action is stated. From the actions, mapped into the semantic network, when some answer is retrieved, the passages are shown with a specific ranking associated. The ranking can be tuned, as the user says if the answer is useful or not. This introduces an interesting concept of machine learning in the knowledge management system.

Any clue to the whereabouts of the lost treasure and how to turn things into gold?

References: Wikipedia, GATE

Read Full Post »

Knowledge Management

Knowledge management is defined as the combination of practices to recognize, infer, represent, and distribute knowledge.

A typical Knowledge Management System usually refers to a system that supports gathering, discovery, cataloging and storage of information, within organizations. The idea behind a Knowledge Management System is to enable users to have ready access to the organization’s documented base of facts, sources of information and solutions. Sharing this information wide along the organization can lead to more effective managing and design processes, while lending a hand to new or improved concepts.

Some of the advantages claimed for Knowledge Management System are :

1. Sharing of valuable organizational information  horizontally and vertically throughout organizational hierarchy.
2. Collaboration in the development and sharing of good practices.
3. Knowledge is  meaningfully accumulated and stored within a context.
4. Recurring processes are not reused, reducing redundant work.
5. May reduce overall learning curve for new users.

At present, Knowledge management is a critical concern for all organizations. One of the main obstacles towards an effective information retrieval is due to the fact that almost 80% of documents in an organization are available in unstructured format. That includes all data types, including email, attachments, presentations, spreadsheets, videos, voice messages, invoices, purchase orders, resumes, work orders and technical reports. This overwhelming amount of data makes processing unstructured data for storage and posterior retrieval a major concern for most organizations, as this information overload leads to less time to share knowledge and also work duplication.

Since a big fraction of corporation knowledge sources are arbitrarily and weakly structured, current data warehouses technologies cannot be applied to exploit their textual contents. That means that 80% of business information can’t be efficiently employed in automated business processes or business intelligence.

The existing statistical text learning algorithms can be trained to approximately categorize documents. A variety of methodologies and tools were developed and used in order to perform automatic document classification. Text Mining techniques merge methods of information retrieval, language processing and Data Mining to analyze large sets of unstructured data.

Due to the continuous growth of the amount of data, automated extraction of previously unknown and potentially useful information, turns into more compulsory to utilize this methodologies.

References: Bolebruch, Murphy, wikipédia

Read Full Post »

Prymas! The ultimate Information Retrieval tool.

Read Full Post »