Archive for the ‘Knowledge Graph’ Category

Holographic Embeddings of Knowledge Graphs [Are You Blinding/Gelding Raw Data?]

Monday, October 19th, 2015

Holographic Embeddings of Knowledge Graphs by Maximilian Nickel, Lorenzo Rosasco, Tomaso Poggio.


Learning embeddings of entities and relations is an efficient and versatile method to perform machine learning on relational data such as knowledge graphs. In this work, we propose holographic embeddings (HolE) to learn compositional vector space representations of entire knowledge graphs. The proposed method is related to holographic models of associative memory in that it employs circular correlation to create compositional representations. By using correlation as the compositional operator HolE can capture rich interactions but simultaneously remains efficient to compute, easy to train, and scalable to very large datasets. In extensive experiments we show that holographic embeddings are able to outperform state-of-the-art methods for link prediction in knowledge graphs and relational learning benchmark datasets.

Heavy sledding but also a good candidate for practicing How to Read a Paper.

I suggest that in part because of this comment by the authors in the conclusion:

In future work we plan to further exploit the fixed-width representations of holographic embeddings in complex scenarios, as they are especially suitable to model higher-arity relations (e.g., taughtAt(John, AI, MIT)) and facts about facts (e.g., believes(John, loves(Tom, Mary))).

Any representation where statements of “higher-arity relations” and “facts about facts” are not easily recorded and processed, is seriously impaired when it comes to capturing human knowledge.

Perhaps capturing only triples and “facts” explains the multiple failures of the U.S. intelligence community. It is working with tools that blind and geld its raw data. The rich nuances of intelligence data are lost in a grayish paste suitable for computer consumption.

A line of research worth following. Maximilian Nickel‘s homepage at MIT is a good place to start.

I first saw this in a tweet by Stefano Bertolo.

Dandlion’s New Bloom:…

Thursday, December 26th, 2013

Dandlion’s New Bloom: A Family Of Semantic Text Analysis APIs by Jennifer Zaino.

From the post:

Dandelion, the service from SpazioDati whose goal is to delivering linked and enriched data for apps, has just recently introduced a new suite of products related to semantic text analysis.

Its dataTXT family of semantic text analysis APIs includes dataTXT-NEX, a named entity recognition API that links entities in the input sentence with Wikipedia and DBpedia and, in turn, with the Linked Open Data cloud and dataTXT-SIM, an experimental semantic similarity API that computes the semantic distance between two short sentences. TXT-CL (now in beta) is a categorization service that classifies short sentences into user-defined categories, says SpazioDati.CEO Michele Barbera.

“The advantage of the dataTXT family compared to existing text analysis’ tools is that dataTXT relies neither on machine learning nor NLP techniques,” says Barbera. “Rather it relies entirely on the topology of our underlying knowledge graph to analyze the text.” Dandelion’s knowledge graph merges together several Open Community Data sources (such as DBpedia) and private data collected and curated by SpazioDati. It’s still in private beta and not yet publicly accessible, though plans are to gradually open up portions of the graph in the future via the service’s upcoming Datagem APIs, “so that developers will be able to access the same underlying structured data by linking their own content with dataTXT APIs or by directly querying the graph with the Datagem APIs; both of them will return the same resource identifiers,” Barbera says. (See the Semantic Web Blog’s initial coverage of Dandelion here, including additional discussion of its knowledge graph.)

The line, “…dataTXT relies neither on machine learning nor NLP techniques,…[r]ather it relies entirely on the topology of our underlying knowledge graph to analyze the text,” caught my eye.

In private beta now but I am interested in how well it works against data in the wild.

Interpreting the knowledge map of digital library research (1990–2010)

Tuesday, May 14th, 2013

Interpreting the knowledge map of digital library research (1990–2010) by Son Hoang Nguyen and Gobinda Chowdhury. (Nguyen, S. H. and Chowdhury, G. (2013), Interpreting the knowledge map of digital library research (1990–2010). J. Am. Soc. Inf. Sci., 64: 1235–1258. doi: 10.1002/asi.22830)


A knowledge map of digital library (DL) research shows the semantic organization of DL research topics and also the evolution of the field. The research reported in this article aims to find the core topics and subtopics of DL research in order to build a knowledge map of the DL domain. The methodology is comprised of a four-step research process, and two knowledge organization methods (classification and thesaurus building) were used. A knowledge map covering 21 core topics and 1,015 subtopics of DL research was created and provides a systematic overview of DL research during the last two decades (1990–2010). We argue that the map can work as a knowledge platform to guide, evaluate, and improve the activities of DL research, education, and practices. Moreover, it can be transformed into a DL ontology for various applications. The research methodology can be used to map any human knowledge domain; it is a novel and scientific method for producing comprehensive and systematic knowledge maps based on literary warrant.

This is a an impressive piece of work and likely to be read by librarians, particularly digital librarians.

That restricted readership is unfortunate because anyone building a knowledge (topic) map will benefit from the research methodology detailed in this article.

O Knoweldge Graph, Where Art Thou?

Monday, February 11th, 2013

O Knoweldge Graph, Where Art Thou? by Matthew Hurst.

From the post:

The web search community, in recent months and years, has heard quite a bit about the ‘knowledge graph’. The basic concept is reasonably straightforward – instead of a graph of pages, we propose a graph of knowledge where the nodes are atoms of information of some form and the links are relationships between those statements. The knowledge graph concept has become established enough for it to be used as a point of comparison between Bing and Google.


Much of what we see out there in the form of knowledge returned for searches is really isolated pockets of related information (the date and place of brith of a person, for example). The really interesting things start happening when the graphs of information become unified across type, allowing – as suggested by this example – the user to traverse from a performer to a venue to all the performers at that venue, etc. Perhaps ‘knowledge engineer’ will become a popular resume-buzz word in the near future as ‘data scientest’ has become recently.

Read Matthew’s post for the details of the comparison.

+1! to going from graphs of pages to graphs of “atoms of information.”

I am less certain about “…graphs of information become unified across type….”

What I am missing is the reason to think that “type,” unlike any other subject, will have a uniform identification.

If we solve the problem of not requiring “type” to have a uniform identification, why not apply that to other subjects as well?

Without an express or implied requirement for uniform identification, all manner of “interesting things” will be happening in knowledge graphs.

(Note the plural, knowledge graphs, not knowledge graph.)