Dissertation Defense

Deep Representation of Lyrical Style and Semantics for Music Recommendation

Abhay L. Kashyap

11:00-1:00 Thursday, 20 July 2017, ITE 346

In the age of music streaming, the need for effective recommendations is important for music discovery and a personalized user experience. Collaborative filtering based recommenders suffer from popularity bias and cold-start which is commonly mitigated by content features. For music, research in content based methods have mainly been focused in the acoustic domain while lyrical content has received little attention. Lyrics contain information about a song’s topic and sentiment that cannot be easily extracted from the audio. This is especially important for lyrics-centric genres like Rap, which was the most streamed genre in 2016. The goal of this dissertation is to explore and evaluate different lyrical content features that could be useful for content, context and emotion based models for music recommendation systems.

With Rap as the primary use case, this dissertation focuses on featurizing two main aspects of lyrics; its artistic style of composition and its semantic content. For lyrical style, a suite of high level rhyme density features are extracted in addition to literary features like the use of figurative language, profanity and vocabulary strength. In contrast to these engineered features, Convolutional Neural Networks (CNN) are used to automatically learn rhyme patterns and other relevant features. For semantics, lyrics are represented using both traditional IR techniques and the more recent neural embedding methods.

These lyrical features are evaluated for artist identification and compared with artist and song similarity measures from a real-world collaborative filtering based recommendation system from Last.fm. It is shown that both rhyme and literary features serve as strong indicators to characterize artists with feature learning methods like CNNs achieving comparable results. For artist and song similarity, a strong relationship was observed between these features and the way users consume music while neural embedding methods significantly outperformed LSA. Finally, this work is accompanied by a web-application, Rapalytics.com, that is dedicated to visualizing all these lyrical features and has been featured on a number of media outlets, most notably, Vox, attn: and Metro.

Committee: Drs. Tim Finin (chair), Anupam Joshi, Tim Oates, Cynthia Matuszek and Pranam Kolari (Walmart Labs)

Analysis of Irregular Event Sequences using Deep Learning, Reinforcement Learning, and Visualization

Filip Dabek

11:00-1:00 Thursday 13 July 2017, ITE 346, UMBC

History is nothing but a catalogued series of events organized into data. Amazon, the largest online retailer in the world, processes over 2,000 orders per minute. Orders come from customers on a recurring basis through subscriptions or as one-off spontaneous purchases, resulting in each customer exhibiting their own behavioral pattern when it comes to the way in which they place orders throughout the year. For a company such as Amazon, that generates over $130 billion of revenue each year, understanding and uncovering the hidden patterns and trends within this data is paramount in improving the efficiency of their infrastructure ranging from the management of the inventory within their warehouses, distribution of their labor force, and preparation of their online systems for the load of users. With the ever increasingly availability of big data, problems such as these are no longer limited to large corporations but are experienced across a wide range of domains and faced by analysts and researchers each and every day.

While many event analysis and time series tools have been developed for the purpose of analyzing such datasets, most approaches tend to target clean and evenly spaced data. When faced with noisy or irregular data, it has been recommended to undergo a pre-processing step of converting and transforming the data into being regular. This transformation technique arguably interferes on a fundamental level as to how the data is represented, and may irrevocably bias the way in which results are obtained. Therefore, operating on raw data, in its noisy natural form, is necessary to ensure that the insights gathered through analysis are accurate and valid.

In this dissertation novel approaches are presented for analyzing irregular event sequences using a variety of techniques ranging from deep learning, reinforcement learning, and visualization. We show how common tasks in event analysis can be performed directly on an irregular event dataset without requiring a transformation that alters the natural representation of the process that the data was captured from. The three tasks that we showcase include: (i) summarization of large event datasets, (ii) modeling the processes that create events, and (iii) predicting future events that will occur.

Committee: Drs. Tim Oates (Chair), Jesus Caban, Penny Rheingans, Jian Chen, Tim Finin

Jindřich Mynarz recently posted a good list of “What I would like to see in SPARQL 1.2” and I thought I’d add a few comments as well as some of my own wished-for features.

Explicit ordering in GROUP_CONCAT, and quads support for both the HTTP Graph Store Protocol and CONSTRUCT queries (items 2, 5, and 8 in Jindřich’s list) seem like obvious improvements to SPARQL with a clear path forward for semantics and implementation.

Here are the some of the other wished-for features:

• Explicitly specify the REDUCED modifier (#1)

  As an implementor, I quite like the fact that REDUCED is “underspecified.” It allows optimization opportunities that are much cheaper than a full DISTINCT would be, while still reducing result cardinality. I think it’s unfortunate that REDUCED hasn’t seen much use over the years, but I’m not sure what a better-specified REDUCED operator would do different from DISTINCT.

• Property path quantifiers (#3)

  The challenge of supporting path quantifiers like elt{n,m} is figuring out what the result cardinality should be. The syntax for this was standardized during the development of SPARQL 1.1, but we couldn’t find consensus on whether elt{n,m} should act like a translation to an equivalent BGP/UNION pattern or like the arbitrary length paths (which do not introduce duplicate results). For small values of n and m, the translation approach seems natural, but as they grow, it’s not obvious that use cases would only want the translation semantics and not the non-duplicating semantics.

  Perhaps a new syntax could be developed which would allow the query author to indicate the desired cardinality semantics.

• Date time/duration arithmetic functions (#6)

  This seems like a good idea, and very useful to some users, though it would substantially increase the size and number of the built-in functions and operators.

• Support for non-scalar-producing aggregates (#9)

  I’m interested to see how this plays out as a SPARQL extension in systems like Stardog. It likely has a lot of interesting uses, but I worry that it would greatly complicate the query and data models, leading to calls to extend the semantics of RDF, and add new query forms, operators, and functions.

• Structured serialization format for SPARQL queries (#10)

  I’m indifferent to this. I suspect some people would benefit from such a format, but I don’t think I’ve ever had need for one (where I couldn’t just parse a query myself and use the resulting AST) and it would be another format to support for implementors.

Beyond that, here are some other things I’d like to see worked on (either standardization, or cross-implementation support):

• Support for window functions

• Explicit support for named graphs in SERVICE blocks

  This can be partially accomplished right now for hard-coded graphs by using an endpoint url with the default-graph-uri query parameter, but I’d like more general support that could work dynamically with the active graph when the SERVICE block is evaluated.

• Structured errors for use in the SPARQL Protocol

  My preference for this would be using the RFC7807 “Problem Details” JSON format, with a curated list of IRIs and associated metadata representing common error types (syntax errors, query-to-complex or too-many-requests refusals, etc.). There’s a lot of potential for smarter clients if errors contain structured data (e.g. SPARQL editors can highlight/fix syntax issues; clients could choose alternate data sources such as triple pattern fragments when the endpoint is overwhelmed).

At the AKSW Colloquium on Friday 7th of July, at 10:40 AM there will be two paper presentations concerning genetic algorithms to learn linkage rules, and differentiable learning of logical rules for knowledge base reasoning.

Tommaso Soru will present the paper Differentiable Learning of Logical Rules for Knowledge Base Reasoning, currently a pre-print, by Fan Yang, Zhilin Yang, and William W. Cohen.

Abstract

“We study the problem of learning probabilistic first-order logical rules for knowledge base reasoning. This learning problem is difficult because it requires learning the parameters in a continuous space as well as the structure in a discrete space. We propose a framework, Neural Logic Programming, that combines the parameter and structure learning of first-order logical rules in an end-to-end differentiable model. This approach is inspired by a recently-developed differentiable logic called TensorLog, where inference tasks can be compiled into sequences of differentiable operations. We design a neural controller system that learns to compose these operations. Empirically, our method obtains state-of-the-art results on multiple knowledge base benchmark datasets, including Freebase and WikiMovies.”

Daniel Obraczka will present the paper Learning Expressive Linkage Rules using Genetic Programming of Isele and Bizer accepted at VLDB 2012. This work presents an algorithm to learn record linkage rules utilizing genetic programming.

Abstract

“A central problem in data integration and data cleansing is to find entities in different data sources that describe the same real-world object. Many existing methods for identifying such entities rely on explicit linkage rules which specify the conditions that entities must fulfill in order to be considered to describe the same real-world object. In this paper, we present the GenLink algorithm for learning expressive linkage rules from a set of existing reference links using genetic programming. The algorithm is capable of generating linkage rules which select discriminative properties for comparison, apply chains of data transformations to normalize property values, choose appropriate distance measures and thresholds and combine the results of multiple comparisons using non-linear aggregation functions. Our experiments show that the GenLink algorithm outperforms the state-of-the-art genetic programming approach to learning linkage rules recently presented by Carvalho et. al. and is capable of learning linkage rules which achieve a similar accuracy as human written rules for the same problem.”

About the AKSW Colloquium

This event is part of a series of events about Semantic Web technology. Please see http://wiki.aksw.org/public/colloquium for further information about previous and future events. As always, Bachelor and Master students are able to get points for attendance and there is complimentary coffee and cake after the session.

Summary

This quarter kicked off with a full program at www 2017 in Perth and continued with ESWC which included an award winning paper on Linked Data Notifications.

Solid gain some traction this quarter going from 500 to 1500 stars on github, after a few articles publicized the protocol.  Linked data continues to evolve in search with with google now including fact check data powered by schema.org.

A new proposal was also introduced called Linked Data Templates, which aims “to provide means to define read-write Linked Data APIs declaratively using SPARQL and specify a uniform interaction protocol for them”.

Communications and Outreach

Following the announcement of Linked Data Templates, there was an invitation to join the declarative apps community group, which some of us have.  Please feel free to join this group and get involved with the evolution of read-write standards.

Community Group

This quarter saw the announcement and call for participation of Linked Data Templates, which interestingly is a protocol, neutral to transport method (HTTP, IPFS etc), however, the specification shall provide bindings for the HTTP protocol.

There was also an excellent blog post from Kingsley show casing a number of the technologies we’ve been looking at over the years, and demonstrating the openlink smart data bot.

Applications

Relatively quiet quarter for apps with much work being done on authentication.  There was also the release of twinql, “A graph query language for the semantic web”.  Work on updating libraries, rdflib, and servers, node solid sever have been bumped to the latest versions.  There was work done on the solid connections ui, an application to manage the solid social graph.

And also the release of the Openlink Smart Data Bot, “Smart Agent that provides a single RESTful interface for interacting with a variety of Actions (operations) provided by APIs”.

Last but not Least…

Congrats to timbl on winning the Turing award, often referred to as the Nobel Prize for Computing, “for inventing the World Wide Web, the first web browser, and the fundamental protocols and algorithms allowing the Web to scale”.  An invention that continues to change all our lives, and hopefully one we can all help to bring to its full potential!

2017-06-30, Registration for DC-2017 is now open at http://dcevents.dublincore.org/IntConf/index/pages/view/reg17. The International Conference takes place on Thursday through Saturday, 26-28 October and the DCMI Annual Meeting occurs on Sunday, 29 October. The full conference rate includes two days of presentations, papers, project reports, posters and special sessions and the third day of workshops. A day rate is available for all three conference days. DC-2017 in Washington DC is collocated in the same venue with the ASIST Annual Meeting that takes place from 27 October through 1 November. Special rates for the ASIST meeting are available to DCMI members. For more information and to register, visit the DC-2017 conference website at http://dcevents.dublincore.org/index.php/IntConf/dc-2017/schedConf/.

Ph.D. Dissertation Defense

Dynamic Data Assimilation for Topic Modeling

Jennifer Sleeman
9:00am Thursday, 29 June 2017, ITE 325b, UMBC

Understanding how a particular discipline such as climate science evolves over time has received renewed interest. By understanding this evolution, predicting the future direction of that discipline becomes more achievable. Dynamic Topic Modeling (DTM) has been applied to a number of disciplines to model topic evolution as a means to learn how a particular scientific discipline and its underlying concepts are changing. Understanding how a discipline evolves, and its internal and external influences, can be complicated by how the information retrieved over time is integrated. There are different techniques used to integrate sources of information, however, less research has been dedicated to understanding how to integrate these sources over time. The method of data assimilation is commonly used in a number of scientific disciplines to both understand and make predictions of various phenomena, using numerical models and assimilated observational data over time.

In this dissertation, I introduce a novel algorithm for scientific data assimilation, called Dynamic Data Assimilation for Topic Modeling (DDATM), which uses a new cross-domain divergence method (CDDM) and DTM. By using DDATM, observational data in the form of full-text research papers can be assimilated over time starting from an initial model. DDATM can be used as a way to integrate data from multiple sources and, due to its robustness, can exploit the assimilating observational information to better tolerate missing model information. When compared with a DTM model, the assimilated model is shown to have better performance using standard topic modeling measures, including perplexity and topic coherence. The DDATM method is suitable for prediction and results in higher likelihood for subsequent documents. DDATM is able to overcome missing information during the assimilation process when compared with a DTM model. CDDM generalizes as a method that can also bring together multiple disciplines into one cohesive model enabling the identification of related concepts and documents across disciplines and time periods. Finally, grounding the topic modeling process with an ontology improves the quality of the topics and enables a more granular understanding of concept relatedness and cross-domain influence.

The results of this dissertation are demonstrated and evaluated by applying DDATM to 30 years of reports from the Intergovernmental Panel on Climate Change (IPCC) along with more than 150,000 documents that they cite to show the evolution of the physical basis of climate change.

Committee Members: Drs. Tim Finin (co-advisor), Milton Halem (co-advisor), Anupam Joshi, Tim Oates, Cynthia Matuszek, Mark Cane, Rafael Alonso

The following is a written version of the lunchtime lecture I gave today at the Open Data Institute. I’ll put in a link to the video when it comes online. It’s not a transcript, I’m just writing down what I had planned to say.

2017-06-19, This webinar, presented by Dave Clarke, co-founder and CEO of the Synaptica® group of companies, will demonstrate how to design and build rich end-user search and discovery applications using Linked Data. The Linked Open Data cloud is a rapidly growing collection of publicly accessible resources, which can be adopted and reused to enrich both internal enterprise projects and public-facing information systems. The webinar will use the Linked Canvas application as its primary use-case. Linked Canvas is an application designed by Synaptica for the cultural heritage community. It enables high-resolution images of artworks and artifacts to be catalogued and subject indexed using Linked Data. The talk will demonstrate how property fields and relational predicates can be adopted from open data ontologies and metadata schemes, such as DCMI, SKOS, IIIF and the Web Annotation Model. Selections of properties and predicates can then be recombined to create Knowledge Organization Systems (KOS) customized for business applications. The demonstration will also illustrate how very-large-scale subject taxonomies and name authority files, such as the Library of Congress Name Authority File, DBpedia, and the Getty Linked Open Data Vocabularies collection, can be used for content enrichment and indexing.

To register and for more information about the webinar and presenter, visit http://dublincore.org/resources/training/#2017clarke.

Uber, whatever its faults, provides value to its users, both the drivers and the riders. People appreciate or even enjoy the service, even if they don’t like the corporate behavior or economic disruption.   Solutions seem to mostly include boycotts (in favor of taxis or competitors like Lyft) and legal action.  But most of those those solutions are pushing water uphill, because people actually like the service.

I have another solution: let’s rebuild the service without any major company involved. Let’s help software eat the world on behalf of the users, not the stockholders. In this post, I’ll explain a way to do it.  It’s certainly not trivial, and has some risks, but I think it’s possible and would be a good thing.

The basic idea is similar to this: riders post to social media describing the rides they want, and drivers post about the rides they are available to give.  They each look around their extended social graph for posts that line up with what they want, and also check for reasons to trust or distrust each other. That’s about it. You could do this today on Twitter, but it would take some decent software to make it pleasant and reliable, to make the user experience as good as with Uber.  To be clear: I’m aiming for a user experience similar to the Uber app; I’m proposing using social media as an underlying layer, not a UI.

What’s deeply different in this model is the provider of the software does not control the market. If I build this today and get millions of users, someone else can come along tomorrow with a slightly better interface or nicer ads, and the customers can move easily, even in the middle of a ride.  In particular, the upstart doesn’t need to convince all the riders and driver to switch in order to bootstrap their system!  The market, with all its relationships and data are outside the ridesharing system. As a user, you wouldn’t even know what software the other riders and drivers are be using, unless they choose to tell you.

With this approach, open source solutions would also be viable.  Then the competition could arise quite literally tomorrow, as someone just forks a product and makes a few small changes.

This is no fun for big money investors looking for their unicorn exit, but it’s great for end users.  They get non-stop innovation, and serious competition for their business.

There are many details, below, including some open issues.  The details span areas of expertise, so I’m sure I’ve gotten parts incomplete or wrong. If this vision appeals to you, please help fill it in, in comments or posts of your own.

Critical Mass

Perhaps the hardest problem with establishing any kind of multi-sided market is getting a critical mass of buyers and sellers.  Why would a rider use the system, if there are not yet any drivers?  Why would drivers bother using the software when there are no riders? Each time someone tries the system, they find no one else there and they go away unhappy.

In this case, however, I think we have some options.   For example, existing drivers, including taxi operators, could start to use the software while they’re doing the driving they already do, with minimum additional effort.  Reasons to do it, in addition to optimistically wanting to help bootstrap this: it could help them keep track of their work, and it could establish a track record for when riders start to show up.

Similarly, riders could start to use it in areas without drivers if they understand they’re priming the pump, helping establish demand, and perhaps there were some fun or useful self-tracking features.

Various communities and businesses, not in the ridesharing business, might benefit from promoting this system: companies who have a lot of employees commuting, large events where parking is a bottleneck, towns with traffic issues, etc.  In these cases, in a niche, it’s much easier to get critical mass, which can then spread outward.

Finally, there are existing ridesharing systems that might choose to play in this open ecosystem, either because their motivation is noncommercial (eg eco carpooling) or because they see a way to share the market and still make their cut (eg taxi companies).

Privacy

In the model as I’ve described it so far, there’s no privacy. If I want a ride from San Francisco to Sebastopol, the whole world could see that. My friends might ask, uncomfortably, what I was doing in Sebastopol that day. This is a tricky problem, and there might not be a perfect solution.

In the worst case, the system ends up as only viable for the kind of trips you’re fine being public, perhaps your commute to work, or your trip to an event you’re going to post about anyway. But we can probably do better than that.  I currently see two imperfect classes of solution:

1. Trust some third party organizations, perhaps to act as information brokers, seeing all the posts from both sides and informing each when there is a match, possibly masking some details. Or perhaps they certify drivers, which gives them access to your data, with an enforceable contract they’ll use it for these purposes only.
2. Trust people to act appropriately when given the right social cues and pressure: basically, use advisory  access control, where anyone can see the data, but only after they clearly agree that they are acting as part of the ridesharing system and that they will only use the data for that purpose. There might be social or legal penalties for violating this agreement.

There might also be cryptographic solutions, perhaps as an application of homomorphic encryption, but I’m not yet aware of any results that would fully address this issue.

Personal Safety

When I was much younger, hitchhiking was common. If you wanted to go somewhere without having a car, you could stand on the side of the road and stick out your thumb. But there was some notion this might be dangerous for either party, and in some places it became illegal. (Plus there was that terrifying Rutger Hauer and C Thomas Howell movie.) There have been a few stories of assaults by Uber drivers, and the company claims to carefully vet drivers. So how could this work without a company like Uber, standing behind the drivers?

There are several approaches here, that can all work together:

1. Remote trust assessment. Each party should be able to see data on the other before agreeing to the ride.  This might include social graph connections to the other person, reviews posted about the other person (and by whom), and official certifications about the other person  (including even: the ride is from a licensed taxicab).  When legally permissible this should, I think, even include information that might be viewed as discriminatory, like

As I remember it, about ten years ago, I started this blog for one main reason. I had just watched a talk from the CTO of Second Life (remember them, when they were hot?) about his vision for how to expand by opening up the system, making it decentralized.  I thought to myself: that’s going to be really hard, but I’ve thought about it a lot; I should blog my ideas about it.

As I dug into the problem, however, I realized how many sub-problems I couldn’t really solve, yet. So I never posted.   (Soon thereafter, Cory Ondrejka left the Second Life project, moving on to run engineering at Facebook.  Not sure if that’s ironic.)

Now, what feels like several lifetimes later, I’m ready to try again.

This time the “industry darling” I want to tackle first is Uber.  Okay, it’s already become widely hated, but the valuation is still, shall we say, … considerable.

So, coming soon: how to decentralized Uber.

The AKSW and Smart Data Analytics groups are happy to announce SANSA 0.2 – the second release of the Scalable Semantic Analytics Stack. SANSA employs distributed computing for semantic technologies in order to allow scalable machine learning, inference and querying capabilities for large knowledge graphs.

• Website: http://sansa-stack.net
• GitHub: https://github.com/SANSA-Stack
• Download: http://sansa-stack.net/downloads-usage/
• ChangeLog: https://github.com/SANSA-Stack/SANSA-Stack/releases

You can find the FAQ and usage examples at http://sansa-stack.net/faq/.

The following features are currently supported by SANSA:

• Reading and writing RDF files in N-Triples format
• Reading OWL files in various standard formats
• Querying and partitioning based on Sparqlify
• RDFS/RDFS Simple/OWL-Horst forward chaining inference
• RDF graph clustering with different algorithms
• Rule mining from RDF graphs

Deployment and getting started:

• There are template projects for SBT and Maven for Apache Spark as well as for Apache Flink available to get started.
• The SANSA jar files are in Maven Central i.e. in most IDEs you can just search for “sansa” to include the dependencies in Maven projects.
• There is example code for various tasks available.
• We provide interactive notebooks for running and testing code via Docker.

We want to thank everyone who helped to create this release, in particular, the projects Big Data Europe,  HOBBIT , SAKE and Big Data Ocean.

SANSA Development Team

Hello Community! The ESWC 2017 just ended and we give a short report of the course at the conference, especially regarding the AKSW-Group.

Our members Dr. Muhammad Saleem, Dr. Mohamed Ahmed Sherif, Claus Stadler, Michael Röder, Prof. Dr. Jens Lehmann and Edgard Marx participated at the conference. They held a number of presentations, workshops and tutorials:

Michael Röder

• Presented the research paper “All That Glitters is not Gold – Rule-Based Curation of Reference Datasets for Named Entity Recognition and Entity Linking”.
• During the Challenge session, he presented the Mighty Storage Challenge (MOCHA) and this years Open Knowledge Extraction (OKE) challenge.
• At the Demo and Poster session he was happy to present the Demo of the EAGLET tool to interested participants. The tool can be used for semi-automatic checks of named entity recognition and entity linking gold standards.https://svn.aksw.org/papers/2017/ESWC_EAGELT_DEMO/public.pdf
• Had the honor to pass the certificates to the winners of the OKE and MOCHA challenges during the closing ceremony

Mohamed Ahmed Sherif

• Gave the half day tutorial: Link Discovery – Algorithms, Approaches and Benchmarks
• Presented the main research track paper „WOMBAT – A Generalization Approach for Automatic Link Discovery“

Muhammad Saleem

• Organized the Querying the Web of Data (QuWeDa 2017) workshop
• Presented the paper „Federated Query Formulation and Processing through BioFed“

Edgard Marx

• Presented a Workshop paper „Exploring the Evolution and Provenance of Git Versioned RDF Data“ by Natanael Arndt, Patrick Naumann and Edgard Marx
• Presented a demo paper „Kbox – Distributing ready-to-query RDF Knowledge Graphs“ by Edgard Marx, Tommaso Soru, Ciro Baron Neto and Sandro Coelho

Claus Stadler

• Presented a Workshop paper in QuWeDa „JPA Criteria Queries over RDF Data“ by Claus Stadler, Jens Lehmann

The final versions of the papers from Edgard and Claus will be made available soon.

As every year the ESWC also awarded the best papers and studies in several categories. The award for Best Challenge Paper went to: “End-to-end Representation Learning for Question Answering with Weak Supervision” by Daniil Sorokin and Iryna Gurevych. The paper is part of the HOBBIT project by AKSW. Congrats to the winners!

Have a look at all the winners at ESWC 2017: http://2017.eswc-conferences.org/awards.

I keep doing this for lasercuts but getting a double outline instead of a single outline, and so a double cut. This is because (

Hello Community! We proudly announce that The International Conference on Web Intelligence (WI) accepted four papers by our group. The WI takes place in Leipzig between the 23th – 26th of August. The accepted papers are:

“An Evaluation of Models for Runtime Approximation in Link Discovery” by Kleanthi Georgala, Michael Hoffmann, and Axel-Cyrille Ngonga Ngomo.

Abstract: Time-efficient link discovery is of central importance to implement the vision of the Semantic Web. Some of the most rapid Link Discovery approaches rely internally on planning to execute link specifications. In newer works, linear models have been used to estimate the runtime the fastest planners. However, no other category of models has been studied for this purpose so far. In this paper, we study non-linear runtime estimation functions for runtime estimation. In particular, we study exponential and mixed models for the estimation of the runtimes of planners. To this end, we evaluate three different models for runtime on six datasets using 400 link specifications. We show that exponential and mixed models achieve better fits when trained but are only to be preferred in some cases. Our evaluation also shows that the use of better runtime approximation models has a positive impact on the overall execution of link specifications.

“CEDAL: Time-Efficient Detection of Erroneous Links in Large-Scale Link Repositories” by Andre Valdestilhas, Tommaso Soru and Axel-Cyrille Ngonga Ngomo.

Abstract: More than 500 million facts on the Linked Data Web are statements across knowledge bases. These links are of crucial importance for the Linked Data Web as they make a large number of tasks possible, including  cross-ontology, question answering and federated queries. However, a large number of these links are erroneous and can thus lead to these applications producing absurd results. We present a time-efficient and complete approach for the detection of erroneous links for properties that are transitive. To this end, we make use of the semantics of URIs on the Data Web and combine it with an efficient graph partitioning algorithm. We then apply our algorithm to the LinkLion repository and show that we can analyze 19,200,114 links in 4.6 minutes. Our results show that at least 13% of the owl:sameAs links we considered are erroneous. In addition, our analysis of the  provenance of links allows discovering agents and knowledge bases that commonly display poor linking. Our algorithm can be easily executed in parallel and on a GPU. We show that these implementations are up to two orders of magnitude faster than classical reasoners and a non-parallel implementation.

“LOG4MEX: A Library to Export Machine Learning Experiments” by Diego Esteves, Diego Moussallem, Tommaso Soru, Ciro Baron Neto, Jens Lehmann, Axel-Cyrille Ngonga Ngomo and Julio Cesar Duarte.

Abstract: A choice of the best computational solution for a particular task is increasingly reliant on experimentation. Even though experiments are often described through text, tables, and figures, their descriptions are often incomplete or confusing. Thus, researchers often have to perform lengthy web searches for reproducing and understanding the results. In order to minimize this gap, vocabularies and ontologies have been proposed for representing data mining and machine learning (ML) experiments. However, we still lack proper tools to export properly these metadata. To this end, we present an open-source library dubbed LOG4MEX which aims at supporting the scientific community to fulfill this gap.

“GENESIS – A Generic RDF Data Access Interface” by Tim Ermilov, Diego Moussallem, Ricardo Usbeck and Axel-Cyrille Ngonga Ngomo

Abstract: The availability of billions of facts represented in RDF on the Web provides novel opportunities for data discovery and access. In particular, keyword search and question answering approaches enable even lay people to access this data. However, the interpretation of the results of these systems, as well as the navigation through these results, remains challenging. In this paper, we present GENESIS, a generic RDF data access interface. GENESIS can be deployed on top of any knowledge base and search engine with minimal effort and allows for the representation of RDF data in a layperson-friendly way. This is facilitated by the modular architecture for reusable components underlying our framework. Currently, these include a generic search back-end, together with corresponding interactive user interface components based on a service for similar and related entities as well as verbalization services to bridge between RDF and natural language.

The final versions of the papers will be made available soon.

Come over to WI 2017 and enjoy the talks. More information on the program can be found here.

The University of Maryland, Baltimore County is looking to hire a Professor of the Practice to head a new graduate program in Data Science. See the job announcement for more information and apply online at Interfolio.

In addition to developing and teaching graduate data science courses, the new faculty member will serve as the Graduate Program Director of UMBC’s program leading to a master’s degree in Data Science. This cross-disciplinary program is offered to professional students through a partnership between the College of Engineering and Information Technology; the College of Arts, Humanities and Social Sciences; the College of Natural and Mathematical Sciences; the Department of Computer Science and Electrical Engineering; and UMBC’s Division of Professional Studies.

At the AKSW Colloquium, on Monday 29th of May 2017, 3 PM, Diego Moussallem will present two papers related to his topic. First paper titled “Using BabelNet to Improve OOV Coverage in SMT” of Du et al., which was presented at LREC 2016 and the second paper titled “How to Configure Statistical Machine Translation with Linked Open Data Resources” of Srivastava et al., which was presented at AsLing 2016.

This is just a quick post to gather together some pointers and links that were shared in answer to a question I asked on twitter yesterday:

Can you publish data from twitter as open data? The short answer is: No. Read on for some notes, pointers and comments.

Twitter’s developer policy places a number of restrictions on your use of their API and the data you get from it. Some of the key ones are:

• In the

I’m interested in how people share information, particularly data, on social networks. I think it’s something to which it’s worth paying attention, so we can ensure that it’s easy for people to share insights and engage in online debates.

There’s lots of discussion at the moment around fact checking and similar ways that we can improve the ability to identify reliable and unreliable information online. But there may be other ways that we can make some small improvements in order to help people identify and find sources of data.

Data forensics is a term that usually 

Ph.D. Dissertation Proposal

Modeling and Extracting information about Cybersecurity Events from Text

Taneeya Satyapanich

Tuesday, 16 May 2017, ITE 325, UMBC

People rely on the Internet to carry out much of the their daily activities such as banking, ordering food and socializing with their family and friends. The technology facilitates our lives, but also comes with many problems, including cybercrimes, stolen data and identity theft. With the large and increasing number of transaction done every day, the frequency of cybercrime events is also increasing. Since the number of security-related events is too high for manual review and monitoring, we need to train machines to be able to detect and gather data about potential cybersecurity threats. To support machines that can identify and understand threats, we need standard models to store the cybersecurity information and information extraction systems that can collect information to populate the models with data from text.

This dissertation will make two major contributions. The first is to extend our current cyber security ontologies with better models for relevant events, from atomic events like a login attempt, to an extended but related series of events that make up a campaign, to generalized events, such as an increase in denial-of-service attacks originating from a particular region of the world targeted at U.S. financial institutions. The second is the design and implementation of a event extraction system that can extract information about cybersecurity events from text and populated a knowledge graph using our cybersecurity event ontology. We will extend our previous work on event extraction that detected human activity events from news and discussion forums. A new set of features and learning algorithms will be introduced to improve the performance and adapt the system to cybersecurity domain. We believe that this dissertation will be useful for cybersecurity management in the future. It will quickly extract cybersecurity events from text and fill in the event ontology.

Committee: Drs. Tim Finin (chair), Anupam Joshi, Tim Oates and Karuna Joshi

Jennifer Sleeman, Milton Halem, Tim Finin, and Mark Cane, Modeling the Evolution of Climate Change Assessment Research Using Dynamic Topic Models and Cross-Domain Divergence Maps, AAAI Spring Symposium on AI for Social Good, AAAI Press, March, 2017.

Climate change is an important social issue and the subject of much research, both to understand the history of the Earth’s changing climate and to foresee what changes to expect in the future. Approximately every five years starting in 1990 the Intergovernmental Panel on Climate Change (IPCC) publishes a set of reports that cover the current state of climate change research, how this research will impact the world, risks, and approaches to mitigate the effects of climate change. Each report supports its findings with hundreds of thousands of citations to scientific journals and reviews by governmental policy makers. Analyzing trends in the cited documents over the past 30 years provides insights into both an evolving scientific field and the climate change phenomenon itself. Presented in this paper are results of dynamic topic modeling to model the evolution of these climate change reports and their supporting research citations over a 30 year time period. Using this technique shows how the research influences the assessment reports and how trends based on these influences can affect future assessment reports. This is done by calculating cross-domain divergences between the citation domain and the assessment report domain and by clustering documents between domains. This approach could be applied to other social problems with similar structure such as disaster recovery.

TL;DR: Some popular fact checking sites are saying that false is true and true is false in their embedded metadata 

I’m a fan of the schema.org claimReview tags for rendering fact checking results as metadata markup embedded in the html that can be easily understood by machines. Google gave a plug for this last Fall and more recently announced that it has broadened its use of the fact checking metadata tags.  It’s a great idea and could help limit the spread of false information on the Web.  But its adoption still has some problems.

Last week I checked to see if the Washington Post is using schema.org’s ClaimReview in their Fact Checker pieces. They are (that’s great!) but WaPo seems to have misunderstood the semantics of the markup by reversing the reviewRating scale, with the result that it assets the opposite of its findings.  For an example, look at this Fact Checker article reviewing claims made by HHS Secretary Tom Price on the AHCA which WaPo rates as being very false, but gives it a high reviewRating of 5 on their scale from 1 to 6.  According to the schema.org specification, this means it’s mostly true, rather than false. ??

WaPo’s Fact Check article ratings assign a checkmark for a claim they find true and from one to four ‘pinocchios‘ for claims they find to be partially (one) or totally (four) false. They also give no rating for claims they find unclear and a ‘flip-flop‘ rating for claims on which a person has been inconsistent. Their reviewRating metadata specifies a worstRating of 1 and a bestRating of 6. They apparently map a checkmark to 1 and ‘four pinocchios‘ to 5. That is, their mapping is {-1:’unclear’; 1:’check mark’, 2:’1 pinocchio’, …, 5:’4 pinocchios’, 6:’flip flop’}. It’s clear from the schema.org ClaimReview examples that that a higher rating number is better and it’s implicit that it is better for a claim to be true.  So I assume that the WaPo FactCheck should reverse its scale, with ‘flip-flop‘ getting a 1, ‘four pinocchios‘ mapped to a 2 and a checkmark assigned a 6.

WaPo is not the only fact checking site that has got this reversed. Aaron Bradley pointed out early in April that Politifact had it’s scale reversed also. I checked last week and confirmed that this was still the case, as this example shows. I sampled a number of Snope’s ClaimCheck ratings and found that all of them were -1 on a scale of -1..+1, as in this example.

It’s clear how this mistake can happen.  Many fact checking sites are motivated by identifying false facts, so have native scales that go from the mundane true statement to the brazen and outrageous completely false.  So a mistake of directly mapping this linear scale into the numeric one from low to high is not completely surprising.

While the fact checking sites that have made this mistake are run by dedicated and careful investigators, the same care has not yet been applied in implementing the semantic metadata embedded in their pages on for their sites.

Sudip Mittal, Aditi Gupta, Karuna Pande Joshi, Claudia Pearce and Anupam Joshi, A Question and Answering System for Management of Cloud Service Level Agreements,  IEEE International Conference on Cloud Computing, June 2017.

One of the key challenges faced by consumers is to efficiently manage and monitor the quality of cloud services. To manage service performance, consumers have to validate rules embedded in cloud legal contracts, such as Service Level Agreements (SLA) and Privacy Policies, that are available as text documents. Currently this analysis requires significant time and manual labor and is thus inefficient. We propose a cognitive assistant that can be used to manage cloud legal documents by automatically extracting knowledge (terms, rules, constraints) from them and reasoning over it to validate service performance. In this paper, we present this Question and Answering (Q&A) system that can be used to analyze and obtain information from the SLA documents. We have created a knowledgebase of Cloud SLAs from various providers which forms the underlying repository of our Q&A system. We utilized techniques from natural language processing and semantic web (RDF, SPARQL and Fuseki server) to build our framework. We also present sample queries on how a consumer can compute metrics such as service credit.

Dear all,

we are happy to announce the 0.2 release of SML-Bench, our Structured Machine Learning benchmark framework. SML-Bench provides full benchmarking scenarios for inductive supervised machine learning covering different knowledge representation languages like OWL and Prolog. It already comes with adapters for prominent inductive learning systems like the DL-Learner, the General Inductive Logic Programming System (GILPS), and Aleph, as well as Inductive Logic Programming ‘classics’ like Golem and Progol. The framework is easily extensible, be it in terms of new benchmarking scenarios, or support for new learning systems. SML-Bench allows to define, run and report on benchmarks combining different scenarios and learning systems giving insight into the performance characteristics of the respective inductive learning algorithms on a wide range of learning problems.

Website: http://sml-bench.aksw.org/
GitHub page: https://github.com/AKSW/SML-Bench/
Change log: https://github.com/AKSW/SML-Bench/releases/tag/0.2

In the current release we extended the options to configure learning systems in the overall benchmarking configuration, and added support for running multiple instances of a learning system, as well as the nesting of instance-specific settings and settings that apply to all instances of a learning system. Besides internal refactoring to increase the overall software quality, we also extended the reporting capabilities of the benchmark results. We added a new benchmark scenario and experimental support for the Statistical Relational Learning system TreeLiker.

We want to thank everyone who helped to create this release and appreciate any feedback.

Best regards,

Patrick Westphal, Simon Bin, Lorenz Bühmann and Jens Lehmann

I spend a lot of my professional life giving people advice. Mostly around how to publish and use open data. In order to make sure I give people the best advice I can, I try and spend a lot of time actually publishing and using open data. A mixture of research and practical work is the best way I’ve found of improving my own

At the AKSW Colloquium, on Monday 8th of May 2017, 3 PM, Lorenz Bühmann will discuss a paper titled “Type-based Semantic Optimization for Scalable RDF Graph Pattern Matching” of Kim et al. Presented at WWW 2017, this work proposes a scalable query processing approach on RDF data that relies on early and aggressive determination and pruning of query-irrelevant data. The paper describes ongoing work as part of the RAPID+ platform project.

Abstract

Scalable query processing relies on early and aggressive determination and pruning of query-irrelevant data. Besides the traditional space-pruning techniques such as indexing, type-based optimizations that exploit integrity constraints defined on the types can be used to rewrite queries into more efficient ones. However, such optimizations are only applicable in strongly-typed data and query models which make it a challenge for semi-structured models such as RDF. Consequently, developing techniques for enabling type-based query optimizations will contribute new insight to improving the scalability of RDF processing systems.

In this paper, we address the challenge of type-based query optimization for RDF graph pattern queries. The approach comprises of (i) a novel type system for RDF data induced from data and ontologies and (ii) a query optimization and evaluation framework for evaluating graph pattern queries using type-based optimizations. An implementation of this approach integrated into Apache Pig is presented and evaluated. Comprehensive experiments conducted on real-world and synthetic benchmark datasets show that our approach is up to 500X faster than existing approaches

About the AKSW Colloquium

This event is part of a series of events about Semantic Web technology. Please see http://wiki.aksw.org/public/colloquium for further information about previous and future events. As always, Bachelor and Master students are able to get points for attendance and there is complimentary coffee and cake after the session.