Selected Readings on Sentiment Analysis

Curated on Posted on by sophie

The Living Library’s Selected Readings series seeks to build a knowledge base on innovative approaches for improving the effectiveness and legitimacy of governance. This curated and annotated collection of recommended works on the topic of sentiment analysis was originally published in 2014.

Sentiment Analysis is a field of Computer Science that uses techniques from natural language processing, computational linguistics, and machine learning to predict subjective meaning from text. The term opinion mining is often used interchangeably with Sentiment Analysis, although it is technically a subfield focusing on the extraction of opinions (the umbrella under which sentiment, evaluation, appraisal, attitude, and emotion all lie).

The rise of Web 2.0 and increased information flow has led to an increase in interest towards Sentiment Analysis — especially as applied to social networks and media. Events causing large spikes in media — such as the 2012 Presidential Election Debates — are especially ripe for analysis. Such analyses raise a variety of implications for the future of crowd participation, elections, and governance.

Selected Reading List (in alphabetical order)

Annotated Selected Reading List (in alphabetical order)

Choi, Eunsol et al. “Hedge detection as a lens on framing in the GMO debates: a position paper.” Proceedings of the Workshop on Extra-Propositional Aspects of Meaning in Computational Linguistics 13 Jul. 2012: 70-79. http://bit.ly/1wweftP

  • Understanding the ways in which participants in public discussions frame their arguments is important for understanding how public opinion is formed. This paper adopts the position that it is time for more computationally-oriented research on problems involving framing. In the interests of furthering that goal, the authors propose the following question: In the controversy regarding the use of genetically-modified organisms (GMOs) in agriculture, do pro- and anti-GMO articles differ in whether they choose to adopt a more “scientific” tone?
  • Prior work on the rhetoric and sociology of science suggests that hedging may distinguish popular-science text from text written by professional scientists for their colleagues. The paper proposes a detailed approach to studying whether hedge detection can be used to understand scientific framing in the GMO debates, and provides corpora to facilitate this study. Some of the preliminary analyses suggest that hedges occur less frequently in scientific discourse than in popular text, a finding that contradicts prior assertions in the literature.

Michael, Christina, Francesca Toni, and Krysia Broda. “Sentiment analysis for debates.” (Unpublished MSc thesis). Department of Computing, Imperial College London (2013). http://bit.ly/Wi86Xv

  • This project aims to expand on existing solutions used for automatic sentiment analysis on text in order to capture support/opposition and agreement/disagreement in debates. In addition, it looks at visualizing the classification results for enhancing the ease of understanding the debates and for showing underlying trends. Finally, it evaluates proposed techniques on an existing debate system for social networking.

Murakami, Akiko, and Rudy Raymond. “Support or oppose?: classifying positions in online debates from reply activities and opinion expressions.” Proceedings of the 23rd International Conference on Computational Linguistics: Posters 23 Aug. 2010: 869-875. https://bit.ly/2Eicfnm

  • In this paper, the authors propose a method for the task of identifying the general positions of users in online debates, i.e., support or oppose the main topic of an online debate, by exploiting local information in their remarks within the debate. An online debate is a forum where each user posts an opinion on a particular topic while other users state their positions by posting their remarks within the debate. The supporting or opposing remarks are made by directly replying to the opinion, or indirectly to other remarks (to express local agreement or disagreement), which makes the task of identifying users’ general positions difficult.
  • A prior study has shown that a link-based method, which completely ignores the content of the remarks, can achieve higher accuracy for the identification task than methods based solely on the contents of the remarks. In this paper, it is shown that utilizing the textual content of the remarks into the link-based method can yield higher accuracy in the identification task.

Pang, Bo, and Lillian Lee. “Opinion mining and sentiment analysis.” Foundations and trends in information retrieval 2.1-2 (2008): 1-135. http://bit.ly/UaCBwD

  • This survey covers techniques and approaches that promise to directly enable opinion-oriented information-seeking systems. Its focus is on methods that seek to address the new challenges raised by sentiment-aware applications, as compared to those that are already present in more traditional fact-based analysis. It includes material on summarization of evaluative text and on broader issues regarding privacy, manipulation, and economic impact that the development of opinion-oriented information-access services gives rise to. To facilitate future work, a discussion of available resources, benchmark datasets, and evaluation campaigns is also provided.

Ranade, Sarvesh et al. “Online debate summarization using topic directed sentiment analysis.” Proceedings of the Second International Workshop on Issues of Sentiment Discovery and Opinion Mining 11 Aug. 2013: 7. http://bit.ly/1nbKtLn

  • Social networking sites provide users a virtual community interaction platform to share their thoughts, life experiences and opinions. Online debate forum is one such platform where people can take a stance and argue in support or opposition of debate topics. An important feature of such forums is that they are dynamic and grow rapidly. In such situations, effective opinion summarization approaches are needed so that readers need not go through the entire debate.
  • This paper aims to summarize online debates by extracting highly topic relevant and sentiment rich sentences. The proposed approach takes into account topic relevant, document relevant and sentiment based features to capture topic opinionated sentences. ROUGE (Recall-Oriented Understudy for Gisting Evaluation, which employ a set of metrics and a software package to compare automatically produced summary or translation against human-produced onces) scores are used to evaluate the system. This system significantly outperforms several baseline systems and show improvement over the state-of-the-art opinion summarization system. The results verify that topic directed sentiment features are most important to generate effective debate summaries.

Schneider, Jodi. “Automated argumentation mining to the rescue? Envisioning argumentation and decision-making support for debates in open online collaboration communities.” http://bit.ly/1mi7ztx

  • Argumentation mining, a relatively new area of discourse analysis, involves automatically identifying and structuring arguments. Following a basic introduction to argumentation, the authors describe a new possible domain for argumentation mining: debates in open online collaboration communities.
  • Based on our experience with manual annotation of arguments in debates, the authors propose argumentation mining as the basis for three kinds of support tools, for authoring more persuasive arguments, finding weaknesses in others’ arguments, and summarizing a debate’s overall conclusions.

When Technologies Combine, Amazing Innovation Happens

Curated on Posted on by Stefaan Verhulst

FastCoexist: “Innovation occurs both within fields, and in combinations of fields. It’s perhaps the latter that ends up being most groundbreaking. When people of disparate expertise, mindset and ideas work together, new possibilities pop up.
In a new report, the Institute for the Future argues that “technological change is increasingly driven by the combination and recombination of foundational elements.” So, when we think about the future, we need to consider not just fundamental advances (say, in computing, materials, bioscience) but also at the intersection of these technologies.
The report uses combination-analysis in the form of a map. IFTF selects 13 “territories”–what it calls “frontiers of innovation”–and then examines the linkages and overlaps. The result is 20 “combinational forecasts.” “These are the big stories, hot spots that will shape the landscape of technology in the coming decade,” the report explains. “Each combinatorial forecast emerges from the intersection of multiple territories.”…

Quantified Experiences

Advances in brain-imaging techniques will make bring new transparency to our thoughts and feelings. “Assigning precise measurements to feelings like pain through neurofeedback and other techniques could allow for comparison, modulation, and manipulation of these feelings,” the report says. “Direct measurement of our once-private thoughts and feelings can help us understand other people’s experience but will also present challenges regarding privacy and definition of norms.”…

Code Is The Law

The law enforcement of the future may increasingly rely on sensors and programmable devices. “Governance is shifting from reliance on individual responsibility and human policing toward a system of embedded protocols and automatic rule enforcement,” the report says. That in turn means greater power for programmers who are effectively laying down the parameters of the new relationship between government and governed….”

Privacy-Invading Technologies and Privacy by Design

Curated on Posted on by Stefaan Verhulst

New book by Demetrius Klitou: “Challenged by rapidly developing privacy-invading technologies (PITs), this book provides a convincing set of potential policy recommendations and practical solutions for safeguarding both privacy and security. It shows that benefits such as public security do not necessarily come at the expense of privacy and liberty overall.
Backed up by comprehensive study of four specific PITs – Body scanners; Public space CCTV microphones; Public space CCTV loudspeakers; and Human-implantable microchips (RFID implants/GPS implants) – the author shows how laws that regulate the design and development of PITs may more effectively protect privacy than laws that only regulate data controllers and the use of such technologies. New rules and regulations should therefore incorporate fundamental privacy principles through what is known as ‘Privacy by Design’.
The numerous sources explored by the author provide a workable overview of the positions of academia, industry, government and relevant international organizations and NGOs.

  • Explores a relatively novel approach of protecting privacy
  • Offers a convincing set of potential policy recommendations and practical solutions
  • Provides a workable overview of the positions of academia, industry, government and relevant international organizations and NGOs”

No silver bullet: De-identification still doesn’t work

Curated on Posted on by Stefaan Verhulst

Arvind Narayanan and Edward W. Felten: “Paul Ohm’s 2009 article Broken Promises of Privacy spurred a debate in legal and policy circles on the appropriate response to computer science research on re-identification techniques. In this debate, the empirical research has often been misunderstood or misrepresented. A new report by Ann Cavoukian and Daniel Castro is full of such inaccuracies, despite its claims of “setting the record straight.” In a response to this piece, Ed Felten and I point out eight of our most serious points of disagreement with Cavoukian and Castro. The thrust of our arguments is that (i) there is no evidence that de-identification works either in theory or in practice and (ii) attempts to quantify its efficacy are unscientific and promote a false sense of security by assuming unrealistic, artificially constrained models of what an adversary might do. Specifically, we argue that:

  1. There is no known effective method to anonymize location data, and no evidence that it’s meaningfully achievable.
  2. Computing re-identification probabilities based on proof-of-concept demonstrations is silly.
  3. Cavoukian and Castro ignore many realistic threats by focusing narrowly on a particular model of re-identification.
  4. Cavoukian and Castro concede that de-identification is inadequate for high-dimensional data. But nowadays most interesting datasets are high-dimensional.
  5. Penetrate-and-patch is not an option.
  6. Computer science knowledge is relevant and highly available.
  7. Cavoukian and Castro apply different standards to big data and re-identification techniques.
  8. Quantification of re-identification probabilities, which permeates Cavoukian and Castro’s arguments, is a fundamentally meaningless exercise.

Data privacy is a hard problem. Data custodians face a choice between roughly three alternatives: sticking with the old habit of de-identification and hoping for the best; turning to emerging technologies like differential privacy that involve some trade-offs in utility and convenience; and using legal agreements to limit the flow and use of sensitive data. These solutions aren’t fully satisfactory, either individually or in combination, nor is any one approach the best in all circumstances. Change is difficult. When faced with the challenge of fostering data science while preventing privacy risks, the urge to preserve the status quo is understandable. However, this is incompatible with the reality of re-identification science. If a “best of both worlds” solution exists, de-identification is certainly not that solution. Instead of looking for a silver bullet, policy makers must confront hard choices.”

Predicting crime, LAPD-style

Curated on Posted on by Stefaan Verhulst

The Guardian: “The Los Angeles Police Department, like many urban police forces today, is both heavily armed and thoroughly computerised. The Real-Time Analysis and Critical Response Division in downtown LA is its central processor. Rows of crime analysts and technologists sit before a wall covered in video screens stretching more than 10 metres wide. Multiple news broadcasts are playing simultaneously, and a real-time earthquake map is tracking the region’s seismic activity. Half-a-dozen security cameras are focused on the Hollywood sign, the city’s icon. In the centre of this video menagerie is an oversized satellite map showing some of the most recent arrests made across the city – a couple of burglaries, a few assaults, a shooting.

Advertisement

On a slightly smaller screen the division’s top official, Captain John Romero, mans the keyboard and zooms in on a comparably micro-scale section of LA. It represents just 500 feet by 500 feet. Over the past six months, this sub-block section of the city has seen three vehicle burglaries and two property burglaries – an atypical concentration. And, according to a new algorithm crunching crime numbers in LA and dozens of other cities worldwide, it’s a sign that yet more crime is likely to occur right here in this tiny pocket of the city.
The algorithm at play is performing what’s commonly referred to as predictive policing. Using years – and sometimes decades – worth of crime reports, the algorithm analyses the data to identify areas with high probabilities for certain types of crime, placing little red boxes on maps of the city that are streamed into patrol cars. “Burglars tend to be territorial, so once they find a neighbourhood where they get good stuff, they come back again and again,” Romero says. “And that assists the algorithm in placing the boxes.”
Romero likens the process to an amateur fisherman using a fish finder device to help identify where fish are in a lake. An experienced fisherman would probably know where to look simply by the fish species, time of day, and so on. “Similarly, a really good officer would be able to go out and find these boxes. This kind of makes the average guys’ ability to find the crime a little bit better.”
Predictive policing is just one tool in this new, tech-enhanced and data-fortified era of fighting and preventing crime. As the ability to collect, store and analyse data becomes cheaper and easier, law enforcement agencies all over the world are adopting techniques that harness the potential of technology to provide more and better information. But while these new tools have been welcomed by law enforcement agencies, they’re raising concerns about privacy, surveillance and how much power should be given over to computer algorithms.
P Jeffrey Brantingham is a professor of anthropology at UCLA who helped develop the predictive policing system that is now licensed to dozens of police departments under the brand name PredPol. “This is not Minority Report,” he’s quick to say, referring to the science-fiction story often associated with PredPol’s technique and proprietary algorithm. “Minority Report is about predicting who will commit a crime before they commit it. This is about predicting where and when crime is most likely to occur, not who will commit it.”…”

Privacy and Open Government

Curated on Posted on by Stefaan Verhulst

Paper by Teresa Scassa in Future Internet: “The public-oriented goals of the open government movement promise increased transparency and accountability of governments, enhanced citizen engagement and participation, improved service delivery, economic development and the stimulation of innovation. In part, these goals are to be achieved by making more and more government information public in reusable formats and under open licences. This paper identifies three broad privacy challenges raised by open government. The first is how to balance privacy with transparency and accountability in the context of “public” personal information. The second challenge flows from the disruption of traditional approaches to privacy based on a collapse of the distinctions between public and private sector actors. The third challenge is that of the potential for open government data—even if anonymized—to contribute to the big data environment in which citizens and their activities are increasingly monitored and profiled.”

Open for Business: How Open Data Can Help Achieve the G20 Growth Target

Curated on Posted on by Stefaan Verhulst

New Report commissioned by Omydiar Network on the Business Case for Open Data: “Economic analysis has confirmed the significant contribution to economic growth and productivity achievable through an open data agenda. Governments, the private sector, individuals and communities all stand to benefit from the innovation and information that will inform investment, drive the creation of new industries, and inform decision making and research. To mark a step change in the way valuable information is created and reused, the G20 should release information as open data.
In May 2014, Omidyar Network commissioned Lateral Economics to undertake economic analysis on the potential of open data to support the G20’s 2% growth target and illustrate how an open data agenda can make a significant contribution to economic growth and productivity. Combining all G20 economies, output could increase by USD 13 trillion cumulatively over the next five years. Implementation of open data policies would thus boost cumulative G20 GDP by around 1.1 percentage points (almost 55%) of the G20’s 2% growth target over five years.
Recommendations
Importantly, open data cuts across a number of this year’s G20 priorities: attracting private infrastructure investment, creating jobs and lifting participation, strengthening tax systems and fighting corruption. This memo suggests an open data thread that runs across all G20 priorities. The more data is opened, the more it can be used, reused, repurposed and built on—in combination with other data—for everyone’s benefit.
We call on G20 economies to sign up to the Open Data Charter.
The G20 should ensure that data released by G20 working groups and themes is in line with agreed open data standards. This will lead to more accountable, efficient, effective governments who are going further to expose inadequacy, fight corruption and spur innovation.
Data is a national resource and open data is a ‘win-win’ policy. It is about making more of existing resources. We know that the cost of opening data is smaller than the economic returns, which could be significant. Methods to respect privacy concerns must be taken into account. If this is done, as the public and private sector share of information grows, there will be increasing positive returns.
The G20 opportunity
This November, leaders of the G20 Member States will meet in Australia to drive forward commitments made in the St Petersburg G20 Leaders Declaration last September and to make firm progress on stimulating growth. Actions across the G20 will include increasing investment, lifting employment and participation, enhancing trade and promoting competition.
The resulting ‘Brisbane Action Plan’ will encapsulate all of these commitments with the aim of raising the level of G20 output by at least 2% above the currently projected level over the next five years. There are major opportunities for cooperative and collective action by G20 governments.
Governments should intensify the release of existing public sector data – both government and publicly funded research data. But much more can be done to promote open data than simply releasing more government data. In appropriate circumstances, governments can mandate public disclosure of private sector data (e.g. in corporate financial reporting).
Recommendations for action

  • G20 governments should adopt the principles of the Open Data Charter to encourage the building of stronger, more interconnected societies that better meet the needs of our citizens and allow innovation and prosperity to flourish.
  • G20 governments should adopt specific open data targets under each G20 theme, as illustrated below, such as releasing open data related to beneficial owners of companies, as well revenues from extractive industries
  • G20 governments should consider harmonizing licensing regimes across the G20
  • G20 governments should adopt metrics for measuring the quantity and quality of open data publication, e.g. using the Open Data Institute’s Open Data Certificates as a bottom-up mechanism for driving the adoption of common standards.

Illustrative G20 examples
Fiscal and monetary policy
Governments possess rich real time data that is not open or accessed by government macro-economic managers. G20 governments should:

  • Open up models that lie behind economic forecasts and help assess alternative policy settings;
  • Publish spending and contractual data to enable comparative shopping by government between government suppliers.

Anti corruption
Open data may directly contribute to reduced corruption by increasing the likelihood corruption will be detected. G20 governments should:

  • Release open data related to beneficial owners of companies as well as revenues from extractive industries,
  • Collaborate on harmonised technical standards that permit the tracing of international money flows – including the tracing of beneficial owners of commercial entities, and the comparison and reconciliation of transactions across borders.

Trade
Obtaining and using trade data from multiple jurisdictions is difficult. Access fees, specific licenses, and non-machine readable formats all involve large transaction costs. G20 governments should:

  • Harmonise open data policies related to trade data.
  • Use standard trade schema and formats.

Employment
Higher quality information on employment conditions would facilitate better matching of employees to organizations, producing greater job-satisfaction and improved productivity. G20 governments should:

  • Open up centralised job vacancy registers to provide new mechanisms for people to find jobs.
  • Provide open statistical information about the demand for skills in particular areas to help those supporting training and education to hone their offerings.

Energy
Open data will help reduce the cost of energy supply and improve energy efficiency. G20 governments should:

  • Provide incentives for energy companies to publish open data from consumers and suppliers to enable cost savings through optimizing energy plans.
  • Release energy performance certifications for buildings
  • Publish real-time energy consumption for government buildings.

Infrastructure
Current infrastructure asset information is fragmented and inefficient. Exposing current asset data would be a significant first step in understanding gaps and providing new insights. G20 governments should:

  • Publish open data on governments’ infrastructure assets and plans to better understand infrastructure gaps, enable greater efficiency and insights in infrastructure development and use and analyse cost/benefits.
  • Publish open infrastructure data, including contracts via Open Contracting Partnership, in a consistent and harmonised way across G20 countries…”

Big Data, My Data

Curated on Posted on by Stefaan Verhulst

Jane Sarasohn-Kahn  at iHealthBeat: “The routine operation of modern health care systems produces an abundance of electronically stored data on an ongoing basis,” Sebastian Schneeweis writes in a recent New England Journal of Medicine Perspective.
Is this abundance of data a treasure trove for improving patient care and growing knowledge about effective treatments? Is that data trove a Pandora’s black box that can be mined by obscure third parties to benefit for-profit companies without rewarding those whose data are said to be the new currency of the economy? That is, patients themselves?
In this emerging world of data analytics in health care, there’s Big Data and there’s My Data (“small data”). Who most benefits from the use of My Data may not actually be the consumer.
Big focus on Big Data. Several reports published in the first half of 2014 talk about the promise and perils of Big Data in health care. The Federal Trade Commission’s study, titled “Data Brokers: A Call for Transparency and Accountability,” analyzed the business practices of nine “data brokers,” companies that buy and sell consumers’ personal information from a broad array of sources. Data brokers sell consumers’ information to buyers looking to use those data for marketing, managing financial risk or identifying people. There are health implications in all of these activities, and the use of such data generally is not covered by HIPAA. The report discusses the example of a data segment called “Smoker in Household,” which a company selling a new air filter for the home could use to target-market to an individual who might seek such a product. On the downside, without the consumers’ knowledge, the information could be used by a financial services company to identify the consumer as a bad health insurance risk.
“Big Data and Privacy: A Technological Perspective,” a report from the President’s Office of Science and Technology Policy, considers the growth of Big Data’s role in helping inform new ways to treat diseases and presents two scenarios of the “near future” of health care. The first, on personalized medicine, recognizes that not all patients are alike or respond identically to treatments. Data collected from a large number of similar patients (such as digital images, genomic information and granular responses to clinical trials) can be mined to develop a treatment with an optimal outcome for the patients. In this case, patients may have provided their data based on the promise of anonymity but would like to be informed if a useful treatment has been found. In the second scenario, detecting symptoms via mobile devices, people wishing to detect early signs of Alzheimer’s Disease in themselves use a mobile device connecting to a personal couch in the Internet cloud that supports and records activities of daily living: say, gait when walking, notes on conversations and physical navigation instructions. For both of these scenarios, the authors ask, “Can the information about individuals’ health be sold, without additional consent, to third parties? What if this is a stated condition of use of the app? Should information go to the individual’s personal physicians with their initial consent but not a subsequent confirmation?”
The World Privacy Foundation’s report, titled “The Scoring of America: How Secret Consumer Scores Threaten Your Privacy and Your Future,” describes the growing market for developing indices on consumer behavior, identifying over a dozen health-related scores. Health scores include the Affordable Care Act Individual Health Risk Score, the FICO Medication Adherence Score, various frailty scores, personal health scores (from WebMD and OneHealth, whose default sharing setting is based on the user’s sharing setting with the RunKeeper mobile health app), Medicaid Resource Utilization Group Scores, the SF-36 survey on physical and mental health and complexity scores (such as the Aristotle score for congenital heart surgery). WPF presents a history of consumer scoring beginning with the FICO score for personal creditworthiness and recommends regulatory scrutiny on the new consumer scores for fairness, transparency and accessibility to consumers.
At the same time these three reports went to press, scores of news stories emerged discussing the Big Opportunities Big Data present. The June issue of CFO Magazine published a piece called “Big Data: Where the Money Is.” InformationWeek published “Health Care Dives Into Big Data,” Motley Fool wrote about “Big Data’s Big Future in Health Care” and WIRED called “Cloud Computing, Big Data and Health Care” the “trifecta.”
Well-timed on June 5, the Office of the National Coordinator for Health IT’s Roadmap for Interoperability was detailed in a white paper, titled “Connecting Health and Care for the Nation: A 10-Year Vision to Achieve an Interoperable Health IT Infrastructure.” The document envisions the long view for the U.S. health IT ecosystem enabling people to share and access health information, ensuring quality and safety in care delivery, managing population health, and leveraging Big Data and analytics. Notably, “Building Block #3” in this vision is ensuring privacy and security protections for health information. ONC will “support developers creating health tools for consumers to encourage responsible privacy and security practices and greater transparency about how they use personal health information.” Looking forward, ONC notes the need for “scaling trust across communities.”
Consumer trust: going, going, gone? In the stakeholder community of U.S. consumers, there is declining trust between people and the companies and government agencies with whom people deal. Only 47% of U.S. adults trust companies with whom they regularly do business to keep their personal information secure, according to a June 6 Gallup poll. Furthermore, 37% of people say this trust has decreased in the past year. Who’s most trusted to keep information secure? Banks and credit card companies come in first place, trusted by 39% of people, and health insurance companies come in second, trusted by 26% of people.
Trust is a basic requirement for health engagement. Health researchers need patients to share personal data to drive insights, knowledge and treatments back to the people who need them. PatientsLikeMe, the online social network, launched the Data for Good project to inspire people to share personal health information imploring people to “Donate your data for You. For Others. For Good.” For 10 years, patients have been sharing personal health information on the PatientsLikeMe site, which has developed trusted relationships with more than 250,000 community members…”

How to Make Government Data Sites Better

Curated on Posted on by Stefaan Verhulst

Flowing Data: “Accessing government data from the source is frustrating. If you’ve done it, or at least tried to, you know the pain that is oddly formatted files, search that doesn’t work, and annotation that tells you nothing about the data in front of you.
The most frustrating part of the process is knowing how useful the data could be if only it were shared more simply. Unfortunately, ease-of-use is rarely the case, and we spend more time formatting and inspecting the data than we do actually putting it to use. Shouldn’t it be the other way around?
It’s this painstaking process that draws so much ire. It’s hard not to complain.
Maybe the people in charged of these sites just don’t know what’s going on. Or maybe they’re so overwhelmed by suck that they don’t know where to start. Or they’re unknowingly infected by the that-is-how-we’ve-always-done-it bug.
Whatever it may be, I need to think out loud about how to improve these sites. Empty complaints don’t help.
I use the Centers for Disease Control and Prevention as the test subject, but most of the things covered should easily generalize to other government sites (and non-government ones too). And I choose CDC not because they’re the worst but because they publish a lot of data that is of immediate and direct use to the general public.
I approach this from the point of view of someone who uses government data, beyond pulling a single data point from a spreadsheet. I’m also going to put on my Captain Obvious hat, because what seems obvious to some is apparently a black box to others.
Provide a useable data format
Sometimes it feels like government data is available in every format except the one that data users want. The worst one was when I downloaded a 2gb file, and upon unzipping it, I discovered it was a EXE file.
Data in PDF format is a kick in the face for people looking for CSV files. There might be ways to get the data out from PDFs, but it’s still a pain when you have more than a handful of files….
Useable data format is the most important, and if there’s just one thing you change, make it this.
(Raw data is fine too)
It’s rare to find raw government data, so it’s like striking gold when it actually happens. I realize you run into issues with data privacy, quality, missing data, etc. For these data sources, I appreciate the estimates with standard errors. However, the less aggregated (the more raw) you can provide, the better.
CSV for that too, please.
Never mind the fancy sharing tools
Not all government data is wedged into PDF files, and some of it is accessible via export tools that let you subset and layout your data exactly how you want it. The problem is that in an effort to please everyone, you end up with a tool shown on the left….
Tell people where to get the data
Get the things above done, and your government data site is exponentially better than it was before, but let’s keep going.
The navigation process to get to a dataset is incredibly convoluted, which makes it hard to find data and difficult to return to it….
Show visual previews
I’m all for visualization integrated with the data search tools. It always sucks when I spend time formatting data only to find that it wasn’t worth my time. Census Reporter is a fine example of how this might work.
That said, visual tools plus an upgrade to the previously mentioned things is a big undertaking, especially if you’re going to do it right. So I’m perfectly fine if you skip this step to focus your resources on data that’s easier to use and download. Leave the visualizing and analysis to us.
Decide what’s important, archive the rest
So much cruft. So many old documents. Broken links. Create an archive and highlight what people come to your site for.
Wrapping up
There’s plenty more stuff to update, especially once you start to work with the details, but this should be a good place to start. It’s a lot easier to point out what you can do to improve government data sharing than it is to actually do it of course. There are so many people, policies, and oh yes, politics, that it can be hard to change.”

How Long Is Too Long? The 4th Amendment and the Mosaic Theory

Curated on Posted on by Stefaan Verhulst

Law and Liberty Blog: “Volume 8.2 of the NYU Journal of Law and Liberty has been sent to the printer and physical copies will be available soon, but the articles in the issue are already available online here. One article that has gotten a lot of attention so far is by Steven Bellovin, Renee Hutchins, Tony Jebara, and Sebastian Zimmeck titled “When Enough is Enough: Location Tracking, Mosaic Theory, and Machine Learning.” A direct link to the article is here.
The mosaic theory is a modern corollary accepted by some academics – and the D.C. Circuit Court of Appeals in Maynard v. U.S. – as a twenty-first century extension of the Fourth Amendment’s prohibition on unreasonable searches of seizures. Proponents of the mosaic theory argue that at some point enough individual data collections, compiled and analyzed together, become a Fourth Amendment search. Thirty years ago the Supreme Court upheld the use of a tracking device for three days without a warrant, however the proliferation of GPS tracking in cars and smartphones has made it significantly easier for the police to access a treasure trove of information about our location at any given time.
It is easy to see why this theory has attracted some support. Humans are creatures of habit – if our public locations are tracked for a few days, weeks, or a month, it is pretty easy for machines to learn our ways and assemble a fairly detailed report for the government about our lives. Machines could basically predict when you will leave your house for work, what route you will take, when and where you go grocery shopping, all before you even do it, once it knows your habits. A policeman could observe you moving about in public without a warrant of course, but limited manpower will always reduce the probability of continuous mass surveillance. With current technology, a handful of trained experts could easily monitor hundreds of people at a time from behind a computer screen, and gather even more information than most searches requiring a warrant. The Supreme Court indicated a willingness to consider the mosaic theory in U.S. v. Jones, but has yet to embrace it…”

The article in Law & Liberty details the need to determine at which point machine learning creates an intrusion into our reasonable expectations of privacy, and even discusses an experiment that could be run to determine how long data collection can proceed before it is an intrusion. If there is a line at which individual data collection becomes a search, we need to discover where that line is. One of the articles’ authors, Steven Bollovin, has argued that the line is probably at one week – at that point your weekday and weekend habits would be known. The nation’s leading legal expert on criminal law, Professor Orin Kerr, fired back on the Volokh Conspiracy that Bollovin’s one week argument is not in line with previous iterations of the mosaic theory.