Text and Data Mining: rights holder licensing tools

Text and Data Mining: international perspectives, licensing and legal aspects was a Journals Publisher Forum seminar organised in conjunction with ALPSP, the Publishers Association and STM held last week in London. This is the last in a series of posts summarising discussions.

Sarah Faulder, Chief Executive of the Publishers Licensing Society announced they are developing PLS Clear – the PLS clearing house – a central window to handle license requests that will be a rights holder search and discovery service.

Text and data mining involves access to, and usage of, articles in bulk. Researchers need to track and contact potentially hundreds of publishers for permission to mine their text. The PLS service will connect researchers to rights owners for search and discovery.

Publishers already entrust licensing their secondary rights to PLS on a non-exclusive basis. As a result PLS has built arguably the most comprehensive database in the UK of publishers and their content (by ISBN/ISSN and, in due course, by DOI). This is a natural role for PLS and the network of Reproduction Rights Organizations all over the world.

They are testing a single discovery portal through which researchers can both find the appropriate publisher(s) and route their permissions requests to the relevant person in the publishing house. The plans are for a generic clearing house. The first application is text and data mining, but it will have wider usage over time.

Text and data mining presents a technical infrastructure problem first and foremost. Licensing is a necessary means of managing access to content where the scale of access increases risk of leakage and therefore piracy, and puts an unacceptable strain on publisher platforms not designed for systematic crawling and scraping.

Carlo Scollo Lavizzari is legal advisor to STM on copyright law, policy and legal affairs. Lavizzari outlined how structuring a license is easy. Leave rhetoric aside and look to business opportunities, it is about defining the terms, what are the sources, input of content, what does the user do with that content? Where is it stored, what is done with it, can it compete or not, etc. Consider the mechanical clause on delivery mechanisms. Should also deal with the end of project – always have an exit strategy! That is the legal skeleton of a legal license.

There are calls for cooperation between those who hold content in public domain, those who hold open access content, those who hold content that is subscribed to or purchased; those who already hold a lot of purchased contents; and researchers who might want to access/mine. The question they haven’t managed to get through with any community is how to combine open access environment and copyright protected license. It is an area where he believes that licensing can provide a solution, but still trying to tackle.

John Billington works in corporate products and services at Copyright Clearance Center, who’ve been working on their Text and Data Mining Pilot Service. They have developed a pilot service that provides licensed, compliant access and retrieval of full text XML and metadata from multiple scientific publishers for the purposes of text mining.

CCC’s role is to provide an authorized means to access and retrieve published content in a standard format. The initial pilot is focused on corporate users with an access, retrieval and licensing layer. Future markets may include corporate marketing users, or academic uses.

He reflected on how it has been challenging to extract full text from different publishers and convert to a normalized format that is usable in text mining technologies. There is a lack of federation and existing tools are still difficult. They are trying to provide a one-stop shop for users and publishers that incorporate standardization, license and business model and access method that works for both sides.

He noted that a researcher wouldn’t want to be limited to what the library is subscribed to. So the tool will show them the metadata for what they aren’t subscribed to. It will filter to help them understand what they have subscribed to or not. They intend to include a purchase mechanism for full text unsubscribed articles. You will be able to download results in normalized XML format. It currently has a web interface, but they are working on an API so they systematize it.

Ed Pentz from CrossRef closed the day by outlining their latest beta application Prospect. They work on the assumption that researchers aren’t doing search or discovery – researchers will know or will have used another tool. Their service relies on DOI Content Negotiation. They are now collecting ORCID IDs for researchers. In text and data mining it is important to have a unique ID for a researcher so you can see who is doing it. They are also including funding information.

DOI content negotiation can serve as a cross-publisher API for accessing full text for TDM purposes. To make use of hits, publishers merely need to register URIs to full-text. NISO is working on a fuller specification on some metadata. They are focusing on an interim solution to at least record URIs to well known licenses. They think it will also be possible to extend to handle embargoes.

He observed that there’s potential to coordinate across initiatives, but only once each organization has individually figured out during their own trial periods. CrossRef are testing the system over the summer and will then assess if it is workable as a production system.

Text and Data Mining: practical aspects of licensing and legal aspects

Alistair Tebbit

Text and Data Mining: international perspectives, licensing and legal aspects was a Journals Publisher Forum seminar organised in conjunction with ALPSP, the Publishers Association and STM held earlier this week in London. This is the second in a series of posts summarising discussions.

Alistair Tebbit, Government Affairs Manager at Reed Elsevier, outlined his company's view on evolving publishers’ solutions in the STM sector. Elsevier have text mining licenses with higher education and research institutions, corporates and app developers. They give access to text miners through transfer of content to user’s system enabled by one of two delivery mechanisms under a separate licence: via API or using ConSyn. The delivery mechanisms have been set up and there are running costs. Their policy to date has been to charge the value add to the services to users for commercial organisations, but not academic.

Why is content delivery managed this way? Platform stability is a critical reason. Data miners want content at scale – they generally don’t do TDM on a couple of articles - but delivering scale via their main platform ScienceDirect.com is sub-optimal. APIs or ConSyn are the solution as they leave ScienceDirect untouched. Effectively they are separating machine-to-machine traffic from traffic created by real users going to ScienceDirect.com. Content security is another key issue. Free-for-all access to miners on ScienceDirect would not allow bona fide users to be checked. XML versions are less susceptible to piracy than PDFs. Why is content delivery managed this way? It’s more efficient for genuine text miners. Most miners prefer to work off XML, not from article versions on ScienceDirect. Their delivery mechanisms put the content into data miners' hands fast.

With text and data mining outputs, they use a CC BY-NC licence when redistributing results of text mining in a research support or other non-commercial tool. They require that the DOI link back to the mined article whenever feasible when displaying extracted content. They grant permission to use snippets around extracted entities to allow context when presenting results, up to a maximum of 200 characters or one complete sentence.

Licensing is working well at Elsevier and will improve further. The demand to mine is being met and there are no extra charges in the vast majority of cases. Additional services to support mining will likely be offered as they improve. However, it’s early days. Mining demand is embryonic with low numbers at the moment. Copyright exceptions are a big cause for concern and there is a major risk of spike in unauthorized redistribution. Platform stability may be threatened and there is a risk of a chilling effect on future service innovation.

Duncan Campbell

Duncan Campbell, Associate Director for Journal Digital Licensing at Wiley-Blackwell, provided an overview of emerging solutions in text and data mining with a publisher perspective on intermediary solutions. Text and data mining is important to publishers as it enriches published content, adds value for customers and aids development of new products. For researchers it helps identify new hypotheses, discover new patterns, facts and knowledge. For corporate research and development, the benefits are as above and in addition it accelerates drug discovery and development and maximises the value of information spend.

There are a number of barriers to text and data mining:

• Access: how can users get hold of content for text mining
• Content formats: there is no standard cross-publisher format
• Evaluation: understanding user needs and use cases
• Uncertainty: what is allowed by law, what is the use of text and data mining output
• Business models: lack of business pricing models e.g. access to unsubscribed content
• Scale: define and manage demand, bilaterial licensing unlikely to be scalable.

There is a potential role for intermediary to help with publisher/end user relationship. This could include as a single point of access and delivery; by providing standard licensing terms as well as speed and ease of access. The intermediary may make mining extensible and scalable and they can cover the long tail of publishers and end-users. It also enables confidential access, especially in pharma.

Andrew Hughes

Andrew Hughes, Commercial Director at the Newspaper Licensing Agency (NLA), provided a different perspective on text and data mining. Text mining requires copying of all data to establish data patterns and connections and computers need to index data. Every word on every page has to be copied. Once the copy exists, it needs to be managed. Copying requires access to data so that indexing can only happen on either the publisher database, but there is a risk of damage and disruption unless managed, and expense; or copy provided to text minders’ database where there are costs and control risks for publishers. He believes that you also need to bear in mind that third party licence partners aren’t always as careful with your data as you are.

In the newspaper sector, press packs are produced by text mining. The NLA eClips is a service where the proprietary way of mining content is withheld and a PDF is supplied of the relevant articles. There are substantial risks for publishers in text mining including the potential for technical errors by miners, challenges around data integrity and commercial malpractice. There are also cost implications including the technical loads on systems, management of copies and uses and opportunity costs.

Hughes cited the Meltwater case where the industry had to tackle the unauthorised use of text and data mining for commercial use. It took a lot of time and litigation, but they are now thriving within the NLA rules. They are licensed by the NLA and their users are licensed. It means they are operating on fair and equal terms with competitors and is an example of how licenses can work to the benefit of all parties.