Collaboration Areas under Consideration

Collaboration Ideas

We are always looking for new value-adding opportunities to carry out with our partners – see some of the areas and topics for commercially-oriented collaboration we’ve identified below.

End-User Focused

EU-NLP-1: Contextual search and discovery across structured and unstructured data sets

In the Consumer market, most individuals will not be Data Scientists, and as such, will face challenges in interacting with and querying their data in order to extract the insights they are looking for. The application of Natural Language Processing (NLP)-based search is expected to provide a powerful tool in addressing this gap, enabling end-users to engage with knowledge discovery across their data assets in a more natural way.

As individuals begin to curate and apply structure to their own data assets, further challenges for NLP begin to emerge:
– Contextualization and personalization of language: Tokenization and discovery mechanisms should take into account context-specific information while also building on the individual’s own vocabulary. The same data item may be referred to in multiple context-dependent ways.
– Searching across both structured and unstructured data: In additional to the contextual use of a data item, the classification and structuring of the data may also change depending on context and usage scenario. NLP-based solutions must consider multiple representations of a unique data item applied in different contexts.

Types of Partners Sought
• Partners with NLP-based search solutions looking for new challenges
• Partners with experience applying NLP techniques successfully to structured (non-text) data
• Partners with experience in integration of open source NLP technologies (Lucene/Solr/OpenNLP, SparkNLP, CoreNLP, etc.)

Submit an Expression of Interest

EU-GAME-1: Gamification for Data Sharing and Incentivization

While many of the early technological solutions in the personal data economy have focused on incentivizing end-users to give up their personal data (e.g. for monetary incentives or other financial rewards), we consider this rather short-sighted: (1) the approach itself is unethical, attempting to identify a point where the individual is willing to ‘give up’ rights to their data, itself a rather dubious legal proposition; (2) this approach does nothing to build trust or foster the relationship with the end-user, limiting the amount of long-term value that can be realized from the relationship; and (3) business models based on this approach are at high risk of being inhibited by both the GDPR and future emerging legislation.

Accordingly, we are looking for alternative means of user incentivization, both in terms of getting the user to share aspects of their data, as well as in encouraging end users to engage more directly with their own data. Gamification approaches are of particular interest, and could be imagined in a range of contexts:

– Users competing against themselves: contrasting current consumption with historical data
– Users competing against their friends: contrasting consumption patterns across a close-knit group
– Reward mechanism for goal achievement: encouraging users to set incremental goals across data sets, and receiving positive reinforcement for goal achievement.

Areas where this could apply are:
– Utility consumption patterns (water, electricity, etc.)
– Carbon offsets vs. multi-modal transport (walking, driving, use of public transport in commuting, etc.)
– Personal/Business use of vehicle (empirical data supporting accounting and annual tax returns)

Types of Partners Sought
• Partners with existing technologies/solutions/platforms for gamification of data.
• Partners with novel (non-monetary) incentivization approaches, at varying stages of readiness (both lab and market tested).
• Partners with experience of gamification in both B2C and B2B applications.

Submit an Expression of Interest

Technology Focused

TECH-POLICY-1: Semantic & Situation-aware Intent-derived Privacy Policies

While consent under the previous Data Protection Directive 95/46/EC was typically of the opt-in/out-out nature, little emphasis was placed on ensuring that the end-user understood what was being consented to in detail. Many research projects during the previous Framework Programme similarly placed an emphasis on making actions across the data processing chain more transparent to the end-user, but have similarly avoided focusing on the intelligibility of the request to the end-user and the distillation of user intent in deriving dynamic and actionable privacy policies.

While the GDPR prepares to bring informed consent to the masses, these problems have now become even more prevalent, and begin to have real operational and financial risk going forward. The main gaps we have identified are:

– Gaps between the end-user’s understanding of the consent request and what organisations (especially technologists) believe they are asking;
– Difficulties in demonstrably conveying adequate information to meet the burden of having ‘informed’ the user;
– Translation of relatively fuzzy user intent into contextual and actionable policy across the data life-cycle.

It is our expectation that semantic policies will play a key role in translating between intent and allowable actions, while aspects of AI and machine learning are expected to provide an additional layer of contextualization/personalization, providing direct linkage to empirical evidence of evolving end-user intent over time. Situational awareness models may similarly be employed to measure the end-user’s evolving mental state across decision checkpoints, and to add further refinement to the policy as the end-user’s awareness evolves over time.

Types of Partners Sought

• Partners with experience in Semantic Policy-based data management.
• Partners with existing solutions for situation-aware decision-support systems and related technologies.
• Partners with experience in simplification of privacy policies (both directly and programmatically).
• Partners with experience in ontological modeling.
• Partners with experience in applying behavioural psychology to product development/UX design.

Submit an Expression of Interest

TECH-AIML-1: Personalized Privacy Recommendations based on Learned Behaviour

As with TECH-AIML-3, a big challenge for keeping individuals engaged with management of consent and privacy preferences is ensuring that the signal-to-noise ratio is kept to a reasonable level. End-users that are flooded with requests may simply accept the defaults outright in order to access a given service, as has been typified by EULAs and excessively verbose privacy policies to date. A relatively straightforward first step to addressing this problem is to build on historical data – keeping track and learning from what kind of requests the individual has generally agreed to or rejected outright in the past, such that recommendations for future requests can be rapidly derived and presented to the end-user.

An additional step would be to examine existing application permissions and to present them to the user for periodic review (e.g. through a risk-sorted dashboard), particularly in cases where the end-user has changed their behaviour, or their general attitude towards privacy can be seen as having changed.

Types of Partners Sought

• Partners with existing applications for mobile application scanning and privacy/security risk assessment.
• Partners with experience in machine-learning and predictive analytics.
• Partners with experience in (or existing solutions for) privacy policy analysis.

Submit an Expression of Interest

TECH-AIML-2: Personalized Newsfeeds - shaping end-user perception of vendor relationships

While early work in vendor relationship management systems and methodologies have highlighted the importance of user-driven monitoring and control of the individual relationship with a given vendor (e.g. through a straightforward vendor-specific rating), this has been primarily left to the end-user to evaluate on their own. Early applications for AI and machine learning, however, have emerged in the area of personalized newsfeed generation, allowing individuals to keep track of information that specifically matters to them. We see the intersection of these two technologies as providing a unique opportunity to keep individuals informed about the vendors they have existing relationships with, while laying the groundwork for suggesting when they may wish to re-evaluate the current relationship as well as any pre-existing consent records. Opportunities for crowd-sourcing and trend analysis should also be considered (e.g. when many users suddenly change their perception of a vendor, others may wish to be informed of the trend, too).

Types of Partners Sought

• Partners with existing solutions for personalized newsfeed generation.
• Partners with experience in machine-learning and NLP-based text analysis.
• Partners with experience in crowd-sourcing trend analysis.

Submit an Expression of Interest

TECH-AIML-3: Learned Consent - Towards User-directed Semi-Autonomous Consent

With the GDPR requiring a great deal of consent decisions to be made by the individual, one can quickly begin to imagine consent requests overloading the end-user, to the point that the end-user no longer has a strong incentive to closely explore each request in detail, and may simply consent outright to make the requests go away in order to access the service (similar to the situation that already exists with EULAs today).

Besides this, there are also some additional dimensions to consider:
1. How much does the user trust the vendor they have an established relationship with;
2. The nature of the consent request – and the associated personal risk contained;
3. How has the end-user responded to similar requests in the past?

In this case, we would take a look a holistic view across these dimensions to determine the overall importance of the consent request to the end-user, the way in which the end-user typically responds to a kind of request, and the extent to which the trust in the vendor relationship influences the extent of consent given or withdrawn.

AI/ML techniques are expected to assist in learning the end-user’s behaviour and in gradually shifting from a mechanism for consent “assistance” (e.g. recommendations on whether to give consent or not) towards a semi-autonomous model in which user’s preferences are respected, open to auditing/adjusting, and are ultimately able to reduce the amount of spurious consent requests (or asked-and-answered style requests) the end-user must deal with over time, ensuring that they remain more actively engaged with the high-impact/high-risk requests that really matter to them.

Aspects to be considered:
– User-directed vendor relationship management / vendor ranking
– Consent classification / risk assessment
– Fully auditable and non-repudiable consent tracking (e.g. through a distributed ledger)
– Personalized consent learning

Types of Partners Sought

• Partners with existing consent management solutions looking for new challenges.
• Partners with novel approaches to consent modeling and weighting, at various levels of readiness.

Submit an Expression of Interest