Machine Learning

Issue emergence and social license estimation through sentiment analysis

One of the current challenges faced by SLO practitioners is the cost and inefficiency associated with collecting data on stakeholder attitudes and relationships, both of which are relevant to the social responsibilities of resource development companies. Through the use of social media and online publication mining, it is possible to estimate the overall level of social license by machine-learning aided sentiment analysis. This has the benefit of uncovering stakeholder opinions in real time, so that strategies may be implemented to manage expectations in real-time. In addition to sentiment analysis, I am currently using an unsupervised machine learning method called latent Dirichlet allocation (LDA) to identify emerging issues from textual and other media data in order to help consolidate stakeholder opinions and understand the mechanisms by which these opinions change (see the socio-political churn model below).

Social network recognition and characterization

Several types of stakeholder networks with different qualitative characteristics have been observed by researchers in the field of social license to operate, but attempts to classify these networks in a categorization framework has so far been problematic. I am interested in training neural networks to recognize the characteristics of observed social networks and aid in their classification. This technique can also be used to identify fracture patterns in rocks and classify them by the mechanisms of fracturing. Hydration-dehydration reactions, for example, yield fractures in rock matrices with specific characteristics. Using already classified samples, it is possible to train neural networks to classify unknown or ambiguous samples and aid petrologists in their understanding of formation conditions of the rock.

Rock fracture matrix with overlayed network map, used for automating characterization of rock fracture mechanism (hydration vs. dehydration reactions)

Agent-based modeling

Social license to operate (SLO)

Social license to operate (or lack thereof) is an emergent social phenomenon that exists when an equilibrium state in opinion is reached. It is relevant to fields from mining and oil and gas to renewable energy and community planning. As a social experience, it is dynamic and complex, but current research in this area is limited to static measurement and prescription, with little or no a priori knowledge of the types and probabilities of outcomes. Agent-based models based in sound decision theory have the potential to be a considerable tool for managers, CSR/ESG practitioners, and local community leaders alike.

Random, scale-free and small-world networks with internal and external opinion leaders (IOL and EOL, respectively). As external opinion leaders add connections to each network, the time to reach consensus and the consensus outcomes go through phases of 1) IOL dominance, 2) conflict, and 3) EOL dominance. The time required for transitions between states depends on the network structure.

Socio-political churn

Public opinion on any given topic (including the social license) is dependent upon the inherent and perceived relationships between issues, and the probability that issues will be discussed given the attitudes and relationships of any given set of people. Robert Boutilier describes this phenomenon as the “socio-political churn,” in which dominant topics emerge from the public discourse and opinion leaders emerge from the social network, as the networks of people and ideas interact. This is quite a complex model, as it relies on knowledge of both the social network (real or online), and knowledge of the network of topic co-mentions in a given discussion sphere. Using both survey and data-mining methods, it is possible to map social networks, and using the LDA topic model mentioned above, it is possible to model this framework in order to understand how these two types of network affect each other

Stakeholders interacting (left) and mentioning a topic (right) with a given probability. New topics are formed when their mentions reach some minimum threshold. Old topics wane and disappear if they are not mentioned frequently enough. Topics are connected by how often they’re mentioned together with other topics.

Transmission of renewable energy from resource centers to population centers

One of the major problems with renewable energy technologies is that they must be located in areas where the energy resources exist. Unlike hydrocarbons that can be transported from remote mines and reservoirs to cities by trucks, trains, and pipelines, renewable technologies such as solar, wind, hydroelectric and geothermal energy must generate electricity on-site, and the most viable current transmission method is through high-powered electric lines. The US is currently funding research into a national “super-grid” that would connect the renewable energy rich areas of the South-West with the population dense areas of the North-East. Agent-based models can explore how cities and individuals make decisions about their energy usage, and the implications that those decisions may ultimately have on the characteristics of a national energy grid.

Discrete Element Modeling (DEM)

Rock Mechanics

Continuum numerical methods such as the finite or boundary element methods have made a large impact in the area of mechanics of materials. One area that they have difficulty, however, is in simulating heterogeneous materials such as rocks, which is best done using a method capable of modeling the discontinuities found in rock matrices. DEM is useful in simulating rock mechanics applications from tunneling to drilling. In geothermal energy production, drilling may constitute more than 50% of the initial cost. This drilling often takes place in extreme environments of heat and pressure, and DEM is one of the most promising avenues for increasing the efficiency and ultimately decreasing the costs associated with geothermal drilling.