NLP Applied to Conflict Resolution

Using surveys at the end of Win-Win sessions and anonymized problem statements collected voluntarily from parties entering into mediation, we will identify relevant iid features (if possible). Subsequently, we will develop models that support communicating productively throughout conflict resolution. The use of AI here will support the greater good, in that marginalized people who use mediators.ai will benefit most from its implementation.

Tech stack:

  - De-identification and anonymization: python, PyTorch, spaCy

  - SVM (RBF) or simple DL model: python, possibly scikit-learn for the initial validation

Relevant experience

Victor Corcino - Ph.D. candidate in the field of ML applied to computational fluid dynamics. Victor was a member of the Catalyst Circle v1, and is a frequent mentor in the catalyst process. He has a specialization in Data Science, AI, and ML.

Eli Selkin - A lifelong researcher with two master's degrees, one of which specialized in machine learning. Co-founder and CTO of upful.ai practicing NLP and ML in the HR space. Eli heavily uses gRPC in almost all of his production services.

Both Victor and Eli are Gimbalabs PPBL team members and are actively working to make the Cardano ecosystem more approachable from the community.

Timeline

March 2022: Create a survey to ask practiced mediators questions about the communication and participation of parties during the session. Additional information from participants to regulate bias in models will be collected (3 weeks)

April 2022: Develop document anonymizer application on SingularityNet (3 weeks)

May 2022: Create a model (probably initially simply logistic regression or an SVM) to give (internal) scores to case descriptions, that can help us guide parties involved in mediation to write improved descriptions (and potentially future documentation). The scores will not be shown to mediators or parties but will add indicators to document content on mediators.ai. (3-4 weeks)

June 2022: More complex models to support more data as we collect more. (3-6 weeks high variability)

Future focus: ML model-based guide that helps mediation seekers decide what additional information they need to provide to support their cases.

KPIs:

  - Validation of anonymization from external sources

  -Tested model and invocation on SingularityNet (usable by anyone)

  - Validated outcomes from exit surveys of participants in conflict resolution

Metrics:

  - Initially: de-identification that can be confirmed with multiple parties.

  - Subsequently validated anonymization.

  - Creation of visual representation of abstracted score to inform the parties of the benefits their inputs will have on the outcome of their mediation or resolution.

Funding breakdown:

12 weeks - 2 developers @ $1000/wk ea =12 * 2000 = $24000