ISSP 2030 AI Forum Week 3

05 Feb 2021 - Arun Balajiee

Semi-supervised Knowledge Graph Cultivation for Contract Segment Classification

Speaker: Mengdi Wang

In this talk Mengdi explored the general approaches to mining information from legal contracts for process automation. In achieving this purpose, Mengdi also talked about extending the ideas beyond English language contracts to Chinese language contracts. Mengdi discussed her implementation which uses the semantic role labeling technique, specially the (Agent, Predicate, Theme) roles for each token in a sentence to construct the legal knowledge graph. This knowledge graph would serve the dual purpose of embedding the data in one unified form as well to capture all the essence into uniform input or output to train/test ML models. Comparisons with several state-of-art models such as BERT-based models or models that use other techniques show the proposed solution is better or on par with them.

The novelty of the implementation was in developing the knowledge graph iteratively using an LSTM seqeunce model until a well representative graph is constructed for each contract text. Another novel aspect of the model was in implementing the model for chinese text and using the semantic role labelling for the Chinese language. This shows generalizability of computational linguistices across domains using first principles and the associated NLP modelling of the training dataset.

CAASI Grief to Action Initiative: co-designing platforms with BLM community partners

Speaker: Dr. Sera Linardi

This walk was specifically about the CASSI system, training and prediction based on the dataset collected and understooed from several policy makers in the field. The process involves four steps - Immersion, Predictive model, prescriptive algorithms and field test & deployment. The key part of the process was involvement of student, local governments, volunteers, community organisations who came together to take definitive actions against the injustices caused to different sections of the society over the past year. They specifically targeted ths issues that can be resolved through technological interventions and understood the process to build the models to fulfill these interventions

The key takeaway from the talk was to understand the process of developing technological intervention in the field of HCI through user studies, design interviews, focus groups and collaborating with the community to understand the grienvances, concerns, needs and develop tools that enable them better and so there is social justice, and the system can be mitigated of any biases.

Further, the research team prototyped two separate applications – PxPUC, an application to log misconduct complaints and 412Connect, an initiative to support business ventures of people of different ethinicity. The former application was prototyped as a part of the CS capstone class, a collaborative effort of the CS department, GSPIA and all stateholders striving to bring a change in the system.