Problem Statements
- October 4 2019 October 5 2019
Please stay tuned in for more information on the datasets for the "featured" problem statements.
1). Data Error Challenge (Sponsored by the Global Ops and Content Systems Teams)
Can you find the most “errors” in our Judge Entity Master data?
You will be given access to our Judge Master viewer tool, YoDa, as well as our judge data (a data dump of the top ~5000 active judges will be loaded into an S3 location listed below).
Obviously, you can go through and check things one-at-a-time by hand…but you won’t win that way!!
Can you use machine learning and/or web-scraping to validate the information?
If you find anything incorrect, provide the correction and the source.
Our judges (LexisNexis Data Stewards) will validate you findings and tally them up.
The team with the most corrections (and most automatable process) wins!
For details on the datasets, please contact your mentors or Rick McFarland/Mike Etgen.
2). Predict user intent in Searches (sponsored by the NARS)
The challenge is to develop a smarter system for recognizing user intent to “get a document” and resolving their input to a specific document or set of documents on our system.
Retrieving a specific document is an extremely common task of legal researches, making up roughly 50% of the “searches” submitted on Lexis Advance.
With the shear volume of submissions, and a variety of user input, there are thousands each month likely intended to retrieve a specific document but instead are not recognized, resulting in various levels of success when falling back to standard search behavior.
The ideal solution would include or consider:
Use of NLP to detect user intent as “get a document” and improve upon the current “black or white” rules approach.
A system that learns and improves over time (machine learning) is desirable.
A more forgiving experience (“fuzzy matching”) is desirable as citation formats are widely varied as is our user input
For details on the datasets, please contact your mentors or Rick McFarland/Mike Etgen.
Scoring
Judges will ask the teams that take up this challenge to run their classifier against a hold-out sample, whoever has the highest F1 scores wins!
3). Make an AI "test taker bot" that takes (and passes!) the Multi-state Bar Exam (sponsored by Jamie Buckley, CPO)
Through this challenge, we want you to create an AI system that can pass the contracts portion of the Bar Exam (more specifically the multi-state, multiple-choice portion).
A sample question would look like the one that can be found here: MBE Practice Test
The overall goal of this Challenge is to assess the state-of-the-art in legal AI, and to measure it against a test that every lawyer knows.
Scoring
Basically, if you make a "bot" that can answer a set of MBE bar questions correctly, you are a winner!
4). Legal Citations Network Visualizations (Sponsored by the UX Team)
Citations play a huge role in the way law is practiced. When attorneys write their arguments, and when judges write their opinions, they will cite "authorities" that support their position or ruling.
The authorities are often prior cases from the highest court in the same jurisdiction, or the US Supreme Court.
Visualizing how cases relate to each other through that citation network is helpful in understanding not only what cases are most authoritative, but also how the legal issue cited within the cases is being "framed" by those judges.
This challenge is to use either the Harvard caselaw data, or the Legal Issue Trail feature in Lexis Advance, to create the best visualization of a citation network for a legal issue described and cited within cases.
For details on the datasets, please contact your mentors or Rick McFarland/Mike Etgen.
Scoring
Judges will examine the submissions for this challenge based upon creativity, usability, and accuracy.
5). Build the best Chatbot to answer LPA practice note questions (Sponsored by the Lexis Practice Advisor)
Lexis Practice Advisor provides practical guidance for attorneys on many different topics.
Some of the "practice notes" are formatted to include "questions and answers", like an FAQ style document.
An issue for customers however is in tracking down the right document with the answers they need using either a browse or keyword search capability.
It just takes too much time and clicks to get to the right content.
This challenge is to provide a natural language question asking interface on top of practice notes such that the user can ask a question and receive a precise and correct answer from a practice note.
For example, "Who must comply with data breach notification rules in California?" should return the matching "answer" portion from the California Data Breach Notification practice note.
Scoring
Judges will ask the teams that take up this challenge to run their classifier against a random sample of questions, whoever answers the most question accurately wins!
6).Choose Your Own Adventure
Create an innovative solution for a LexisNexis product (current or new).
Ideally the solution should not be on the product backlog or roadmap.
The idea is your own take on future of the product or market need. No idea is too big or too small, as long as it scratches a real itch.
Feel free to use any technologies, any LexisNexis product, or a combination of things.
Mobile Apps, Data Visualizations, Chatbots, Voice Recognition (Alexa, Google Home), Virtual Reality, Web Apps, Digital Media, etc...
Choose your own problem to solve and convince us of your future vision!