Retrieval Model for Legal Answers
Revolutionizing the Work of Lawyers
Lawyers are often faced with a large number of questions from clients, many of which require extensive research and time-consuming answers. Such work can be quite monotonous, especially in similar cases. We have created a new solution to help lawyers in this task. Our retrieval model is an AI tool designed to find legal answers to questions asked by clients of the law firm. This model is able to automate and significantly optimize the work of lawyers, helping them find answers to questions faster and more accurately.
Model
The model is based on the transformer architecture, a design that is recognized for its ability to capture long-range dependencies and relationships between words in a sentence, making it an ideal choice for our retrieval system. The model has been specifically designed to work for the Polish language, which is known for its grammatically complicated structure. Despite these challenges, the model has been highly effective in finding answers to legal questions in Polish.
In our case, we trained BERT on a dataset of over 10,000 real legal inquiries to lawyers. The model takes in a new inquiry, transforms it into embedding using the encoder, and then compares it to the embeddings of the other inquiries. Finally, it returns the answer corresponding to the most similar query.
Validation strategy
To ensure the quality of our model we employed a two-fold validation strategy. The first validation method was an automated check of the model's response accuracy. The second and more critical method involved manual validation by legal experts. These experts assessed the model's ability to provide helpful suggestions in generating answers to new queries. According to expert estimations, a model that provides the two most likely answers can be useful in more than 80% of scenarios.
Conclusion
The legal answer retrieval model is a valuable asset for lawyers. It automates the search process especially repetitive cases, allowing lawyers to focus on more demanding tasks, and has a high level of accuracy.