Vector-based indexing of Federal Supreme Court judgments - On November 13, 2024

Thomas Murphy reported at the DA-3 workshop on the retrieval effectiveness of LLM-based search methods to find relevant judgments of the Swiss Federal Supreme Court. To determine the search quality - whether relevant judgments are effectively found or not - a test collection is essential. Thomas uses a test collection from Eurospider with 123 test queries for which the relevant judgments are known. He compared the retrieval effectiveness of different search methods based on their yield and precision precision and recall. After numerous experiments, an improvement was achieved on the test collection with a combination of LLM-based vector retrieval and text retrieval optimized for legal information.

More Information This email address is being protected from spambots. You need JavaScript enabled to view it..

Information Retrieval

The objective of Information Retrieval (IR) is to search large data collections for information relevant to a user’s information requirements. The term “information retrieval” was coined by Calvin Mooers in 1950. Like “research” the word “retrieval” does not refer to refinding something. It rather relates to the information retrieval paradox: “If I knew what I was searching for, I wouldn’t be searching for it.”

Information retrieval is focuses on three dimensions: systems and applications, theory and models, evaluation. Various retrieval models exist, such as Vector Space Model (VSM) and probabilistic and language models. For evaluatio,n recall and precision are often used. SMART was an early retrieval system that dealt with all three aspects. RankBrain is a more recent retrieval system based on TensorFlow.

WebGND

The Integrated Authority File (German: Gemein­same Norm­datei or GND) is an inter­national authority file used and maintained by the German National Library (German: Deutsche National­bibliothek or DNB), all German-language library associations, the Zeit­schrift­en­daten­bank (ZDB) and many other insti­tutions. WebGND is an online application that supports navigation and search within this large database which consists of more than 11 million records covering personal names, corporate names, meeting names, geographic names, topical terms and uniform work titles.

Eurospider Information Technology AG
Winterthurerstrasse 92
8006 Zürich

 

Cookies make it easier for us to provide you with our services. With the usage of our services you permit us to use cookies.
More information Ok Decline