The Text-to-SQL task is defined as “given a relational database and a natural language sentence that describes a question on the database, generate an SQL query over the database that expresses the question”. Text-to-SQL strategies based on Large Language Models (LLMs) achieve remarkable performance on well-known benchmarks, but their performance is significantly less for real-world databases. Some of the reasons for this performance decrease lie in the mismatch between the end user’s view of the data and the organization and naming conventions of the database schema, and in the differences between the end user’s data semantics and the encoding of such semantics in the database. This article then proposes an LLM-based strategy to compile natural language questions into SQL queries that uses a knowledge graph and incorporates a dynamic few-shot examples technique. The implementation of the strategy leverages a database keyword search tool and specific naming conventions to expose the knowledge graph to an LLM thereby simplifying the text-to-SQL task. The article includes experiments with a real-world, proprietary relational database and the Mondial database to assess the performance of the proposed strategy. The experiments suggest that the strategy achieves an accuracy on challenging relational databases that surpasses state-of-the-art approaches on the same databases.

This paper describes the steps taken to construct a Natural Language interface based on Large Language Models (LLMs) for Busca360, an application in production at an energy company. The paper analyses how technologies developed for text-to-SQL perform on the Busca360 database and proposes solutions to overcome the limitations observed. The analysis is based on a benchmark that uses the Busca360 database and a set of 100 Natural Language (NL) questions and their ground truth SQL translations. The paper first investigates how text-to-SQL strategies perform on the Busca360 relational schema. The experiments showed that the performance of the strategies tested was significantly less than that reported in the leaderboards, mainly because schema linking failed. The Busca360 database was then equipped with views that rename the table and column names and have new columns that predefine joins that follow foreign keys. A second set of experiments using such views corroborated that schema linking largely improved, and the predefined joins reduced the complexity of the NL question translation. To enhance the results further, a third set of experiments tested a dynamic few-shot examples technique that prompted the LLM with examples of NL question/SQL translation, retrieved from a synthetic dataset pre-computed from the database. The paper concludes by outlining the current LLM-based Natural Language interface using views and the dynamic few-shot examples technique, and leveraging the Busca360 database keyword search services.

This work focuses on analyses of cooperative protocols to enhance the performance of power line communication systems. Based on a measurement campaign and considering a sum power constraint, achievable data rates for amplify-and-forward (AF) and decode-and-forward (DF) protocols are analyzed. Similar investigations are performed for the maximum data rates attained using Hermitian-symmetric orthogonal frequency division multiplexing (HS-OFDM) together with equal gain combining (EGC), selection combining (SC) and maximal ratio combining (MRC) techniques. The influences of optimally and uniformly allocated transmission power and frequency bandwidth are are also analyzed and the efficiency of combination before and after equalization is compared. Results show that the relative distances among source, relay and destination nodes significantly impact system performance. Also, they reveal a range of total transmission power and bandwidth in which benefits can be verified. Among combining techniques, MRC and SC present similar results, but MRC offers a slightly better performance. In relation to computational complexity, SC is the most favorable. Maximum data rate analyses of HS-OFDM with frequency domain equalization based on zero forcing and minimum mean square error criteria show that the former scheme offers almost the same performance as the latter. Furthermore, it is shown that equalization after combination is more advantageous.