PricePediaGPT: An Application for Accessing PricePedia Content Through Dialogue

Generative artificial intelligence in 2023 has captured the broad public's attention with ChatGPT. This system is capable of reading a question or a command written by a user and providing an answer in human language, constructing, through a statistical algorithm, sentences relevant to the question. In this activity, ChatGPT deploys three different areas of expertise:

1. the ability to understand the question posed by the user;
2. the capability to generate content that contains an answer to the question;
3. the ability to transform the generated content into high-quality human text.

Those who have used it have been impressed by its ability to understand and produce quality texts. However, there have been several concerns regarding its ability to always produce content relevant to the posed question. ChatGPT's goal of conversing on any lawful topic of human knowledge represents an ambitious challenge that is continuously being refined.

GPT Applications

ChatGPT has made its dialogue environment available to partner developers, allowing them to develop applications (called GPTs) that leverage the system's capabilities to understand questions and generate responses. Developers are tasked with integrating and consulting their own information systems to enrich the responses, tailoring them to the specific requests of users. Thanks to these functionalities, PricePedia has built an application within the ChatGPT environment that allows users to receive answers based also on the nearly 1500 articles published in the last five years in the PricePedia magazine.
The ability to access the contents of PricePedia articles makes the PricePediaGPT application particularly useful for the economic and conjunctural analysis of commodity markets. Although ChatGPT has been trained on a wide range of general content up to the beginning of 2023, it does not possess specialized knowledge or post-training updates on commodity markets. Therefore, PricePediaGPT utilizes PricePedia articles to provide updated and specific analyses.

By accessing chat.openai.com (with a PLUS subscription^[1]), the application can be searched using the feature from the main menu.

The PricePediaGPT application allows a user to have normal conversations with ChatGPT, but with the ability to search within the articles of the PricePedia magazine for content to answer the user's question.

Once the question is entered, ChatGPT indicates that the system wants to access the pricepedia.it website and asks the user to confirm the site's reliability:

Once this confirmation is given, the application accesses the articles from the magazine to search for information from which to derive the answer to the user's question.

A Look at the PricePediaGPT Application

Some information about the operation of the PricePediaGPT application may be useful to potential users.

For every question posed by the user via the ChatGPT interface, the PricePediaGPT application performs this process:

• interpretation of the question: this task is performed by the GPT4 model which interprets the user's question and transforms it into a list of keywords;
• search for paragraphs starting from the list of keywords provided by GPT4, a search system developed by Vicem srl searches for paragraphs in PricePedia articles that contain information relevant to the question;
• generation of the answer: using the content of the selected paragraphs, the GPT4 model generates the response to be presented to the user.

The most consolidated stages of this process are naturally the interpretation of the question and the generation of the answer. The component with the greatest room for improvement concerns the search for content within the PricePedia articles. These developments are possible in two directions:

• expansion of the set of accessible articles: it is clear that the more content accessible for search, the higher the quality of the response to possible user questions will be;
• accuracy of the search: given a list of words, there are many algorithms capable of searching within a database of paragraphs for those that are "closest" to them. The study developed by Vicem has identified that the combination of two algorithms (one literal and one semantic) is currently the most effective. However, it is natural that this is only the first step of what may be a long journey.

Project Objectives

With this project, PricePedia aims to develop an access system to knowledge about commodity markets that the PricePedia team has described in articles published in the PricePedia Magazine.
We hope that many will use it and share their opinions with us.