Abstract
Outdated or inaccurate information about a business or other point of interest in a digital map can be problematic, as users may waste time looking for a business that has closed or that is not open at the time of a user’s visit. This disclosure describes the use of a large language model (LLM) to determine attribute values for a business based on authoritative online sources. The attribute values are compared to the information about the business stored on the digital map to determine a freshness score. In some implementations, a smaller student model may be trained from a LLM for the specific purpose of business information update. The model may be instruction-tuned and have a long context to enable it to process long prompts that include instructions and data. The model may be executed in offline mode to obtain information updates at a low computational cost. The freshness score about the business and about individual attributes of the business is provided to users of the digital map and serves to inform them of the reliability and freshness of the information. User feedback regarding stale information is obtained and utilized to update business information on the digital map. Contributing users can be provided a measure of their contributions to improving the freshness score for various businesses.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Roy, Subhasish and Kar, Debabrat, "Freshness Score for Point of Interest Data in Digital Maps Using Large Language Models", Technical Disclosure Commons, (February 23, 2024)
https://www.tdcommons.org/dpubs_series/6712