» » »

How to Build Personalization into LLM Recommendations

We enable Large Language Models (LLM) with personalization capability. This is not specific to the LLM (Open AI's ChatGPT, Athropic's Claude, Meta's Llama 2, Googles,...)

Today, LLMs are not good at personalization and providing recommendations. They may advise physicians and financial advisors to "ask professionals" in their respective fields for help, even having user information available. When answering questions for software professionals, the LLM may need to deliver in-depth answers with code or algorithms, whereas for professionals in other fields would need definitions and main concepts.

The intent of this project is to make LLMs provide answers tailored to the needs of a specific user, taking into account available information about that individual. To do that, we need to generalize available documents about a person. Based on the needs of the application and with the permission of the individual being served, information used could include: their LinkedIn profile, visited web pages, investment history extracted from tax documents, and health forms (while maintaining the privacy of this person). We rely on meta-learning techniques to design an LLM prompt to produce a personalization prompt to obtain suitable relevant information. Such a “meta-prompt” is produced by a generalization operation applied to available documents for the user. These documents need to be de-identified so that they are sufficient for personalization, on one hand, and will maintain user privacy on the other hand.

A personalization profile is built from the link provided by the user.

Then, given a user question, this system will use the LLM to generate a set of queries. The URLs from search results are stored internally in a self.urls. A check is performed for any new URLs that haven't been processed yet (not in self.url_database). Only these new URLs are loaded, transformed, and added to the vector store. The vector store is queried for relevant documents based on the questions generated by the LLM. Only unique documents are returned as the final result.

This project build is in https://github.com/bgalitsky/LLM-personalization

Attend in person or online

Speaker: Boris Galitsky

Monday, 10/23/23


Website: Click to Visit



Save this Event:

Google Calendar
Yahoo! Calendar
Windows Live Calendar

Hacker Dojo

855 Maude Avenue
Mountain View, CA 94043