Dynamic Long Short-Term Memory Based Memory Storage For Long Horizon LLM Interaction

TL;DR

This paper introduces Pref-LSTM, a lightweight framework combining a BERT classifier and LSTM memory to enhance long-term personalization in large language models, emphasizing preference filtering over extensive fine-tuning.

Contribution

The paper presents a novel, scalable approach for user preference modeling in LLMs using a BERT-based classifier with LSTM memory, avoiding heavy fine-tuning.

Findings

BERT classifier reliably identifies user preferences.

LSTM memory encoder did not show strong results.

Preference filtering is a viable path for scalable personalization.

Abstract

Memory storage for Large Language models (LLMs) is becoming an increasingly active area of research, particularly for enabling personalization across long conversations. We propose Pref-LSTM, a dynamic and lightweight framework that combines a BERT-based classifier with a LSTM memory module that generates memory embedding which then is soft-prompt injected into a frozen LLM. We synthetically curate a dataset of preference and non-preference conversation turns to train our BERT-based classifier. Although our LSTM-based memory encoder did not yield strong results, we find that the BERT-based classifier performs reliably in identifying explicit and implicit user preferences. Our research demonstrates the viability of using preference filtering with LSTM gating principals as an efficient path towards scalable user preference modeling, without extensive overhead and fine-tuning.