Training-Free Lexical-Dense Fusion for Conversational-Memory Retrieval

• URL: http://arxiv.org/abs/2606.04194v1
• Date: Tue, 02 Jun 2026 20:22:16 GMT
• Title: Training-Free Lexical-Dense Fusion for Conversational-Memory Retrieval
• Abstract summary: Retrieving the few past turns that answer a new query is the retrieval bottleneck behind long-term conversational memory.<n>Recent work, Nano-Memory, shows that scoring a session by the maximum query-turn similarity beats mean-pooled session embeddings.<n>We replicate it and ask what a training-free, CPU-only retrieval stage should add around it.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Retrieving the few past turns that answer a new query across long multi-session histories is the retrieval bottleneck behind long-term conversational memory (LoCoMo, LongMemEval). Recent concurrent work, Nano-Memory, shows that scoring a session by the maximum query-turn similarity (late interaction, "Turn Isolation Retrieval") beats mean-pooled session embeddings. We do not claim that effect; we replicate it and ask what a training-free, CPU-only retrieval stage should add around it. We report four findings. (1) Fuse: score-level fusion of the late-interaction dense score with BM25, under a single leave-one-conversation-out weight, adds +8.8 to +17.2 points of LoCoMo Hit@1 over late interaction alone across six encoders (all p<1e-4), reaching Hit@1 0.752 / NDCG@5 0.829 (e5-large-v2), +11.2 pp over BM25. (2) An off-the-shelf web-search cross-encoder reranker over the fused top-10 hurts here, degrading Hit@1 by 6.9 pp (one reranker, one configuration). (3) A pooling-operator ablation shows top-k late interaction matches max-similarity, but a naive smooth-max (log-sum-exp) collapses for half the encoders. (4) The late-minus-early gap is large for all six encoders and tends to be larger for larger ones, while the marginal fusion gain shrinks; on LongMemEval-S, a lexical regime where BM25 saturates, the net fusion gain over BM25 is small and not significant. A per-category analysis frames the gain as a division of labor: dense late interaction helps most on multi-hop and temporal questions but trails BM25 on adversarial ones. The contribution is a controlled, reproducible account of a strong training-free retrieval recipe, not the late-interaction retriever itself (Nano-Memory's). We make no claim to a complete memory architecture; this is a retrieval-stage study.

Related papers

• PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents [85.82624962221026]
  We introduce PARSER, which decouples reading from reasoning.<n>A bank of lightweight subagents each bound to a single chunk read the entire document in parallel.<n>On multi-hop QA with contexts ranging from 7K to 896K tokens, PARSER with a 4B backbone outperforms the strongest sequential memory baseline.
  arXiv  Detail & Related papers  (2026-09-06T16:19:01Z)
• Entity-Memory Graph Retrieval Improves Evidence Coverage in Long-Conversation Question Answering [0.0]
  On 1,986 questions from ten LoCoMo conversations, graph retrieval raises official evidence recall at top-k 25 from 79.7468% to 84.4842%.<n>Four paper-eligible requested configurations support empirical robustness across the tested GPT-3.5 and DeepSeek extractors on both outcomes.
  arXiv  Detail & Related papers  (2026-08-28T05:03:33Z)
• LINE Conversation History Retrieval for Personal Memory RAG: Evaluating Search Representations and Hybrid Retrieval [0.0]
  This study presents a retrieval-only case study over one user's LINE conversation history.<n>We segmented 358,896 messages into 22,329 temporally coherent chunks and constructed three search representations.<n>We compared BM25, dense vector retrieval, and linear hybrid retrieval on 100 evaluation questions verified by a single annotator.
  arXiv  Detail & Related papers  (2026-08-28T00:52:11Z)
• PUMA: Post-Hoc Sparsification of Universal Multimodal Embeddings for Efficient Retrieval [50.986807025432476]
  We introduce PUMA, a sparse autoencoder recipe that maps universal multimodal embeddings to compact sparse codes without retraining the backbone.<n>We evaluate on five benchmarks covering text-to-image and composed image retrieval.<n>Puma reduces vector storage by 8-16x (FP32) and is up to 25x faster than exact dense scoring on larger candidate pools.
  arXiv  Detail & Related papers  (2026-08-26T13:26:39Z)
• Evidence-Unit Fairness and the Limits of Query-Adaptive Sparse-Dense Fusion in Financial Document Retrieval [4.371552631761563]
  We study sparse-dense hybrid retrieval on FinDER, a benchmark of expert-annotated questions over corporate 10-K filings.<n>We find that if the retrieval unit is larger than the dense encoder's input window, the dense model never sees a large share of the labeled evidence.<n>On the corrected corpus, fusing BM25 and a compact dense encoder improves reference-level Hit@10 by roughly 28 percent over either component.
  arXiv  Detail & Related papers  (2026-07-31T18:06:58Z)
• Don't Ask the LLM to Track Freshness: A Deterministic Recipe for Memory Conflict Resolution [0.0]
  We show that replacing the LLM-judgment answer pipeline with candidate-extraction plus Python max(serial) yields +10.8 points on FC-SH (gpt-4o-mini)<n>The recipe reaches 78.0% on FC-SH (gpt-4o-mini), 94.8% (gpt-4o), and 30.2% on FC-MH (gpt-4o-mini) via a per-hop deterministic extension of Self-Ask.
  arXiv  Detail & Related papers  (2026-05-31T20:13:43Z)
• AgentIR: A Workload-Adaptive Cascade Retrieval Substrate for Long-Term Conversational Memory [1.8479558716666358]
  Lucene-class engines treat the index as static and the query as stateless, leaving the workload's structure unexploited.<n>A cascade router decides from the BM25 top-k margin alone and re-tunes across workloads without retraining.<n>Capacity on a shared 8-core VM rises from 154 to 1,400 concurrent agents (9x)
  arXiv  Detail & Related papers  (2026-05-24T14:14:13Z)
• Cross-Session Threats in AI Agents: Benchmark, Evaluation, and Algorithms [0.0]
  CSTM-Bench is 26 executable attack classified by kill-chain stage and cross-session operation.<n>Released on Hugging Face as intrinsec-ai/cstm-bench with two 54-scenario splits: dilution (compositional) and cross_session.
  arXiv  Detail & Related papers  (2026-04-22T22:40:31Z)
• Test-Time Strategies for More Efficient and Accurate Agentic RAG [58.44913384057518]
  Retrieval-Augmented Generation (RAG) systems face challenges with complex, multihop questions.<n>Such approaches can introduce inefficiencies, including repetitive retrieval of previously processed information.<n>We investigate test-time modifications to the Search-R1 pipeline to mitigate these identified shortcomings.
  arXiv  Detail & Related papers  (2026-03-12T19:18:59Z)
• Test-Time Scaling with Diffusion Language Models via Reward-Guided Stitching [66.39914384073145]
  We propose a self-consistency framework that turns cheap diffusion-sampled reasoning into a reusable pool of step-level candidates.<n>We find that step-level recombination is most beneficial on harder problems.<n>Our training-free framework improves average accuracy by up to 2 across six math and coding tasks.
  arXiv  Detail & Related papers  (2026-02-26T11:08:39Z)
• Domain-Adaptive and Scalable Dense Retrieval for Content-Based Recommendation [0.0]
  We present a scalable dense retrieval system based on a two-tower bi-encoder, fine-tuned on the Amazon Reviews 2023 (Fashion) subset.<n>We construct training pairs from review text (as a query proxy) and item metadata (as the positive document) and fine-tune on 50,000 sampled interactions with a maximum sequence length of 500 tokens.<n>On a review-to-title benchmark over 826,402 catalog items, our approach improves Recall@10 from 0.26 (BM25) to 0.66.
  arXiv  Detail & Related papers  (2026-01-31T20:58:23Z)
• CourseTimeQA: A Lecture-Video Benchmark and a Latency-Constrained Cross-Modal Fusion Method for Timestamped QA [3.0867286767516937]
  Given a natural-language query, the system retrieves relevant timestamped segments and synthesizes a grounded answer.<n>We present CourseTimeQA (52.3 h, 902 queries across six courses) and a lightweight, latency-constrained cross-modal retriever (CrossFusion-RAG)<n>On CourseTimeQA, CrossFusion-RAG improves nDCG@10 by 0.10 and MRR by 0.08 over a strong BLIP-2 retriever while achieving approximately 1.55 s median end-to-end latency on a single A100.
  arXiv  Detail & Related papers  (2025-11-29T07:06:51Z)
• MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learning [73.27233666920618]
  We propose MemSearcher, an agent workflow that iteratively maintains a compact memory and combines the current turn with it.<n>At each turn, MemSearcher fuses the user's question with the memory to generate reasoning traces, perform search actions, and update memory to retain only information essential for solving the task.<n>We introduce multi-context GRPO, an end-to-end RL framework that jointly optimize reasoning, search strategies, and memory management of MemSearcher Agents.
  arXiv  Detail & Related papers  (2025-11-04T18:27:39Z)
• Squeezed Attention: Accelerating Long Context Length LLM Inference [61.787865959140994]
  We propose Squeezed Attention to accelerate applications where a large portion of the input context is fixed.<n>During inference, we compare query tokens from the user input with the centroids to predict which keys from the fixed context are semantically relevant.<n>We also present a hierarchical version of our algorithm which can reduce the complexity of attention from linear to logarithmic with respect to the fixed context length.
  arXiv  Detail & Related papers  (2024-11-14T18:54:19Z)
• Efficient k-NN Search with Cross-Encoders using Adaptive Multi-Round CUR Decomposition [77.4863142882136]
  Cross-encoder models are prohibitively expensive for direct k-nearest neighbor (k-NN) search. We propose ADACUR, a method that adaptively, iteratively, and efficiently minimizes the approximation error for the practically important top-k neighbors.
  arXiv  Detail & Related papers  (2023-05-04T17:01:17Z)
• Improving Dual-Encoder Training through Dynamic Indexes for Negative Mining [61.09807522366773]
  We introduce an algorithm that approximates the softmax with provable bounds and that dynamically maintains the tree. In our study on datasets with over twenty million targets, our approach cuts error by half in relation to oracle brute-force negative mining.
  arXiv  Detail & Related papers  (2023-03-27T15:18:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.