Dingyang Lyu, Zhengjia Xu, Jey Han Lau, Jianzhong Qi
Human mobility prediction, a core task of location-based web services, forecasts a user's next Point of Interest (POI) visit from historical trajectories, supporting applications from recommendation to urban planning. A general perception in the literature has been that visits to popular POIs are relatively easy to predict. Our analysis challenges this perception and shows that, even for visits to popular POIs, many predictions fail. The underlying cause is often transition-level sparsity, where a transition is a move from a source POI to a destination POI. While a POI may be popular, its transitions from other POIs may still be very sparse, since there are many POIs in a dataset. As a result, predicting a popular next POI can still be challenging if the transition from the source POI to it never appears in the training data. We therefore argue that a core bottleneck in human mobility prediction lies in the transition-level long tail, and propose REtrieval Combined with pArametric memory for next-POI prediction (RECAP). RECAP learns frequent transitions in the parameters of a small Transformer trained from scratch. Long-tail transitions give the parameters little or no training signal. RECAP retrieves evidence for them from two sources: (i) multi-hop transitivity learned from a global transition graph of all users, and (ii) revisit evidence from a target user's historical trajectory. RECAP further uses two-stage training to let the revisit component also learn from frequently seen transitions. Experiments on three real-world check-in datasets show that RECAP improves prediction accuracy, with clear gains on tail transitions. Our code is available at https://github.com/NiallLDY/RECAP.
Linh Tran, Yulong Li, Radu Florian, Stacy Patterson, Ana Milanova, Wei Sun
The strong zero-shot and long-context capabilities of recent Large Language Models (LLMs) have paved the way for highly effective re-ranking systems. Attention-based re-rankers leverage attention weights from transformer heads to produce relevance scores, but not all heads are created equally: many contribute noise and redundancy, thus limiting performance. To address this, we introduce CoRe heads, a small set of retrieval heads identified via a contrastive scoring metric that explicitly rewards high attention heads that correlate with relevant documents, while downplaying nodes with higher attention that correlate with irrelevant documents. This relative ranking criterion isolates the most discriminative heads for re-ranking and yields a state-of-the-art list-wise re-ranker. Extensive experiments with three LLMs show that aggregated signals from CoRe heads, constituting less than 1% of all heads, substantially improve re-ranking accuracy over strong baselines. We further find that CoRe heads are concentrated in middle layers, and pruning the computation of final 50% of model layers preserves accuracy while significantly reducing inference time and memory usage.
Local-commerce and food-delivery marketplaces run large ads businesses: across hundreds of thousands of stores, merchants pay to be surfaced on browse and search surfaces, and a ranker decides which ads each consumer sees. Ranking well demands strong per-consumer personalization, yet a consumer's raw interaction history is sparse and high-cardinality. Our production ranker had personalized only through hand-engineered signals: aggregate engagement counts, categorical embeddings, and order-agnostic pooled history, with no learned per-consumer sequence representation. We learn one by training a GPT-style generative model of the consumer's engagement sequence to predict their next few engagements, and read its last-token hidden state as the consumer embedding. That single frozen vector becomes one shared personalization input for every ad surface: a daily job refreshes it, and each ranker consumes it as an ordinary dense feature. Offline it is the largest single-feature gain in our ranking-AUC ladder, a relative Eval-AUC gain of +0.59% on a browse-driven feed and +0.57% on a query-driven search surface. Online, it shipped as one component of bundled ranker retrains on both surfaces: Discovery (revenue +0.79%, conversion rate +1.17%) and Search (revenue +0.94%, conversion rate +1.50%). Three design choices adapt the generative objective to a repurchase-heavy marketplace, and are the core of our contribution: an exploration-weighted loss that up-weights businesses new to the consumer over ones already engaged; a two-stage curriculum: a click pretrain that keeps the representation broad and exploratory, then a conversion fine-tune that sharpens purchase intent; and a designed negative set that pairs a fixed popular-business pool and businesses shown but not ordered with random and in-batch negatives.
Sedigheh Eslami, Markus Krimmel, William Barber, Antoine Chaffin, Maximilian Schall, Jonas Golde, Louis Milliken, Ekaterina Glazkova 等 10 位作者
Contextual embeddings represent chunks of a document with respect to the document's global context. These models are typically trained and evaluated under the assumption that a relevant document contains only a single relevant chunk. In this work, we argue that in real-world use cases a retriever should not only find answers spread over several chunks but also surface the evidence needed to verify them. To address this, we introduce pplx-embed-v2-context, a family of contextual embedding models trained by distilling graded chunk-level relevance from a query-aware context compression model. This removes the need for gold-chunk labels and makes the training targets independent of the chunking. We utilize a vision-language backbone for training and release pplx-embed-v2-context in 9B and 0.6B sizes with compatible embedding spaces. On the evaluation side, we introduce two benchmarks that label both the chunks that answer a query and the evidence needed to verify the answer. Context-bench is a controlled diagnostic, and Q2D-Chunk is built from real-world search queries. To encourage further research, we also publish Q2D-Chunk as a public leaderboard. Our 9B model sets a new state of the art on both benchmarks and ConTEB. We find that the visual retrieval capabilities of our 9B model transfer from its vision-language base model and propose to extend chunking to image inputs. We demonstrate that our models achieve state-of-the-art performance among dense retrievers on visual ViDoRe-V3 and that performance improves with granular image chunking.
Chenyang Yu, Zhaomin Xiao, Yan Huang, Eduardo Blanco
Gardening in the backyard implies living there, while checking into an Airbnb more likely means staying for a few days. Such duration judgments are important for understanding human mobility, travel behavior, and social context. However, duration is rarely stated explicitly in social media posts and must instead be inferred from noisy textual, visual, and contextual cues. In this paper, we investigate whether vision-language models possess such duration commonsense. We create a new corpus of geotagged tweets co-annotated by humans and LLMs to serve as a testbed for this task. We find that state-of-the-art models struggle with duration inference, even when guided by multi-turn Chain-of-Thought prompting. Error analyses further reveal that models often fail to integrate critical contextual information, or they overly rely on prominent visual details while overlooking subtle hashtags and semantic nuances. These findings underscore the significant challenges remaining in context-aware multimodal understanding.
Mainak Ghosh, Michael E. Rose, Sebastian Erhardt, Erik Buunk, Dietmar Harhoff
PaECTER is an open-source document-level encoder specific for patents. We fine-tune BERT for Patents with examiner-added citation information to generate numerical representations for patent documents. PaECTER performs better in similarity tasks than current state-of-the-art models used in the patent domain. More specifically, our model outperforms the patent specific pre-trained language model (BERT for Patents) and general-purpose text embedding models (e.g., E5, GTE, and BGE) on our patent citation prediction test dataset on different rank evaluation metrics. PaECTER predicts at least one most similar patent at a rank of 1.32 on average when compared against 25 irrelevant patents. Numerical representations generated by PaECTER from patent text can be used for downstream tasks such as classification, tracing knowledge flows, or semantic similarity search. Semantic similarity search is especially relevant in the context of prior art search for both inventors and patent examiners.
Ranking aggregation from multiple retrievers is a powerful yet under-explored technique in modern retrieval and RAG systems. We introduce LLM-IRA, the first extension of the quantum-inspired interactive ranking aggregation framework to document reranking. By leveraging large language models (LLMs) solely for pairwise judgments on high-disagreement document pairs, our method learns fusion weights in a principled, interactive way. The final ranking is a simple, ultra-fast pointwise linear fusion of scores, no full pairwise matrix or expensive listwise prompting is required. Experiments on the BEIR benchmark and TREC Deep Learning tracks with an open-weight instruction-tuned judge (Llama-3.1-8B-Instruct) show that LLM-IRA matches or numerically exceeds strong static fusion baselines on 5 of 6 evaluated benchmarks. We report two weight-update variants: momentum blending and a full accumulated-history refit, which together reach statistical significance (Wilcoxon signed-rank, p<0.05) over the best unsupervised aggregator on 4 of 6 benchmarks (DL19 under momentum; DL20, DBpedia, and NFCorpus under accumulated refit, which is the stronger of the two configurations overall). On the benchmarks we profiled (DL19, DL20), it already clears the best unsupervised baseline within as few as 5 LLM calls per query, well before exhausting its typical ~25--30 call budget. LLM-IRA also achieves cost-effectiveness comparable to state-of-the-art zero-shot LLM rerankers at an order of magnitude fewer LLM calls, and requires no training data or model fine-tuning.
Personalized recommendation depends on learning semantic and collaborative relations among items from online content and user interactions. Generative recommendation (GR) offers a promising paradigm by autoregressively generating discrete semantic IDs (SIDs) of items. However, randomly initialized SID token embeddings and GR models trained from scratch do not directly inherit the semantic relations inherent in items, while inter-item collaborative relations are learned largely implicitly through token-level attention and next-token supervision. To address this limitation, we propose REGER, a unified framework that explicitly incorporates inter-item relations into semantic pretraining, history encoding, and SID generation. We pretrain the GR model through generative semantic retrieval, enabling it to inherit inter-item semantic similarity by learning to generate semantic neighbors' SIDs. We combine global attention with local attention aligned with SID boundaries to capture historical context and directed transitions between adjacent items. We further develop a theoretical framework for hierarchical preference decomposition over the SID prefix tree, proving an exact factorization of collaborative preference distributions and a weighted decomposition of item-level KL divergence into prefix-conditional KL divergences. Building on this analysis, we distill a discriminative teacher's full conditional distributions over all legal next tokens at ground-truth prefixes along the target SID path, jointly optimizing this objective with cross-entropy. Extensive experiments on real-world datasets demonstrate consistent improvements over existing GR baselines and support the effectiveness of the proposed preference distillation.
Pengyue Li, Sheng Wang, Hua Dai, Zhiyu Chen, Zhifeng Bao, Brian D. Davison
High-quality datasets are typically required for accomplishing data-driven tasks, such as training medical diagnosis models, predicting real-time traffic conditions, or conducting experiments to validate research hypotheses. Consequently, open dataset search, which aims to ensure the efficient and accurate fulfillment of users' dataset requirements, has emerged as a critical research challenge and has attracted widespread interest. Recent studies have made notable progress in enhancing the flexibility and intelligence of open dataset search, and large language models (LLMs) have demonstrated strong potential in addressing long-standing challenges in this area. Therefore, a systematic and comprehensive review of the open dataset search problem is essential, detailing the current state of research and exploring future directions. In this survey, we focus on recent advances in open dataset search beyond traditional approaches that rely on metadata and keywords. From the perspective of dataset modalities, we place particular emphasis on example-based dataset search, advanced similarity measurement techniques based on dataset content, and efficient search acceleration techniques. In addition, we emphasize the mutually beneficial relationship between LLMs and open dataset search. On the one hand, LLMs help address complex challenges in query understanding, semantic modeling, and interactive guidance within open dataset search. In turn, advances in dataset search can support LLMs by enabling more effective integration into retrieval-augmented generation (RAG) frameworks and data selection processes, thereby enhancing downstream task performance. Finally, we summarize open research problems and outline promising directions for future work. This work aims to offer a structured reference for researchers and practitioners in the field of open dataset search.
Retrieval-augmented generation (RAG) systems ground large language models (LLMs) by retrieving external evidence and conditioning generation on the retrieved context. Although dense retrieval is effective for local semantic matching, independently ranked retrieval can return redundant passages, miss cross-chunk dependencies, and provide incomplete evidence for multi-hop reasoning. We propose SpectRAG, a spectral graph-augmented retrieval framework that combines hybrid candidate seeding, weighted chunk graph construction, graph-based expansion, spectral relevance propagation, and token-budgeted context selection. Document chunks are represented by dense semantic embeddings and graph-spectral coordinates derived from a weighted chunk graph whose edges encode semantic similarity, document adjacency, section hierarchy, shared entities, citation links, and similarity between LLM-generated utility questions. At query time, SpectRAG builds a high-recall seed set using exact inner-product search and hybrid relevance signals, expands it over the chunk graph, re-scores candidates using graph proximity and spectral information, and assembles the final context under a fixed token budget. The goal is not merely to retrieve individually similar chunks, but to construct a context that is relevant, diverse, structurally connected, and sufficient for grounded reasoning. Experiments on MultiHopRAG, UltraDomain, and long-document question-answering settings show that SpectRAG improves broad evidence recovery and achieves competitive generation quality against dense, hybrid, HyDE, LightRAG, and NaiveRAG baselines. The results further indicate that graph-spectral structure is particularly useful for inference-oriented and cross-document retrieval, while highlighting trade-offs in early precision, computational cost, and domain-dependent performance.
Jun Yang, Jiasi Chen, Jiawei Sun, Ziyue Guo, Jianan Yan, Tao Luo
Semantic-ID-based generative recommendation unifies retrieval and ranking through autoregressive decoding but differs structurally from discriminative models, which can directly access rich target-item side features. The attributes are compressed into a finite codebook during quantization, while the target item remains undetermined during decoding and its features cannot be looked up. This target-side feature vacuum weakens generation at each semantic-ID level through high conditional entropy and cascading errors, and final ranking through missing target-side feature crossing. Existing methods interact with candidate representations or generate discrete attribute tokens, but none reconstructs a continuous target-side representation inside decoding to support both stepwise generation and ranking. We propose TacGR (Stepwise Target-Side Feature Anticipation and Crossing for Generative Recommendation). Before decoding each codeword, TacGR anticipates a continuous target-side feature vector and crosses it with the semantic prefix and user behavior sequence inside the decoder. The crossed representation conditions current-level generation and is directly used for ranking. Continuous-vector regression risks a learnable target collapsing with its prediction and prediction noise propagating into generation through attention. TacGR addresses these challenges with a semantic-anchoring auxiliary tower and a unidirectional visibility mask, respectively. For extremely long industrial behavior histories, it introduces gated convolutional denoising to suppress noise and adapts the pyramid memory funnel to reduce inference cost. Offline experiments on a large-scale Lazada dataset show significant gains over OneRec, UniRec, and UniSGR. Online A/B testing improves IPV, Order, and GMV by 0.79%, 5.55%, and 3.09%, respectively.
Retrieval-Augmented Generation (RAG) has become the standard paradigm for grounding Large Language Model outputs in external knowledge. Lumer et al. presented the first systematic evaluation comparing vector-based agentic RAG against hierarchical node-based reasoning systems for financial document QA across 1,200 SEC filings, finding vector-based systems achieved a 68% win rate. Concurrently, the PageIndex framework reported 98.7% accuracy on FinanceBench through purely reasoning-based retrieval. This paper extends their work by: (i) implementing and evaluating three retrieval configurations: Vector RAG, Tree Reasoning, and the proposed Adaptive Hybrid Retrieval (AHR), on a small controlled corpus of financial, legal, and medical documents; (ii) introducing a four-tier query complexity benchmark; and (iii) employing LLM-as-judge evaluation with GPT-4o-mini. Experiments reveal that Tree Reasoning achieves the highest overall score (0.900), but no single paradigm dominates across all tiers: AHR scores highest on cross-reference queries (0.850), all three methods tie on multi-section queries (0.929), and Vector RAG scores highest on the single multi-document query (0.900). Vector RAG has the lowest mean latency (1.61 s against 3.40 s for Tree Reasoning). On the first 50 publicly released FinanceBench questions, using the evidence snippets supplied with the benchmark as the retrieval corpus, Tree Reasoning scores 0.938, Hybrid AHR 0.901, and Vector RAG 0.821. Because the configurations also differ in document access, generation prompts and output length limits, these differences cannot be attributed to retrieval architecture alone. In these preliminary results the adaptive router does not outperform fixed tree reasoning overall (0.845 vs. 0.900); they motivate a larger evaluation of when query-based routing helps, rather than establishing its benefit.
Generative recommendation (GR) has emerged as a promising paradigm that predicts target items by autoregressively generating their semantic identifiers (SID). Most GR methods follow a quantization-representation-generation pipeline. While most research focuses on quantization and generation, the representation that bridges the two remains underexplored. Existing representation constructions mainly take two forms: either directly flattening the SID code embeddings into a sequence or merging them into a compact vector via external networks. While effective, they still expose two practical challenges: \textbf{1) Loss of item semantics:} Information lost by quantization or compression needs to be recovered during the representation stage. \textbf{2) Insufficient modeling of structured SID relations:} Intra-item code composition patterns and inter-item prefix-transition patterns are not explicitly preserved for representation and subsequent SID generation. To overcome these challenges, we propose \textbf{ComeIR}, a \textbf{\uline{Co}}nditional \textbf{\uline{me}}mory Enhanced \textbf{\uline{I}}tem \textbf{\uline{R}}epresentation framework that reconstructs SID code embeddings into item-aware inputs and preserves structured SID evidence for prediction. Item-guided Token Scoring estimates the contribution of each SID code using the item embedding employed by the quantizer. Dual-level Conditional Memory captures intra-item prefix composition and inter-item prefix transitions, while Memory Merge integrates these signals into each historical item representation. The Memory-restoring Prediction Head further conditions SID-level prediction on the generated prefix and historical transition evidence. Extensive experiments demonstrate the effectiveness and flexibility of ComeIR, and further reveal scalable gains from enlarging conditional memory.
Workloads such as query rewriting, deep research and group-relative policy optimization (GRPO) issue batches of semantically related vector queries. Existing disk-resident approximate nearest neighbor (ANN) systems process these queries independently. On three workloads, 31 to 67 percent of the I/Os of a batch fetch a page that another query of the batch has already fetched. A page cache shared by the batch removes these I/Os, yet it makes a single-thread search only $1.1$ to $1.2{\times}$ faster, because the search time goes to distance computation and the cache does not change the walk of any query. We present CoWalk, which runs the queries of a batch as cooperating walks on one thread. Every query keeps its own candidate list, a page is read once per batch, and the full-precision vector on every fetched page is ranked by every query of the batch. A query therefore reaches the target recall with a shorter candidate list, which computes fewer distances and issues fewer I/Os. At recall@10 of 0.95, CoWalk answers a batch on one thread $1.6$ to $1.9{\times}$ faster than the fastest per-query baseline and issues $2.8$ to $6.9{\times}$ fewer I/Os. On a server it serves $3.0$ to $6.9{\times}$ the throughput of that baseline at a similar batch latency.
Generative recommendation (GR) reformulates item retrieval as autoregressive generation of discrete semantic identifiers (SIDs), offering a unified paradigm for personalized recommendation. GR4AD, as an industrial generative recommendation framework, has greatly improved advertising revenue while enabling efficient real-time serving in Kuaishou.However, continued deployment has exposed three limitations: deterministic SID assignment remains sensitive to semantic boundaries; conventional backbone scaling provides diminishing revenue gains under strict serving constraints; and sequence-level value optimization provides limited credit assignment to intermediate SID decisions, causing likelihood-based beam search to prematurely prune potentially high-value candidates. We present GR4AD-X, which addresses these limitations through reliable SID construction, serving-efficient capacity scaling, and unified value alignment. Specifically, GR4AD-X improves SID reliability through Multi-Path assignment. It expands model capacity along multiple complementary dimensions rather than relying solely on wider or deeper backbones, achieving a better effectiveness-efficiency trade-off under strict serving constraints. Finally, Unified Value-Guided Generation and Discrimination (UniVGD) aligns generation with commercial value through prefix-level reward optimization, value-guided beam pruning, and discriminative reranking within a unified model. Large-scale online A/B tests on the Kuaishou advertising platform demonstrate that GR4AD-X achieves a cumulative revenue improvement of 3.07% over the deployed GR4AD baseline while satisfying strict latency and throughput constraints.
Gangzhen Qian, Kainan Zhou, Chuhong Xu, Hang Xiao, Zhi Jing, Rong Bao, Lu Yi, Zhaoyi Li
The external sources retained in our non-systematic audit did not isolate the incremental security or utility effect of consuming /llms.txt while ordinary retrieval remained fixed. The initial audit covered 39 sources through August 16, 2026, with later targeted updates through October 14. A purposive 15-tool study observed no automatic requests in 45 trials at one endpoint; a non-probability log study found that 97% of domains with detected files received no recorded requests in May 2026. We separate publication, fetch, model use, and action; define five operational threat paths and paired security-utility estimands; and exercise deterministic pre-model controls with 30 author-defined parser fixtures. We also execute a reduced pilot: 24 tasks across eight sites, five conditions, and 120 real model responses, with ordinary evidence held fixed. Frozen automatic scoring identifies attacker-target adoption in $21/24$ raw adversarial responses, with one further response unresolved; the adversarial URLs-only condition has no known adoptions and one unresolved response. The pilot measures output changes conditional on supplied evidence, without live retrieval or external actions. Together, the framework and pilot distinguish manifest-content effects from publication, discovery, and downstream execution, while leaving full retrieval and agent evaluation open.
Dataset search remains a bottleneck in data analysis workflows, as existing systems largely rely on keyword matching or shallow retrieval that fail to capture user intent. To facilitate intent-driven search, we present two retrieval components to equip multi-faceted dataset search interfaces through natural language. First, we propose Hypothetical Schema Embeddings (HySE) that uses a language model to generate task-aligned hypothetical table schemas which are used for embedding-based retrieval, bridging the gap between intent and table content. We introduce KaggleDS, a corpus comprising 6,503 tables and diverse task-queries derived from Kaggle, and show that HySE consistently improves recall over strong keyword and embedding-based baselines, as well as over a dense table retriever fine-tuned on KaggleDS. We also show that HySE generalizes to end-to-end question answering on, for example, the NQ-Tables dataset, a corpus of 169,898 Wikipedia tables with human-written queries, where HySE improves Recall@10 by +6.0 pp over dense retrieval baselines. Second, we introduce FlexPrune, which translates dataset constraints expressed in natural-language into SQL predicates over a customizable metadata index, enabling filtering across arbitrary metadata attributes. Combined with HySE, FlexPrune increases Recall@10 by +6.0 pp over dense retrieval followed by tag-based filtering. Our results show that natural-language-based retrieval supporting task and metadata queries offers a viable alternative to schema-level keyword search in dataset search.
Generative engine optimization needs a scorer that ranks candidate pages by the share of an answer they will receive as citations. We measure pretrained rerankers, a hosted typed decision model (Jev), eight open decision models and distilled students on GEO-Bench queries answered by ten engines that receive the same candidate sources in their prompt. Where several engines answer the same query, they agree with one another at a within-group Spearman correlation of 0.704. Jev and two open 27B decision models, used without training, are the strongest scorers at 0.575 to 0.578. A Qwen3-Reranker-4B fine-tuned with low-rank adapters on teacher labels for other GEO-Bench queries, released as GEO Citation Ranker 4B, reaches 0.542 to 0.549 across three seeds, and its differences with the 27B models and with Jev are not distinguishable from zero on 229 queries. Most of the remaining distance between the student and Jev disappears when Jev scores one passage at a time, as the student does. Students trained on teacher labels and on observed engine citations reach the same level, and labels from Jev alone match the two-teacher target at 0.6B. We release the model, the benchmark, the predictions of every reported scorer and the code.
LLM-based generative recommendation reads long, text-rich interaction histories but generates only short semantic identifiers (SIDs). This asymmetry makes supervised fine-tuning inefficient: encoding the history is expensive, yet each history usually provides supervision for only one target. We present Merge and Branch (M&B) to reduce input cost and learn more from each encoded history. Merge compresses each historical item's SID and text into a small number of learned tokens. Branch uses the compressed history to predict multiple target SIDs. To share history computation while keeping these targets independent, M&B organizes each example as a shallow, wide tree. We develop tree FlashAttention to execute this layout efficiently, processing the shared history once and propagating gradients from every branch for end-to-end training. A tree-structured key/value cache extends history sharing to constrained decoding. Experiments on public benchmarks show that M&B outperforms strong recommendation baselines while substantially improving training and decoding efficiency.
Carlos Peláez-González, Francisco Herrera, Andrés Herrera-Poyatos
Large Language Models (LLMs) are increasingly used as recommendation generators in conversational recommender systems (RS) via retrieval-augmented generation (RAG). This flexible design introduces new sources of structural bias. This paper studies positional bias in top-1 and single-turn LLM-based RS: systematic dependence of the recommended item on the order in which candidates are presented in the prompt. We propose two complementary detection methodologies: (i) POB-ICM, based on contingency matrix analysis and Chi-square independence testing, and (ii) POB-UNBP, a sample-efficient approach derived from a uniformity null model under permutation-based evaluation, enabling stochastic decoding with formal statistical guarantees. Using a controlled RAG testbed for product recommendation, built on a real product catalog, a fixed prompt template and 2000 template-generated user queries with 20 permutations each (40,000 instances), both methodologies reject the null hypothesis of positional independence and uniformity in all evaluated state-of-the-art models. We introduce three complementary metrics -- permutation sensitivity, positional bias, and strong positional bias -- to quantify this effect, the latter targeting cases where the position displaces the recommendation towards clearly less relevant items rather than only choosing among comparably relevant ones. A robustness analysis across candidate-set size and decoding temperature shows that the bias is structural and model-dependent, not an artifact of deterministic decoding or candidate-set size. These results establish positional bias as a measurable, reproducible property of prompt-mediated recommendation pipelines in the evaluated setting and motivate routine auditing and governance controls (e.g., order randomization, standardized prompting) for deployed LLM-based RS. Source code is publicly released.
Food similarity underlies retrieval, recommendation, substitution, and other food computing tasks, yet food representations are typically evaluated by identity preservation. We ask what happens to food similarity beyond identity: how different visual representations organize foods into relational neighborhoods, what culinary relationships these neighborhoods reveal or obscure, and how consistently these structures emerge across representations. Indian cuisine provides a challenging setting in which visually similar foods may have different culinary identities, while foods within a culinary family can vary substantially in appearance. Using an expanded Khana dataset and seven pretrained visual representations spanning convolutional, transformer-based, self-supervised, and image-text aligned models, we evaluate image-to-image retrieval and image-to-dish neighborhoods, examining culinary-family structure, contrast cases, and cross-representation agreement. We also aggregate image representations to compare dish-level visual similarity across models and against recipe-derived similarity for the same dishes. Image-level neighborhoods vary substantially across representations, including similarities across culinary-family boundaries. At the dish level, aggregation reveals common visual structure despite local variation, but this structure corresponds less strongly with recipe-derived similarity. These findings show that recognition performance alone cannot characterize relational structure: food similarity depends on representation, modality, and abstraction level rather than being a fixed property of the food collection. Understanding these structures is therefore a prerequisite to combining multiple food information modalities into a coherent notion of similarity.
Abhinav Goel, Agostino Capponi, Alfio Gliozzo, Chaitya Shah
AI agents can now write software from a plain-English description, but code that runs is not the same as code that does what was asked. This gap matters most for smart contracts, the programs behind Web3 applications that move money on a blockchain and cannot be fixed once deployed. We ask how to check whether an AI-written contract behaves as it should, and where the requested behavior gets lost. We design an evaluation built on execution rather than opinion. We replay the same transactions on an AI-written contract and on the original it should reproduce and measure how often they agree. We test every grading tool on planted mistakes, compare against an answer key written without the description, and audit our own scorer until known wrong answers cannot pass. Common methods, such as AI judges and standard analysis tools, miss obvious mistakes, and scoring choices matter too. Checking only the functions every system includes hid about half the difference between pipelines. We then propose an elicitation-first architecture. Before writing code, the agent asks the client a short checklist of questions, checks the contract against an independently written executable answer key, and returns disagreements to the client instead of silently rewriting code. Asking first matters far more than reviewing or patching afterwards, across three AI models, while automated security fixes often break requested behavior. A few well-chosen questions recover much of what a short description omits, and their value can be estimated without the original contract. To our knowledge, this is the first study in smart contract synthesis to gather missing requirements before generation and check the result against an independently written executable model. The lessons extend to any setting where AI agents turn requests into programs. We release our tools, 600 tasks and all results.
Retrieval-augmented generation (RAG) lets a large language model (LLM) answer questions by retrieving relevant passages from a large corpus. However, RAG struggles with multi-hop questions, which require chaining facts spread across several documents through intermediate entities that the question never mentions. Graph-based RAG methods retrieve once from the entities a question names and so rarely recover the full evidence chain. Iterative RAG methods recover it step by step, but their accuracy is limited by the many tasks they ask the LLM to perform, such as planning the sub-questions, deciding their dependencies, rewriting them as answers arrive, and keeping track of what is known. We propose DeFT-RAG, a RAG framework built on three components. First, DeFT-RAG generates one self-contained sub-question at a time from the original question and the sub-questions answered so far with their answers (QA pairs). Each sub-question names every entity it asks about, which removes the need to plan, track dependencies or rewrite, and lets the LLM focus on a single fine-grained task at a time. Second, DeFT-RAG moves the remaining LLM work outside the LLM: rule-based updates to a QA table that keeps track of what is known, and yes/no decisions made by a fine-tuned lightweight decision model. Third, DeFT-RAG caches the QA pairs and reuses an answer when a later question asks a sub-question with the same meaning. On MuSiQue, 2WikiMultiHopQA and HotpotQA, DeFT-RAG achieves the best accuracy among thirteen baselines with both gpt-4o-mini and Qwen3.6 as the backbone, while giving the LLM fewer input tokens than any other iterative method. With a QA cache shared across 1,000 MuSiQue questions, 23.1% of sub-questions are answered from the cache, which reduces LLM calls by 12.9% without lowering accuracy.
We present FunnelAL, a retrieve-then-rank active learning system for single-class discovery, which adapts the multi-stage funnel architecture of industrial recommender systems to data annotation. Starting from one positive and one negative example, FunnelAL repeats three stages: (1) embedding-based retrieval that narrows the corpus to a candidate set, (2) a ranking stage that exploits a learned ranker (RankNet) and, once batch precision falls, mixes in committee-based exploration (QBC), and (3) the annotator's labels, which refine both stages for the next round. We evaluate on five datasets, three benchmarks and two large corpora of up to 2.7 million images, each with two different pretrained encoders, DINOv2 and CLIP. With clean labels and DINOv2, FunnelAL has the best mean rank over the five datasets, and with CLIP it shares the best rank with its ablation Funnel-RankNet. On the three benchmarks it discovers positives significantly faster than the classical and uncertainty-based baselines and than PF-MA, and no significant difference from PF-MA's final F1 was detected. With DINOv2, no significant difference from SEALS's active-search mode (most-likely-positive selection) was detected on most datasets. With CLIP, FunnelAL discovers significantly faster than it on three of the five. On iNaturalist's working set of 2.15 million images FunnelAL selects eight times faster than the neighborhood baselines as we implemented them (SEALS in both its modes and AnchorAL, which select only among the nearest neighbors of labeled items). There, PF-MA and uncertainty sampling, which label items near the decision boundary across the whole pool, reach a higher F1 under our evaluation protocol. Under 10% annotator label noise it has a higher F1 than every baseline on three of the five datasets with DINOv2 and on all five with CLIP, and SEALS's active-search mode is more robust at higher rates.
Nesta Midavaine, Floris Fok, Paulo Mendes, Joao Silva, Önder Akaçık, Dimas Lima, Zülküf Genç
We present LCM-Promo, an LLM-based uplift model for promotion allocation. LCM-Promo is a task-specific Large Commerce Model (LCM): a finetuned LLM trained on transactional commerce data to predict user intent and promotion response. It learns monotonic promotion-response curves from production logs by interpreting single-token outputs as conversion-threshold probabilities, while conditioning on session context such as day and delivery shift. In a production A/B test on churn reactivation, LCM-Promo increased conversion by 1% and reduced promotion cost by 6% relative to the Segment-EMA production baseline.
Zhenye Yang, Jinpeng Chen, Wenbo Fu, Jun Ma, Huan Li, Qimin Zhou, Tao Wang, Li Sun 等 9 位作者
Multimodal Recommender Systems (MMRS) leverage rich item features to mitigate data sparsity, yet similarity-driven graph construction and propagation can favor semantically similar substitutes over functionally related complements. When behavioral associations connect semantically distant items, complementary signals can remain geometrically isolated, yielding fragmented "Similarity Islands" and semantic drift when standard propagation is applied. To overcome these limitations, we propose MAGE (Manifold-Augmented Graph Embedding), a framework for personalized multimodal ranking that decouples complementarity modeling into offline geometric warping and online topological propagation. The offline phase enhances item representations by extracting complement-aware residual directions via a Soft Conceptor mechanism and injecting them under a local gap-preserving constraint, incorporating co-purchase evidence while controlling the magnitude of geometric perturbation. Synergistically, the online phase employs an asymmetric dual-stream GNN that treats visual features as stable anchors and augmented textual features as flexible probes, fused through a self-calibrated cross-modal mechanism to reconcile shared semantics and modality-specific residual cues. Extensive experiments on four Amazon datasets show that MAGE yields consistent improvements over strong multimodal recommendation baselines. Together with the supporting analyses in the appendix, these results suggest that geometry-aware augmentation can alleviate the semantic mismatch that arises in complementary recommendation settings. Our code is available at https://github.com/Alex-bupt/mage.
Retrieval-augmented generation (RAG) can benefit from more sophisticated retrieval, but extra retrieval may be unnecessary for many questions and gains in evidence retrieval may not translate into faithful or correct answers. We study a cost-aware, two-action retrieval policy that chooses between a dense-retrieval-plus-reranking baseline and a hybrid lexical--dense retriever. Rather than classifying baseline retrieval as simply successful or unsuccessful, we estimate the question-specific incremental evidence-retrieval benefit of hybrid retrieval using features available before the routing decision. A regularized linear model and decision threshold are selected using document-grouped cross-validation and frozen prior to held-out evaluation. On 60 previously unseen QASPER training-split articles (188 questions, 169 evidence-evaluable), the policy triggers hybrid retrieval on 46 questions (24.5\%) and obtains the same evidence Hit@10 as always-hybrid retrieval (78.70\%, versus 76.92\% for always-keep). Paired, paper-cluster bootstrap intervals do not establish a reliable utility advantage. We further compare generated answers from a 4-bit Qwen3-4B model under the three retrieval policies. Provisional answer-level assessments show no clear correctness or faithfulness improvement after partial review, despite the evidence-retrieval result. The generation assessments remain model-assisted and lack independent human validation. These results expose an important distinction between retrieving relevant evidence efficiently and improving the reliability of the generated answer. The work is presented as a reproducible empirical study, not a claim of statistically established answer-quality superiority.
Shangxin Yang, Zongwei Wang, Jing Long, Junliang Yu, Min Gao
Semantic ID-based recommendation represents each item as a sequence of semantic tokens obtained through residual quantization. This paradigm generates the next item's tokens autoregressively from coarse to fine, entailing each prediction attending to historical tokens at all semantic levels. However, our inference-time analysis reveals that historical tokens at the target-token's semantic level tend to be more informative, whereas other levels can introduce interference. Motivated by these findings, we propose LESSRec that separately encodes the historical token sequence at each semantic level and restricts attention over historical tokens to the sequence at the target token's semantic level during prediction. To complement these level-specific histories, LESSRec explicitly provides an item-level semantic sequence that represents each historical item's overall semantics in chronological order. The decoder accesses the level-aligned token sequence and the item-level semantic sequence through two separate cross-attention branches and fuses their contexts. We further introduce LESSRec+, which uses separate encoders for different semantic-level histories with the same total parameter count to reduce encoder computation. Extensive experiments across three public datasets and three tokenizers demonstrate the recommendation effectiveness of LESSRec and LESSRec+, with efficiency analyses highlighting the computational benefits of LESSRec+.
Mingzhe Lu, Jian Qin, Ruize Qin, Jinhao Liu, Yunpeng Li, Hang Su
Narrative-driven recommendation allows users to express multifaceted needs through free-form text where stated preferences naturally qualify and condition one another. Existing methods typically treat narratives as flat keyword combinations or unconstrained generative queries, failing to capture how complex preferences hierarchically refine and contextualize each other. We propose the Preference Dependency Graph Framework (PDG-F), a narrative-driven recommendation framework centered on explicit preference dependencies. Specifically, PDG-A (PDG-Agent) first parses unstructured requests to organize user preferences and their mutual qualifications into a directed acyclic Preference Dependency Graph (PDG). Through this graph, candidate evidence from the open Web is acquired via directed path retrieval, and final recommendations are determined by topological path ranking. Comprehensive experiments across three diverse domains spanning English and Chinese contexts confirm that PDG-F consistently achieves superior recommendation performance over competitive baselines, open-domain retrieval agents, and frontier language models. Further diagnostic analyses show that our dependency-guided retrieval markedly raises the attainable candidate-pool ceiling, demonstrating the fundamental value of modeling preference dependencies in narrative-driven recommendation.
Text-based Person Anomaly Search (TPAS) retrieves a target pedestrian using natural-language descriptions of appearance and normal or anomalous behavior. Unlike conventional text-based person retrieval, TPAS must preserve identity cues while distinguishing fine-grained behaviors. It presents two challenges: same-identity, different-behavior hard negatives cause gradient conflict during granular alignment, while strong appearance cues can obscure weak behavioral evidence during aggregation. We propose GradCO (behavior-aware Gradient-Conflict Offloading), comprising three complementary components. Behavior-Aware Mixture-of-Experts (B-MoE) forms appearance, normal-behavior, and anomalous-behavior streams whose roles are learned from retrieval objectives without expert-specific semantic labels. Conflict Offload Learning (COL) uses pair-conditioned offload anchors in similarity computation to reduce the conflicting-gradient burden on the main text representations. Hierarchical Intent Alignment (HIA) aggregates expert similarities and features using query-conditioned weights. Experiments on five TPAS/TIPR benchmarks show GradCO significantly surpasses SOTA methods and validate our design.