As global e-commerce platforms continue to expand, companies are entering new markets where they encounter cold-start challenges due to limited human labels and user behaviors. In this paper, we share our experiences in Coupang to provide a competitive cold-start performance of relevance matching for emerging e-commerce markets. Specifically, we present a Cold-Start Relevance Matching (CSRM) framework, utilizing a multilingual Large Language Model (LLM) to address three challenges: (1) activating cross-lingual transfer learning abilities of LLMs through machine translation tasks; (2) enhancing query understanding and incorporating e-commerce knowledge by retrieval-based query augmentation; (3) mitigating the impact of training label errors through a multi-round self-distillation training strategy. Our experiments demonstrate the effectiveness of CSRM-LLM and the proposed techniques, resulting in successful real-world deployment and significant online gains, with a 45.8% reduction in defect ratio and a 0.866% uplift in session purchase rate.

A long-standing challenge for pushing sensor-based human activity recognition (HAR) to industrial usage is the distribution shift between training data and testing data: significant variations in data distribution lead to a notable decline in performance. Recently, Large Language Models (LLMs) have demonstrated exceptional generalization capability, which provides a new opportunity to mitigate the distribution shift problem of HAR. However, since LLMs are inherently designed and trained on textual data, their potential to enhance generalization in HAR applications remains an open question. In this paper, we introduce LLM4HAR, a novel LLM-based model to improve cross-domain HAR. LLM4HAR consists of three main modules: (i) the Sensor Data Adaptation module, which aligns IMU signals with LLMs via sensor embedding(ii) the Sensor Knowledge Learning module, which injects sensor knowledge into LLMs for activity recognition, and (iii) the Efficiency Enhancement module, which employs a partial training strategy and reduces the model size by more than 10 times. Extensive evaluations show that LLM4HAR outperforms the existing methods by 13.82% in average F1 score, demonstrating the feasibility and effectiveness of transferring knowledge from pretrained LLMs to enhance HAR. Further, LLM4HAR has been adopted by JD Logistics to support downstream applications such as Courier Welfare Improvement and Map Data Generation.

Last-mile Delivery Parcel Volume (LDPV) quantifies the number of parcels destined for a specific region, particularly a manually divided Area-Of-Interest (AOI). Accurate prediction of AOI-level LDPV is crucial for the efficient management of logistics resources. However, the straightforward adaptation of existing prediction models often falls short, primarily due to (I) a lack of consideration for the intuition behind AOI divisions, and (II) a reliance solely on fully observed historical data, which may not inform future trends. To overcome the above pitfalls, leveraging rich AOI data and advanced parcel travel time estimation services in JD Logistics, this paper introduces a novel framework called Dual-view Prediction Networks (DualPNs). It combines a Vector-Quantified AutoEncoder (VQ-AE) and a Template-Augmented Zero-Inflated Poisson (TA-ZIP), enabling both point and probabilistic distribution predictions of AOI-level LDPV. Specifically, VQ-AE utilizes a vector quantization technique to distill a large number of AOIs into representative templates, thereby addressing the first pitfall. Subsequently, TA-ZIP dynamically integrates fully observed and lookahead features, aligning them with template-specific decoders to parameterize the probabilistic distributions, thus resolving the second pitfall. We conduct extensive experiments in two cities, comprising over 47,000 and 126,000 AOIs respectively, to demonstrate the superiority of our DualPNs over other baselines. Moreover, a real-world case study highlights the effectiveness of DualPNs for enhancing downstream courier allocation by yielding an average improvement of 1.51% in the on-time delivery rate.

The$V$-matrix Support Vector Machine (VSVM) is an innovative machine learning method recently proposed by Vapnik and Izmailov, which integrates positional relationships among training samples into the model learning, yielding the decision via conditional probability. But it overlooks the distribution information hidden in the data which plays a pivotal role in the training process and neglects the utilization of testing samples. To fully exploit the distribution information of the data, this paper proposes a novel Distribution Metric Based$V$-matrix Support Vector Machine (DVSVM) building upon VSVM. DVSVM incorporates the distributional information implicit in the data by measuring the distances between samples using the Wasserstein distance. Compared to VSVM, it also additionally accounts for the positional relationships of testing samples. It is further theoretically proved that VSVM can degenerate from DVSVM under certain conditions. Experimental results on several synthetic datasets and real-world disease datasets demonstrate the superiority of DVSVM.

Domain adaptation has found widespread applications in real-life scenarios, especially when the target domain has limited labeled samples. However, most of the domain adaptation models only utilize one type of knowledge from the source domain, which is usually achieved by strong mode of convergence. To fully incorporate multiple knowledge from the source domain, for binary classification, this paper studies a novel learning paradigm for Domain Adaptation via Learning Using Statistical Invariant by simultaneously combining the strong and weak modes of convergence in a Hilbert space. The strong mode of convergence undertakes the mission of learning a least squares probability output binary classification task in a general hypothesis space, while the weak mode of convergence integrates diverse knowledge by constructing meaningful statistical invariants that embody the concept of intelligence. The utilization of weak convergence shrinks the admissible set of approximation functions, and subsequently accelerates the learning process. In this paper, several statistical invariants that represent sample, feature and parameter information from the source domain are constructed. By taking an appropriate statistical invariant, DLUSI realizes some existing methods. Experimental results on synthetic data as well as the widely used Amazon Reviews and 20 News data demonstrate the superiority of the proposed method.

Efficient order assignment in last-mile delivery benefits customers, couriers, and the platform. State-of-the-practice order assignment is based on the static delivery area partition, which cannot adapt well to the dynamic order quantity and destination distributions on different days. State-of-the-art methods focus on balancing order amounts or the payoff among couriers dynamically, neglecting the courier's workload in delivering orders. This paper explores the courier's heterogeneous behaviors for delivering orders to different destinations to measure the courier's workload and then achieve more efficient order assignments under the fair workload constraint. We design a workload-constrained order assignment system, calledWORD, to reduce the cost of the last-mile delivery, i.e., the couriers’ total travel distance and overdue order rate. Specifically, the heterogeneous behaviors for delivering orders are first utilized for workload calculation. Then a two-stage order assignment framework is designed, including a sort-based initialization algorithm for initializing the assignment under the fair workload constraint and a coalition-game-based improvement algorithm for improving the assignment. Extensive evaluation results with real-world logistics data from one of the largest logistics companies in China show thatWORDreduces the cost of the order assignment by up to 51.9% under the fair workload constraint compared to the state-of-the-art methods.

Order Transfer from the transfer center to delivery stations is an essential and expensive part of the logistics service chain. In practice, one vehicle sends transferred orders to multiple delivery stations in one transfer trip to achieve a better trade-off between the transfer cost and time. A key problem is generating the vehicle’s route for efficient order transfer, i.e., minimizing the order transfer time. In this paper, we explore fine-grained delivery station features, i.e., downstream couriers’ remaining working times in last-mile delivery trips and the transferred order distribution to design a Prediction-and-Scheduling framework for efficient Order Transfer called PSOT, including two components: i) a Courier’s Remaining Working Time Prediction component to predict each courier’s working time for conducting heterogeneous tasks, i.e., order pickups and deliveries, with a context-aware location embedding and an attention-based neural network; ii) a Vehicle Scheduling component to generate the vehicle’s route to served delivery stations with an order-transfer-time-aware heuristic algorithm. The evaluation results with real-world data from one of the largest logistics companies in China show PSOT improves the courier’s remaining working time prediction by up to 35.6% and reduces the average order transfer time by up to 51.3% compared to the state-of-the-art methods.

On-demand delivery has become an increasingly popular urban service in recent years as it facilitates citizens’ daily lives significantly. In the fulfillment cycle, the order preparation time estimation is extremely important and can be used for many applications, such as improving order dispatching and fulfillment time estimation. Existing work is generally based on high-cost physical devices or large-scale labeled training data, which are not feasible in on-demand delivery services. We solve this problem based on already collected different kinds of data from the on-demand delivery platform, e.g., the courier’s reported arrival time to the merchant. Our intuition is that the couriers’ reported time implicitly reflects the order preparation time, which leads to a challenge: complicated correlations between the couriers’ reported arrival time and the order preparation time. To solve this challenge, we design an order preparation time inference framework OPTI, which first constructs a self-supervised classification task based on the couriers’ reported arrival time to infer the coarse-grained order preparation time and then exploits semi-supervised learning to transfer the coarse-grained time to fine-grained time inference. Experimental results show that OPTI can improve the accuracy of inference by 5% to 17% compared to the state-of-the-art solutions.

The process industry implements important maintenance strategies on the basis of the health status of system operation to reduce engineering maintenance cost and concurrently ensure reliability and safety. This study presents an architecture for knowledge-based prognostics and health management (K-PHM) by applying the trapezoidal interval type-2 fuzzy linguistic term sets. The proposed K-PHM methodology successfully provides operational information for diagnostics, prognostics, and subsequent actionable knowledge for health management in subjective decision making. It emphasizes knowledge-driven maintenance strategies to increase system reliability and safety within linguistic decision making. The main objective of this study is to focus on the uncertainty within the knowledge formation derived from data and subsequent information expressed with linguistic terms generated from domain experts in the K-PHM methodology. Fuzzy sets and fuzzy logic reasoning techniques are used in conjunction to enhance the capability of handling uncertainty throughout the PHM process. The effectiveness of the proposed K-PHM is verified with the authentic case study for the predictive maintenance of a chemical pumping system against abnormal vibration.

We study the problem of query attribute value extraction, which aims to identify named entities from user queries as diverse surface form attribute values and afterward transform them into formally canonical forms. Such a problem consists of two phases: named entity recognition (NER) and attribute value normalization (AVN). However, existing works only focus on the NER phase but neglect equally important AVN. To bridge this gap, this paper proposes a unified query attribute value extraction system in e-commerce search named QUEACO, which involves both two phases. Moreover, by leveraging large-scale weakly-labeled behavior data, we further improve the extraction performance with less supervision cost. Specifically, for the NER phase, QUEACO adopts a novel teacher-student network, where a teacher network that is trained on the strongly-labeled data generates pseudo-labels to refine the weakly-labeled data for training a student network. Meanwhile, the teacher network can be dynamically adapted by the feedback of the student's performance on strongly-labeled data to maximally denoise the noisy supervisions from the weak labels. For the AVN phase, we also leverage the weakly-labeled query-to-attribute behavior data to normalize surface form attribute values from queries into canonical forms from products. Extensive experiments on a real-world large-scale E-commerce dataset demonstrate the effectiveness of QUEACO.

Sequence labeling aims to predict a finegrained sequence of labels for the text.However, such formulation hinders the effectiveness of supervised methods due to the lack of token-level annotated data.This is exacerbated when we meet a diverse range of languages.In this work, we explore multilingual sequence labeling with minimal supervision using a single unified model for multiple languages.Specifically, we propose a Meta Teacher-Student (MetaTS) Network, a novel meta learning method to alleviate data scarcity by leveraging large multilingual unlabeled data.Prior teacher-student frameworks of self-training rely on rigid teaching strategies, which may hardly produce high-quality pseudo-labels for consecutive and interdependent tokens.On the contrary, MetaTS allows the teacher to dynamically adapt its pseudoannotation strategies by the student's feedback on the generated pseudo-labeled data of each language and thus mitigate error propagation from noisy pseudo-labels.Extensive experiments on both public and real-world multilingual sequence labeling datasets empirically demonstrate the effectiveness of MetaTS 1 .

On-demand delivery has become an increasingly popular urban service in recent years as it facilitates citizens' daily lives significantly. Different from traditional logistics services, e.g., FedEx and UPS, on-demand delivery is expected to be completed within a relatively short time, e.g., 30 minutes to 1 hour. In the fulfillment cycle, the order preparation time estimation is extremely important, which can be used for many applications such as improving order dispatching and fulfillment time estimation. Existing work is generally based on high-cost physical devices or large-scale labeled training data, which are not feasible in on-demand delivery services. We solve this problem based on already collected different kinds of data from the on-demand delivery platform, e.g., the courier's reported arrival time to the merchant. Our intuition is that couriers' reported time implicitly reflects the order preparation time, which leads to a challenge: complicated correlations between the couriers' reported arrival time and the order preparation time. To solve this challenge, we design an order preparation time inference framework OPTI, which first constructs a self-supervised classification task based on the couriers' reported arrival time to infer the coarse-grained order preparation time, and then exploits semi-supervised learning to transfer the coarse-grained time to fine-grained time inference. We implement and evaluate OPTI in the [anonymous] platform, which is one of the largest on-demand delivery platforms in the world. Experimental results show that OPTI can improve the accuracy of inference by 5% to 17% compared to the state-of-the-art solutions.

We study the problem of semantic matching in product search, that is, given a customer query, retrieve all semantically related products from the catalog. Pure lexical matching via an inverted index falls short in this respect due to several factors: a) lack of understanding of hypernyms, synonyms, and antonyms, b) fragility to morphological variants (e.g. "woman" vs. "women"), and c) sensitivity to spelling errors. To address these issues, we train a deep learning model for semantic matching using customer behavior data. Much of the recent work on large-scale semantic search using deep learning focuses on ranking for web search. In contrast, semantic matching for product search presents several novel challenges, which we elucidate in this paper. We address these challenges by a) developing a new loss function that has an inbuilt threshold to differentiate between random negative examples, impressed but not purchased examples, and positive examples (purchased items), b) using average pooling in conjunction with n-grams to capture short-range linguistic patterns, c) using hashing to handle out of vocabulary tokens, and d) using a model parallel training architecture to scale across 8 GPUs. We present compelling offline results that demonstrate at least 4.7% improvement in [email protected] and 14.5% improvement in mean average precision (MAP) over baseline state-of-the-art semantic search methods using the same tokenization method. Moreover, we present results and discuss learnings from online A/B tests which demonstrate the efficacy of our method.

We consider an L-user multiple access channel where transmitter m has access to the linear equation um= ⊕l=1Lfmlwlof independent messages wl∈ Fpkl with fml∈ Fp, and the destination wishes to recover all L messages. This problem may be motivated as the last hop in a network where relay nodes employ the compute-and-forward strategy and decode linear equations of messages; we seek to do the reverse and extract messages from sums over a multiple access channel. In particular, we exploit the particular form of dependencies between the equations at the different relays to improve the reliable communication rates beyond those achievable by simply forwarding all equations to the destination independently. The presented achievable rate region for the discrete memoryless channel model is shown to be capacity for the additive white Gaussian noise channel.

Full-duplex allows for the simultaneous flow of information in two directions and, in point-to-point Gaussian two-way channels, doubles capacity. A two-way line network where two sources exchange messages through multiple serial relays is considered. It is shown that when all nodes are full duplex, one may achieve to within a constant gap, independent of the number of relays, of the capacity of two one-way line networks. This shows that, even in the presence of relays that carry information in two directions, full duplex is able to approximately double capacity. A novel lattice coding scheme is developed for the two-way line network with two relays, which may be extended to an arbitrary number of relays and to half-duplex scenarios. The key technical contribution is the achievability strategy, where each relay decodes the sum of several signals (using lattice codes) and then re-encodes it into another lattice codeword. This allows other nodes to again decode sums of codewords. The presented lattice-coding-based scheme ensures that both directions simultaneously fully utilize the relays' powers, even for asymmetric channels. The symmetric rate achieved by the proposed scheme is within 0.5 log 5 bit/Hz/s of the symmetric rate capacity regardless of the number of relays.

We consider a three user multiple access channel where transmitter m has access to the linear equation um= Σ3l = 1fmlwlof independent messages w1ϵ Fpk1, w2 ϵ Fpk2, w3 ϵ Fpk3(and fmlϵ Fp), and the destination wishes to recover all three messages. This problem is motivated as the last hop in a network where relay nodes employ the Compute-and-Forward strategy and decode linear equations of messages; we seek to do the reverse and extract messages from sums over a multiple access channel. An achievable rate region for the two user problem was previously derived; here we extend and strengthen this work to show capacity for the two and three user Gaussian channel models subject to invertability conditions on the matrix of coefficients describing the given linear equations of messages. The optimal transmission scheme is not to independently send the three equations umover the MAC but rather to exploit their special correlation structure.

We develop a novel lattice coding scheme for the Two-way Two-relay Channel: 1 ↔ 2 ↔ 3 ↔ 4, where Node 1 and 4 communicate with each other through two relay nodes 2 and 3. Each node only communicates with its neighboring nodes. The key technical contribution is the lattice-based achievability strategy, where each relay is able to remove the noise while decoding the sum of several signals in a Block Markov strategy and then re-encode the signal into another lattice codeword using the so-called “Re-distribution Transform”. This allows nodes further down the line to again decode sums of lattice codewords. The symmetric rate achieved by the proposed lattice coding scheme is within 1/2 log 3 bit/Hz/s of the symmetric rate capacity.

Lattice codes are known to achieve capacity in the Gaussian point-to-point channel, achieving the same rates as i.i.d. random Gaussian codebooks. Lattice codes are also known to outperform random codes for certain channel models that are able to exploit their linearity. In this paper, we show that lattice codes may be used to achieve the same performance as known i.i.d. Gaussian random coding techniques for the Gaussian relay channel, and show several examples of how this may be combined with the linearity of lattices codes in multisource relay networks. In particular, we present a nested lattice list decoding technique in which lattice codes are shown to achieve the decode-and-forward (DF) rate of single source, single destination Gaussian relay channels with one or more relays. We next present two examples of how this DF scheme may be combined with the linearity of lattice codes to achieve new rate regions which for some channel conditions outperform analogous known Gaussian random coding techniques in multisource relay channels. That is, we derive a new achievable rate region for the two-way relay channel with direct links and compare it to existing schemes, and derive a new achievable rate region for the multiple access relay channel. We furthermore present a lattice compress-and-forward (CF) scheme for the Gaussian relay channel which exploits a lattice Wyner-Ziv binning scheme and achieves the same rate as the Cover-El Gamal CF rate evaluated for Gaussian random codes. These results suggest that structured/lattice codes may be used to mimic, and sometimes outperform, random Gaussian codes in general Gaussian networks.

We consider the transmission of independent messages over a Gaussian relay network with interfering links. Using the compute-and-forward framework, relays can efficiently decode equations of the transmitted messages. The relays can then send their collected equations to the destination, which solves for its desired messages. Here, we study a special case of the inverse compute-and-forward problem: transmitting the equations to a single destination over a multiple-access channel. We observe that if the underlying messages have unequal rates, the set of possible values of an equation is constrained by the value of the other equations. We use this fact to improve the rate region for downloading equations. Interestingly, the rate region achieved over relay networks with interfering links using a combination of compute-and-forward and inverse compute-and-forward is larger than the best rate region achievable in the absence of interfering links. This verifies that interference may be used to beneficially “mix” messages over a wireless network.

We present a nested lattice-code-based strategy that achieves the random-coding based Compress-and-Forward (CF) rate for the three node Gaussian relay channel. To do so, we first outline a lattice-based strategy for the (X + Z1, X + Z2) Wyner-Ziv lossy source-coding with side-information problem in Gaussian noise, a re-interpretation of the nested lattice-code-based Gaussian Wyner-Ziv scheme presented by Zamir, Shamai, and Erez. We use the notation (X + Z1, X + Z2) Wyner-Ziv to mean that the source is of the form X + Z1and the side-information at the receiver is of the form X+Z2, for independent Gaussian X, Z1and Z2. We use this (X + Z1, X + Z2) Wyner-Ziv scheme to implement a “structured” or lattice-code-based CF scheme for the Gaussian relay channel which achieves the same rate as the Cover-El Gamal CF rate achieved by random Gaussian codebooks.