Query Accuracy Research Articles

PALM: Personalized Attention-based Language Model for Long-tail Query Understanding in Enterprise Search Systems

Enterprise search systems face significant challenges in handling long-tail queries, which constitute a substantial portion of search traffic but often receive inadequate attention in traditional systems. This paper introduces PALM (Personalized Attention-based Language Model), a novel framework designed to enhance long-tail query understanding in enterprise search environments. PALM integrates personalization capabilities with an advanced attention mechanism to improve search accuracy for infrequent queries while maintaining high performance on common queries. The framework employs a unique hierarchical architecture that combines user context, query semantics, and organizational knowledge through a sophisticated attention mechanism. The system features an innovative query embedding approach that adapts to individual user contexts while leveraging collective organizational knowledge. Extensive experiments on a large-scale enterprise dataset, comprising over 5 million queries from 50,000 users, demonstrate PALM's superior performance compared to state-of-the-art baselines. The results show significant improvements across multiple metrics, with a 17.5% increase in MAP for ultra-rare queries and a 10.4% overall improvement in NDCG@10. The framework exhibits robust performance across different organizational units and query types, making it particularly valuable for enterprise environments where query patterns are highly diverse and context-dependent. Our ablation studies confirm the effectiveness of each component in the PALM architecture, while case analyses provide insights into the framework's practical applications.

A uniformed adaptive early termination model through probabilistic feature to speed up quantization-based search

The quantization-based approaches are the effective techniques for addressing the problem of approximate nearest neighbor search. However, most of these methods typically employ a fixed number of nProbes as the termination condition for each query. This practice can lead to unnecessary query processing time because many queries can be terminated early, before completing the greedy search in the nearest nProbes subspace. To address this issue, we propose an early termination method to speed up the search processing of quantization-based methods. At the heart of our proposal is to exploit an effective probabilistic feature generated by locality sensitive hashing to learn a uniformed prediction model, allowing the termination condition for each query to be predicted adaptively. Experimental results over various real-world high dimensional datasets show that our proposal outperforms competitors in both query accuracy and efficiency, achieving an average running time reduction of up to around 7x.

Improving Search Query Accuracy for Specialized Websites Through Intelligent Text Correction and Reconstruction Models

Improving Search Query Accuracy for Specialized Websites Through Intelligent Text Correction and Reconstruction Models

GBNSS: A Method Based on Graph Neural Networks (GNNs) for Global Biological Network Similarity Search

Biological network similarity search plays a crucial role in the analysis of biological networks for human disease research and drug discovery. A biological network similarity search aims to efficiently identify novel networks biologically homologous to the query networks. Great progress has been achieved in biological network similarity searches. However, it remains a challenge to mine the biological network information fully to improve the accuracy of query results without increasing time overheads. In this study, we propose a biological network similarity search method based on graph neural networks named GBNSS, which combines topological and biological information (GO annotations) of biological networks into graph neural networks to find topologically and biologically similar biological networks in the database. Additionally, GBNSS is a topology-free biological network similarity search method with an arbitrary network structure. The experimental results on four benchmark datasets show that GBNSS outperforms the existing methods in terms of computational efficiency and search accuracy. Case studies further demonstrate that GBNSS is capable of searching similar networks in real-world biological networks.

The Sharing Model of Kindergarten Art Enlightenment Education Curriculum Resources Based on Deep Learning Algorithm

In order to meet the autonomous needs of children’s art learning, this paper studies the sharing model of kindergarten art enlightenment education curriculum resources under the deep learning algorithm. Build a kindergarten art enlightenment education curriculum resource-sharing model that includes the classification and storage layer of education curriculum resources, the metadata annotation layer of education curriculum resources and the shared business service layer. The deep learning algorithm is used in the classification and storage layer of education curriculum resources. The education curriculum resources are divided into courseware resources, case resources and media materials resources through unsupervised restricted Boltzmann machine (RBM) coding and supervised training process. The ontology model of various educational curriculum resources is constructed, and the semantic information of educational curriculum resources is described using ontology. The metadata annotation layer of education curriculum resources annotates the metadata semantics of education curriculum resources through the constructed ontology, which improves the consistency and sharing of education curriculum resources management. The implementation of resource sharing business service is based on semantic analysis and query of educational curriculum resources. Therefore, the shared business service layer adopts a semantic query method based on improved similarity algorithm, which combines semantic distance, node density, node depth and relationship type to build a semantic similarity calculation method to achieve semantic analysis and query of educational curriculum resources. The experimental results show that the model is suitable for multi-classification of shared resources in the kindergarten art enlightenment education curriculum, and the semantic query accuracy is high in the process of sharing. When the threshold is [Formula: see text] 0.5, the query effect of the sharing model in this paper is ideal, which can meet the needs of users’ autonomous learning.

Optimizing Video Queries with Declarative Clues

Video Database Management Systems (VDBMS) leverage advancements in computer vision and deep learning for efficient video data analysis and retrieval. This paper introduces the concept of user-specified Clues, allowing users to incorporate domain-specific knowledge, referred to as Clues, into query optimization. Clues are expressed as Clue types, each associated with optimization rules, and applied to queries through Clue instances. The extensible ClueVQS system we present to incorporate these ideas, optimizes queries automatically, utilizing Clues to improve processing efficiency. We also introduce algorithms to optimize queries using Clues allowing for trade-offs between speed and query accuracy. Our proposals and system address challenges such as data-dependent Clue effectiveness, limiting search space, and accuracy-efficiency trade-offs. Detailed experimental results demonstrate query speedups of up to two orders of magnitude compared to other applicable approaches, and a reduction of the query optimizer time by up to 95% while respecting user-specified accuracy constraints, showcasing the effectiveness of the proposed framework.

SIM: A fast real-time graph stream summarization with improved memory efficiency and accuracy

A graph stream composed of sequentially approaching arriving edges is commonly utilized to represent complicated structured data in interactive application systems. Since graph streams are extraordinarily vast and high velocity, efficient storage and analysis of graph streams face serious challenges. Current graph stream summarization schemes have effectively achieved the storage and management of graph streams. Unfortunately, they either cannot accomplish real-time queries or have low overall query accuracy. To this end, a novel summary structure, named Shared Interaction Matrix (SIM), is proposed for real-time queries rapidly and accurately with smaller memory. SIM is designed as a two-layer adjacency matrix with different structures to improve memory efficiency while preserving the key of heavy edges to support real-time measurement. Moreover, SIM leverages shared hash technology and an integral replacement strategy to boost insertion query speed and query accuracy. The performance of SIM is evaluated by conducting extensive experiments on the CPU and OVS platform. The experimental results show that SIM significantly enhances measurement accuracy and reduces insertion and query processing latency by 39.21%–93.50% while achieving real-time queries, compared with the state-of-the-art schemes.

DET-LSH: A Locality-Sensitive Hashing Scheme with Dynamic Encoding Tree for Approximate Nearest Neighbor Search

Locality-sensitive hashing (LSH) is a well-known solution for approximate nearest neighbor (ANN) search in high-dimensional spaces due to its robust theoretical guarantee on query accuracy. Traditional LSH-based methods mainly focus on improving the efficiency and accuracy of the query phase by designing different query strategies, but pay little attention to improving the efficiency of the indexing phase. They typically fine-tune existing data-oriented partitioning trees to index data points and support their query strategies. However, their strategy to directly partition the multi-dimensional space is time-consuming, and performance degrades as the space dimensionality increases. In this paper, we design an encoding-based tree called Dynamic Encoding Tree (DE-Tree) to improve the indexing efficiency and support efficient range queries based on Euclidean distance. Based on DE-Tree, we propose a novel LSH scheme called DET-LSH. DET-LSH adopts a novel query strategy, which performs range queries in multiple independent index DE-Trees to reduce the probability of missing exact NN points, thereby improving the query accuracy. Our theoretical studies show that DET-LSH enjoys probabilistic guarantees on query accuracy. Extensive experiments on real-world datasets demonstrate the superiority of DET-LSH over the state-of-the-art LSH-based methods on both efficiency and accuracy. While achieving better query accuracy than competitors, DET-LSH achieves up to 6x speedup in indexing time and 2x speedup in query time over the state-of-the-art LSH-based methods.

Category-Level Contrastive Learning for Unsupervised Hashing in Cross-Modal Retrieval

Unsupervised hashing for cross-modal retrieval has received much attention in the data mining area. Recent methods rely on image-text paired data to conduct unsupervised cross-modal hashing in batch samples. There are two main limitations for existing models: (1) learning of cross-modal representations is restricted to batches; (2) semantically similar samples may be wrongly treated as negative. In this paper, we propose a novel category-level contrastive learning for unsupervised cross-modal hashing, which alleviates the above problems and improves cross-modal query accuracy. To break the limitation of learning in small batches, a selected memory module is first proposed to take global relations into account. Then, we obtain pseudo labels through clustering and combine the labels with the Hadamard Matrix for category-centered learning. To reduce wrong negatives, we further propose a memory bank to store clusters of samples and construct negatives by selecting samples from different categories for contrastive learning. Extensive experiments show the significant superiority of our approach over the state-of-the-art models on MIRFLICKR-25K and NUS-WIDE datasets.

One Seed, Two Birds: A Unified Learned Structure for Exact and Approximate Counting

The modern database has many precise and approximate counting requirements. Nevertheless, a solitary multidimensional index or cardinality estimator is insufficient to cater to the escalating demands across all counting scenarios. Such approaches are constrained either by query selectivity or by the compromise between query accuracy and efficiency. We propose CardIndex, a unified learned structure to solve the above problems. CardIndex serves as a versatile solution that not only functions as a multidimensional learned index for accurate counting but also doubles as an adaptive cardinality estimator, catering to varying counting scenarios with diverse requirements for precision and efficiency. Rigorous experimentation has showcased its superiority. Compared to the state-of-the-art (SOTA) autoregressive data-driven cardinality estimation baselines, our structure achieves training and updating times that are two orders of magnitude faster. Additionally, our CPU-based query estimation latency surpasses GPU-based baselines by two to three times. Notably, the estimation accuracy of low-selectivity queries is up to 314 times better than the current SOTA estimator. In terms of indexing tasks, the construction speed of our structure is two orders of magnitude faster than RSMI and 1.9 times faster than R-tree. Furthermore, it exhibits a point query processing speed that is 3%-17% times faster than RSMI and 1.07 to 2.75 times faster than R-tree and KDB-tree. Range queries under specific loads are 20% times faster than the SOTA indexes.

Predictive and Near-Optimal Sampling for View Materialization in Video Databases

Scalable video query optimization has re-emerged as an attractive research topic in recent years. The OTIF system, a video database with cutting-edge efficiency, has introduced a new paradigm of utilizing view materialization to facilitate online query processing. Specifically, it stores the results of multi-object tracking queries to answer common video queries with sub-second latency. However, the cost associated with view materialization in OTIF is prohibitively high for supporting large-scale video streams. In this paper, we study efficient MOT-based view materialization in video databases. We first conduct a theoretical analysis and establish two types of optimality measures that serve as lower bounds for video frame sampling. In order to minimize the number of processed video frames, we propose a novel predictive sampling framework, namely LEAP, exhibits near-optimal sampling performance. Its efficacy relies on a data-driven motion manager that enables accurate trajectory prediction, a compact object detection model via knowledge distillation, and a robust cross-frame associator to connect moving objects in two frames with a large time gap. Extensive experiments are conducted in 7 real datasets, with 7 baselines and a comprehensive query set, including selection, aggregation and top-k queries. The results show that with comparable query accuracy to OTIF, our LEAP can reduce the number of processed video frames by up to 9× and achieve 5× speedup in query processing time. Moreover, LEAP demonstrates impressive throughput when handling large-scale video streams, as it leverages a single NVIDIA RTX 3090ti GPU to support real-time MOT-based view materialization from 160 video streams simultaneously.

Navigating the Maps: Euclidean vs. Road Network Distances in Spatial Queries

Using spatial data in mobile applications has grown significantly, thereby empowering users to explore locations, navigate unfamiliar areas, find transportation routes, employ geomarketing strategies, and model environmental factors. Spatial databases are pivotal in efficiently storing, retrieving, and manipulating spatial data to fulfill users’ needs. Two fundamental spatial query types, k-nearest neighbors (kNN) and range search, enable users to access specific points of interest (POIs) based on their location, which are measured by actual road distance. However, retrieving the nearest POIs using actual road distance can be computationally intensive due to the need to find the shortest distance. Using straight-line measurements could expedite the process but might compromise accuracy. Consequently, this study aims to evaluate the accuracy of the Euclidean distance method in POIs retrieval by comparing it with the road network distance method. The primary focus is determining whether the trade-off between computational time and accuracy is justified, thus employing the Open Source Routing Machine (OSRM) for distance extraction. The assessment encompasses diverse scenarios and analyses factors influencing the accuracy of the Euclidean distance method. The methodology employs a quantitative approach, thereby categorizing query points based on density and analyzing them using kNN and range query methods. Accuracy in the Euclidean distance method is evaluated against the road network distance method. The results demonstrate peak accuracy for kNN queries at k=1, thus exceeding 85% across classes but declining as k increases. Range queries show varied accuracy based on POI density, with higher-density classes exhibiting earlier accuracy increases. Notably, datasets with fewer POIs exhibit unexpectedly higher accuracy, thereby providing valuable insights into spatial query processing.

Practical and privacy-preserving geo-social-based POI recommendation

With the rapid development of location-based services and online social networks, POI recommendation services considering geographic and social factors have received extensive attention. Meanwhile, the vigorous development of cloud computing has prompted service providers to outsource data to the cloud to provide POI recommendation services. However, there is a degree of distrust of the cloud by service providers. To protect digital assets, service providers encrypt data before outsourcing it. However, encryption reduces data availability, making it more challenging to provide POI recommendation services in outsourcing scenarios. Some privacy-preserving schemes for geo-social-based POI recommendation have been presented, but they have some limitations in supporting group query, considering both geographic and social factors, and query accuracy, making these schemes impractical. To solve this issue, we propose two practical and privacy-preserving geo-social-based POI recommendation schemes for single user and group users, which are named GSPR-S and GSPR-G. Specifically, we first utilize the quad tree to organize geographic data and the MinHash method to index social data. Then, we apply BGV fully homomorphic encryption to design some private algorithms, including a private max/min operation algorithm, a private rectangular set operation algorithm, and a private rectangular overlapping detection algorithm. After that, we use these algorithms as building blocks in our schemes for efficiency improvement. According to security analysis, our schemes are proven to be secure against the honest-but-curious cloud servers, and experimental results show that our schemes have good performance.

Research and Application of Legal Consultation database based on multimedia Network

In order to relieve the pressure of people's appeal for legal inquiry, the research and application of legal consultation database based on multimedia network is proposed. The legal consultation system designed in this essay, combined with the current situation of the legal industry, takes the Chinese question answering system as the prototype, combined with the open source data retrieval project Lucene.net, expands the data storage types. In this essay, the Chinese word segmentation system developed by Chinese Academy of Sciences is integrated into Lucene.Net platform to make up for its word segmentation deficiency. And the mutual information technology is used to make the legal related words with the same meaning be searched first. When extracting answers in Chinese question answering system, there are often "missing" and "wrong" answers. In this essay a method of sentence similarity calculation of questions and answers based on Latent Semantic analysis (LSA) is proposed, which uses spatial vector model as representation method and latent semantic analysis theory. A low-dimensional semantic space is constructed by using the dimension reduction method of singular value decomposition, and the sentence similarity between question and answer is computed on the semantic space. Experimental results show that most retrievals can be completed within 20 seconds, and the average unit retrieval time is kept at the millisecond level. Conclusion: This system has more accurate query accuracy and less running time.

Intelligent Interaction of Fluorine containing materials Based on NL2SQL

The data on the production and development of fluorine containing materials are characterized by a large amount of data and a high degree of dimensionality of physical and chemical property characterization indicators. The manual way of analyzing the data item by item not only has high interaction cost, but also is difficult to analyze and explore the data intuitively. In order to efficiently utilize the data, this paper firstly constructs a dataset of fluorine-containing materials and proposes the Mengzi-ITPT model based on it, which takes Mengzi as the encoder and uses the attention mechanism to enhance the representation of the listed information. Meanwhile, for the data characteristics of fluorine containing materials, the training strategy of ITPT is adopted to improve the accuracy of the model. The experimental results show that the accuracy of the Mengzi-ITPT model query reaches 86.9% when the model is trained under the fluorine-containing material dataset.

Locality sensitive hashing scheme based on online-learning

Locally Sensitive Hashing (LSH) algorithms are classical algorithms commonly used on the c-Approximate Nearest Neighbor (c-ANN) search problem. When using Euclidean distance to measure sample similarity and solve the c-ANN problem, the traditional approach is to utilize the Exact Euclidean Locality Sensitive Hashing (E2LSH) algorithm based on the p-stable distribution. However, the uncertainty of the p-stable distribution causes the hash buckets constructed by the E2LSH algorithm to vary in queries. Therefore, this paper proposes the OLLSH algorithm based on the Weighted Majority algorithm in the Online-Learning framework, which selects the hash buckets with more stable query accuracy by weighted voting on the hash buckets generated by the E2LSH algorithm. Then, we conduct simulation experiments on synthetic dataset and four real data sets and conclude that the proposed OLLSH algorithm improves the accuracy compared to the original algorithm with the same memory usage.

Dichotomy Graph Sketch: Summarizing Graph Streams with High Accuracy Based on Deep Learning

In many applications, data streams are indispensable to describe the relationships between nodes in networks, such as social networks, computer networks, and hyperlink networks. Fundamentally, a graph stream is a dynamic representation of a graph, which is usually composed of a sequence of edges, where each edge is represented by two endpoints and a weight. As a result of its large volume and highly dynamic nature, several graph sketches were proposed for the purposes of summarizing large-scale graph streams and enabling fast query processing. By using a compact data structure with hash functions, the graph sketches sequentially store the edges. Nevertheless, the existing graph sketches suffer from low performance on graph query tasks as a result of unpredictable collisions between heavy edges and light edges. To store heavy edges and light edges, this paper introduces a novel learning-based Dichotomy Graph Sketch (DGS) mechanism that uses two separate graph sketches, a heavy sketch and a light sketch. During a graph stream session, DGS obtains heavy edges and light edges, and uses these edges as training samples for a deep neural network (DNN) based binary classifier. The DNN-based classifier is then used to determine whether the upcoming edges are heavy or not. We will store the edges that are classified as heavy edges in the heavy sketch, and those that are classified as light edges in the light sketch. By combining the learnable classifier and Dichotomy Graph Sketches, the proposed mechanism resolves the hashing collision problem in conventional graph sketches and significantly improves graph query accuracy. The DGS algorithm outperforms the state-of-the-art graph sketches in a variety of graph query tasks based on extensive experiments that were conducted on four real-world graph stream datasets.

A Learned Cuckoo Filter for Approximate Membership Queries over Variable-sized Sliding Windows on Data Streams

Designing a space-efficient data structure to answer membership queries while ensuring high accuracy and real-time response is a challenging task in the field of stream processing. Many techniques have been developed to answer these queries in a sliding windows manner. However, assuming the user will conduct the query with the presupposed window size is not always practical. In this paper, we introduce a novel data structure called Learned Cuckoo Filter (LCF). It can provide satisfactory results for the approximate membership query on data streams, regardless of the user-defined query windows. LCF operates by adaptively maintaining cuckoo filters with the assistance of a well-trained oracle that learned the frequency feature of the data within the stream. To further enhance memory utilization, we develop a compact version of LCF (denoted by LCF_C), which selectively removes redundant information to reduce space consumption without compromising query accuracy. Furthermore, we conduct a thorough theoretical analysis of query accuracy and provide detailed guidelines for optimal parameter selection (denoted by LCF_O). Extensive experimental studies on synthetic and real-world datasets demonstrate the superiority of the proposed methods in terms of both space consumption and accuracy. Compared to the state-of-the-art algorithms, LCF_O can reduce up to 61% of space cost at the same error level, and achieve up to 12× improved accuracy with the same space cost.

Neural-based inexact graph de-anonymization

Graph de-anonymization is a technique used to reveal connections between entities in anonymized graphs, which is crucial in detecting malicious activities, network analysis, social network analysis, and more. Despite its paramount importance, conventional methods often grapple with inefficiencies and challenges tied to obtaining accurate query graph data. This paper introduces a neural-based inexact graph de-anonymization, which comprises an embedding phase, a comparison phase, and a matching procedure. The embedding phase uses a graph convolutional network to generate embedding vectors for both the query and anonymized graphs. The comparison phase uses a neural tensor network to ascertain node resemblances. The matching procedure employs a refined greedy algorithm to discern optimal node pairings. Additionally, we comprehensively evaluate its performance via well-conducted experiments on various real datasets. The results demonstrate the effectiveness of our proposed approach in enhancing the efficiency and performance of graph de-anonymization through the use of graph embedding vectors.

VFLR: An Efficient and Privacy-Preserving Vertical Federated Framework for Logistic Regression

With the explosive growth of data volume and computing capability, federated learning, which involves constructing global models over multiple data islands, has demonstrated its advantages and vast prospects in the field of machine learning. However, due to commonly vertically partitioned data, coupled with privacy concerns about data leakage, there are still some challenging issues in traditional federated learning. To tackle these challenges, in this paper, we propose an efficient and privacy-preserving vertical federated learning framework for logistic regression, named VFLR, where multiple participants can collaboratively perform global model training and query over their vertically partitioned data. Specifically, we first design a data aggregation matrix construction algorithm, with which the vertically partitioned data can be aggregated for high-accuracy global model training. Then, by utilizing a novel symmetric homomorphic encryption, our framework can ensure that the whole training and query processes do not leak any private information. Moreover, based on the data aggregation matrix, multi-round interactions are not required in VFLR, improving training efficiency significantly. Detailed security analysis shows that VFLR can well protect data and model information from inference attacks. In addition, extensive experiments demonstrate that VFLR has high training and query accuracy and low computation and communication overhead.