Qiushi Sun, Mukai Li, Zhoumianze Liu, Zhihui Xie, Fangzhi Xu, Zhangyue Yin, Kanzhi Cheng, Zehao Li, Zichen Ding, Qi Liu, Zhiyong Wu, Zhuosheng Zhang, Ben Kao, Lingpeng Kong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Graphical User Interface (GUI) agents powered by Vision-Language Models (VLMs) have demonstrated human-like computer control capability. Despite their utility in advancing digital automation, a critical bottleneck persists: collecting high-quality trajectory data for training. Common practices for collecting such data rely on human supervision or synthetic data generation through executing pre-defined tasks, which are either resource-intensive or unable to guarantee data quality. Moreover, these methods suffer from limited data diversity and significant gaps between synthetic data and real-world environments. To address these challenges, we propose OS-Genesis, a novel GUI data synthesis pipeline that reverses the conventional trajectory collection process. Instead of relying on pre-defined tasks, OS-Genesis enables agents first to perceive environments and perform step-wise interactions, then retrospectively derive high-quality tasks to enable trajectory-level exploration. A trajectory reward model is then employed to ensure the quality of the generated trajectories. We demonstrate that training GUI agents with OS-Genesis significantly improves their performance on highly challenging online benchmarks. In-depth analysis further validates OS-Genesis's efficiency and its superior data quality and diversity compared to existing synthesis methods. Our codes, data, and checkpoints are available at OS-Genesis Homepage.

Prompt compression for Retrieval-Augmented Generation (RAG) often fails by treating all retrieved information uniformly. This undifferentiated approach neglects the critical distinction between foundational core knowledge and illustrative practical knowledge, a failure especially damaging in hierarchical domains like law where essential principles can be discarded for redundant details, diminishing information gain.

Retrieval of legal knowledge by the general public is a challenging problem due to the technicality of the professional knowledge and the lack of fundamental understanding by laypersons on the subject. Traditional information retrieval techniques assume that users are capable of formulating succinct and precise queries for effective document retrieval. In practice, however, the wide gap between the highly technical contents and untrained users makes legal knowledge retrieval very difficult. We propose a methodology, called QBR, which employs a Questions Bank (QB) as an effective medium for bridging the knowledge gap. We show how the QB is used to derive training samples to enhance the embedding of knowledge units within documents, which leads to effective fine-grained knowledge retrieval. We discuss and evaluate through experiments various advantages of QBR over traditional methods. These include more accurate, efficient, and explainable document retrieval, better comprehension of retrieval results, and highly effective fine-grained knowledge retrieval. We also present some case studies and show that QBR achieves social impact by assisting citizens to resolve everyday legal concerns.

Analyzing flow of objects or data at different granularities of space and time can unveil interesting insights or trends.For example, transportation companies, by aggregating passenger travel data (e.g., counting passengers travelling from one region to another), can analyze movement behavior.In this paper, we study the problem of finding important trends in passenger movements between regions at different granularities.We define Origin (𝑂), Destination (𝐷), and Time (𝑇 ) patterns (ODT patterns) and propose an algorithm that enumerates them.We propose optimizations that greatly reduce the search space and the computational cost of pattern enumeration.We also propose pattern variants (constrained patterns and top-𝑘 patterns) that could be useful to different application scenarios.We evaluate our methods on three real datasets and identify significant ODT flow patterns in them.

We explore the use of multi-stage prompting with retrieval-augmented generation (RAG) in generating legal analysis. We break the complex legal problems into small steps, including initial analysis, legal query formulation, legal sources retrieval, legal reasoning, result consolidation, and output generation. The design of this framework emulates the process a legal professional in a common law system approaches the problem.

Global popularity (GP) bias is the phenomenon that popular items are recommended much more frequently than they should be, which goes against the goal of providing personalized recommendations and harms user experience and recommendation accuracy. Many methods have been proposed to reduce GP bias but they fail to notice the fundamental problem of GP, i.e., it considers popularity from a global perspective of all users and uses a single set of popular items, and thus cannot capture the interests of individual users. As such, we propose a user-aware version of item popularity named personal popularity (PP), which identifies different popular items for each user by considering the users that share similar interests. As PP models the preferences of individual users, it naturally helps to produce personalized recommendations and mitigate GP bias. To integrate PP into recommendation, we design a general personal popularity aware counterfactual (PPAC) framework, which adapts easily to existing recommendation models. In particular, PPAC recognizes that PP and GP have both direct and indirect effects on recommendations and controls direct effects with counterfactual inference techniques for unbiased recommendations. All codes and datasets are available at https://github.com/Stevenn9981/PPAC.

Entity alignment (EA), a crucial task in knowledge graph (KG) research, aims to identify equivalent entities across different KGs to support downstream tasks like KG integration, text-to-SQL, and question-answering systems. Given rich semantic information within KGs, pre-trained language models (PLMs) have shown promise in EA tasks due to their exceptional context-aware encoding capabilities. However, the current solutions based on PLMs encounter obstacles such as the need for extensive training, expensive data annotation, and inadequate incorporation of structural information. In this study, we introduce a novel zero-training EA framework, ZeroEA, which effectively captures both semantic and structural information for PLMs. To be specific, Graph2Prompt module serves as the bridge between graph structure and plain text by converting KG topology into textual context suitable for PLM input. Additionally, in order to provide PLMs with concise and clear input text of reasonable length, we design a motif-based neighborhood filter to eliminate noisy neighbors. The comprehensive experiments and analyses on 5 benchmark datasets demonstrate the effectiveness of ZeroEA, outperforming all leading competitors and achieving state-of-the-art performance in entity alignment. Notably, our study highlights the considerable potential of EA technique in improving the performance of downstream tasks, thereby benefitting the broader research field.

We study the problem of semantically annotating textual documents that are complex in the sense that the documents are long, feature rich, and domain specific. Due to their complexity, such annotation tasks require trained human workers, which are very expensive in both time and money. We propose CEMA, a method for deploying machine learning to assist humans in complex document annotation. CEMA estimates the human cost of annotating each document and selects the set of documents to be annotated that strike the best balance between model accuracy and human cost. We conduct experiments on complex annotation tasks in which we compare CEMA against other document selection and annotation strategies. Our results show that CEMA is the most cost-efficient solution for those tasks.

The Hong Kong Legal Information Institute (HKLII) provides a repository of legal documents in Hong Kong and such as ordinances and historical court judgments. HKLII provides a search facility through which users retrieve relevant documents. We perform statistical analysis on HKLII access log over a 5-year period categorizing user search queries and discovering interesting user access patterns. Based on this study, we propose enhancing user experience through identifying the search intent of the user. We classify a user query into one of several types and customize the search behavior based on the query type to enhance user experience. Our study provides an example of leveraging log analysis in legal information system design.

Graph neural networks (GNNs) have emerged as the state-of-the-art paradigm for collaborative filtering (CF). To improve the representation quality over limited labeled data, contrastive learning has attracted attention in recommendation and benefited graph-based CF model recently. However, the success of most contrastive methods heavily relies on manually generating effective contrastive views for heuristic-based data augmentation. This does not generalize across different datasets and downstream recommendation tasks, which is difficult to be adaptive for data augmentation and robust to noise perturbation. To fill this crucial gap, this work proposes a unified Automated Collaborative Filtering (AutoCF) to automatically perform data augmentation for recommendation. Specifically, we focus on the generative self-supervised learning framework with a learnable augmentation paradigm that benefits the automated distillation of important self-supervised signals. To enhance the representation discrimination ability, our masked graph autoencoder is designed to aggregate global information during the augmentation via reconstructing the masked subgraph structures. Experiments and ablation studies are performed on several public datasets for recommending products, venues, and locations. Results demonstrate the superiority of AutoCF against various baseline methods. We release the model implementation at https://github.com/HKUDS/AutoCF.

We propose knowledge internalization (KI), which aims to complement the lexical knowledge into neural dialog models.Instead of further conditioning the knowledge-grounded dialog (KGD) models on externally retrieved knowledge, we seek to integrate knowledge about each input token internally into the model's parameters.To tackle the challenge due to the large scale of lexical knowledge, we adopt the contrastive learning approach and create an effective token-level lexical knowledge retriever that requires only weak supervision mined from Wikipedia.We demonstrate the effectiveness and general applicability of our approach on various datasets and diversified model structures.

Heterogeneous Information Networks (HINs) are labeled graphs that depict relationships among different types of entities (e.g., users, movies and directors). For HINs,meta-path-based recommenders (MPRs) utilize meta-paths (i.e., abstract paths consisting of node and link types) to predict user preference, and have attracted a lot of attention due to their explainability and performance. We observe that the performance of MPRs is highly sensitive to the meta-paths they use, but existing works manually select the meta-paths from many possible ones. Thus, to discover effective meta-paths automatically, we propose the Reinforcement learning-based Meta-path Selection (RMS) framework. Specifically, we define a vector encoding for meta-paths and design a policy network to extend meta-paths. The policy network is trained based on the results of downstream recommendation tasks and an early stopping approximation strategy is proposed to speed up training. (RMS) is a general model, and it can work with all existing MPRs. We also propose a new MPR called RMS-HRec, which uses an attention mechanism to aggregate information from the meta-paths. We conduct extensive experiments on real datasets. Compared with the manually selected meta-paths, the meta-paths identified by (RMS) consistently improve recommendation quality. Moreover, RMS-HRec outperforms state-of-the-art recommender systems by an average of 7% in hit ratio. The codes and datasets are available on https://github.com/Stevenn9981/RMS-HRec.

We study the problem of machine comprehension of court judgments and generation of descriptive tags for judgments. Our approach makes use of a legal taxonomy D, which serves as a dictionary of canonicalized legal concepts. Given a court judgment J, our method identifies the key contents of J and then applies Word2Vec and BERT-based models to select a short list TJ of terms/phrases from the taxonomy D as descriptive tags of J. The tag set TJ suggests concepts that are relevant to or associative with J and provides a simple mechanism for readers of J to compose associative queries for effective judgment recommendation. Our prototype system implemented on the Hong Kong Legal Information Institute (HKLII) platform shows that our method provides a highly effective tool that assists users in exploring a judgment corpus and in obtaining relevant judgment recommendation.

The impressive accuracy of deep neural networks (DNNs) has created great demands on practical analytics over video data. Although efficient and accurate, the latest video analytic systems have not supported analytics beyond selection and aggregation queries. In data analytics, Top-K is a very important analytical operation that enables analysts to focus on the most important entities. In this demonstration, we present Everest, the first system that supports efficient and accurate Top-K video analytics. Everest ranks and identifies the most interesting frames/clips from videos with probabilistic guarantees. Furthermore, it supports user-defined functions to rank frames/clips based on different semantics using different deep vision models. Everest leverages techniques from computer vision, uncertain databases, and Top-K query processing to return results quickly.

A heterogeneous information network (HIN) is one whose nodes model objects of different types and whose links model objects’ relationships. To enrich its information, objects in an HIN are typically associated with additional attributes. We call such an HIN anAttributed HINor AHIN. We study the problem of clustering objects in an AHIN, taking into account objects’ similarities with respect to both object attribute values and their structural connectedness in the network. We show how supervision signal, expressed in the form of amust-link setand acannot-link set, can be leveraged to improve clustering results. We put forward the SCHAIN algorithm to solve the clustering problem, and two highly efficient variants, SCHAIN-PI and SCHAIN-IRAM, which employ thepower iteration based methodand theimplicitly restarted Arnoldi methodrespectively to compute eigenvectors of a matrix. We conduct extensive experiments comparing SCHAIN-based algorithms with other state-of-the-art clustering algorithms. Our results show that SCHAIN-IRAM outperforms other competitors in terms of clustering effectiveness and is highly efficient.

Zhiyong Wu, Lingpeng Kong, Wei Bi, Xiang Li, Ben Kao. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

A heterogeneous information network (HIN) has as vertices objects of different types and as edges the relations between objects, which are also of various types. We study the problem of classifying objects in HINs. Most existing methods perform poorly when given scarce labeled objects as training sets, and methods that improve classification accuracy under such scenarios are often computationally expensive. To address these problems, we propose ConCH, a graph neural network model. ConCH formulates the classification problem as a multitask learning problem that combines semi-supervised learning with self-supervised learning to learn from both labeled and unlabeled data. ConCH employs meta-paths, which are sequences of object types that capture semantic relationships between objects. ConCH co-derives object embeddings and context embeddings via graph convolution. It also uses the attention mechanism to fuse such embeddings. We conduct extensive experiments to evaluate the performance of ConCH against other 15 classification methods. Our results show that ConCH is an effective and efficient method for HIN classification.

Online legal document libraries, such as WorldLII, are indispensable tools for legal professionals to conduct legal research. We study how topic modeling techniques can be applied to such platforms to facilitate searching of court judgments. Specifically, we improve search effectiveness by matching judgments to queries at semantics level rather than at keyword level. Also, we design a system that summarizes a retrieved judgment by highlighting a small number of paragraphs that are semantically most relevant to the user query. This summary serves two purposes: (1) It explains to the user why the machine finds the retrieved judgment relevant to the user’s query, and (2) it helps the user quickly grasp the most salient points of the judgment, which significantly reduces the amount of time needed by the user to go through the returned search results. We further enhance our system by integrating domain knowledge provided by legal experts. The knowledge includes the features and aspects that are most important for a given category of judgments. Users can then view a judgement’s summary focusing on particular aspects only. We illustrate the effectiveness of our techniques with a user evaluation experiment on the HKLII platform. The results show that our methods are highly effective.

Online tutoring is an increasingly popular learning and business model, which matches students with amateur tutors recruited through a platform. While such platforms offer a unique opportunity to study how online and on-demand learning happens at scale, little is still known about how tutors engage on the platform. We present an exploratory analysis of tutor engagement on a large online tutoring platform, using an anonymized dataset of virtual sessions between tutors and students. Specifically, we show significant association between tutor engagement and associated factors like the subject and difficulty level of questions, and the experience of tutors on the platform. Moreover, we highlight important heterogeneities in tutor behavior, particularly in their work intensity. Our findings hold important implications for tutor retention and learning effectiveness on such platforms.

The impressive accuracy of deep neural networks (DNNs) has created great demands on practical analytics over video data. Although efficient and accurate, the latest video analytic systems have not supported analytics beyond selection and aggregation queries. In data analytics, Top-K is a very important analytical operation that enables analysts to focus on the most important entities. In this paper, we present Everest, the first system that supports efficient and accurate Top-K video analytics. Everest ranks and identifies the most interesting frames/moments from videos with probabilistic guarantees. Everest is a system built with a careful synthesis of deep computer vision models, uncertain data management, and Top-K query processing. Evaluations on real-world videos and the latest Visual Road benchmark show that Everest achieves between 14.3x to 20.6x higher efficiency than baseline approaches with high result accuracy.

By introducing a small set of additional parameters, a probe learns to solve specific linguistic tasks (e.g., dependency parsing) in a supervised manner using feature representations (e.g., contextualized embeddings).The effectiveness of such probing tasks is taken as evidence that the pre-trained model encodes linguistic knowledge.However, this approach of evaluating a language model is undermined by the uncertainty of the amount of knowledge that is learned by the probe itself.Complementary to those works, we propose a parameter-free probing technique for analyzing pre-trained language models (e.g., BERT).Our method does not require direct supervision from the probing tasks, nor do we introduce additional parameters to the probing process.Our experiments on BERT show that syntactic trees recovered from BERT using our method are significantly better than linguistically-uninformed baselines.We further feed the empirically induced dependency structures into a downstream sentiment classification task and find its improvement compatible with or even superior to a human-designed dependency schema.

Judgment prediction is the task of predicting various outcomes of legal cases of which sentencing prediction is one of the most important yet difficult challenges. We study the applicability of machine learning (ML) techniques in predicting prison terms of drug trafficking cases. In particular, we study how legal domain knowledge can be integrated with ML models to construct highly accurate predictors. We illustrate how our criminal sentence predictors can be applied to address four important issues in legal knowledge management, which include (1) discovery of model drifts in legal rules, (2) identification of critical features in legal judgments, (3) fairness in machine predictions, and (4) explainability of machine predictions.

We study the problem of applying spectral clustering to cluster multi-scale data, which is data whose clusters are of various sizes and densities. Traditional spectral clustering techniques discover clusters by processing a similarity matrix that reflects the proximity of objects. For multi-scale data, distance-based similarity is not effective because objects of a sparse cluster could be far apart while those of a dense cluster have to be sufficiently close. Following [16], we solve the problem of spectral clustering on multi-scale data by integrating the concept of objects' "reachability similarity" with a given distance-based similarity to derive an objects' coefficient matrix. We propose the algorithm CAST that applies trace Lasso to regularize the coefficient matrix. We prove that the resulting coefficient matrix has the "grouping effect" and that it exhibits "sparsity". We show that these two characteristics imply very effective spectral clustering. We evaluate CAST and 10 other clustering methods on a wide range of datasets w.r.t. various measures. Experimental results show that CAST provides excellent performance and is highly robust across test cases of multi-scale data.

A knowledge-based question-answering (KB-QA) system is one that answers natural-language questions by accessing information stored in a knowledge base (KB). Existing KB-QA systems generally register an accuracy of 70-80% for simple questions and less for more complex ones. We observe that certain questions are intrinsically difficult to answer correctly with existing systems. We propose the PERQ framework to address this issue. Given a question q, we perform three steps to boost answer accuracy: (1) (Prediction) We predict if q can be answered correctly by a KB-QA system S. (2) (Explanation) If S is predicted to fail q, we analyze them to determine the most likely reasons of the failure. (3) (Rectification) We use the prediction and explanation results to rectify the answer. We put forward tools to achieve the three steps and analyze their effectiveness. Our experiments show that the PERQ framework can significantly improve KB-QA systems' accuracies over simple questions.

Novel road-network applications often recommend a moving object (e.g., a vehicle) about interesting services or tasks on its way to a destination. A taxi-sharing system, for instance, suggests a new passenger to a taxi while it is serving another one. The traveling cost is then shared among these passengers. A fundamental query is: given two nodes s and t, and an area A on road network graph , is there a “good” route (e.g., short enough path) P from s to t that crosses A in G? In a taxi-sharing system, s and t can be a taxi's current and destined locations, and A contains all the places to which a person waiting for a taxi is willing to walk. Answering this Route and Area Matching (ROAM) Query allows the application involved to recommend appropriate services to users efficiently. In this paper, we examine efficient ROAM query algorithms. Particularly, we develop solutions for finding a ρ-route, which is an s-t path that passes A, with a length of at most (1 + ρ) times the shortest distance between s and t. The existence of a ρ-route implies that a service or task located at A can be found for a given moving object m, and that m only deviates slightly from its current route. We present comprehensive studies on index-free and index-based algorithms for answering ROAM queries. Comprehensive experiments show that our algorithm runs up to 30 times faster than baseline algorithms.

Personalized PageRank (PPR) is a classic measure of the relevance among different nodes in a graph, and has been applied in numerous systems, such as Twitter's Who-To-Follow and Pinterest's Related Pins. Existing work on PPR has mainly focused on three general types of queries, namely, single-pair PPR, single-source PPR, and all-pair PPR. However, we observe that there are applications that rely on a new query type (referred to as batch one-hop PPR), which takes as input a set S of source nodes and, for each nodes E ϵ S and each of s's neighbor v, asks for the PPR value of v with respect to s. None of the existing PPR algorithms is able to efficiently process batch one-hop queries, due to the inherent differences between batch one-hop PPR and the three general query types. To address the limitations of existing algorithms, this paper presents Baton, an algorithm for batch one-hop PPR that offers both strong theoretical guarantees and practical efficiency. Baton leverages the characteristics of one-hop PPR to avoid unnecessary computation, and it incorporates advanced mechanisms to improve the cost-effectiveness of PPR derivations. Extensive experiments on benchmark datasets show that Baton is up to three orders of magnitude faster than the state of the art, while offering the same accuracy.

A heterogeneous information network (HIN) is one whose objects are of different types and links between objects could model different object relations. We study how spectral clustering can be effectively applied to HINs. In particular, we focus on how meta-path relations are used to construct an effective similarity matrix based on which spectral clustering is done. We formulate the similarity matrix construction as an optimization problem and propose the SClump algorithm for solving the problem. We conduct extensive experiments comparing SClump with other state-of-the-art clustering algorithms on HINs. Our results show that SClump outperforms the competitors over a range of datasets w.r.t. different clustering quality measures.

We study the problem of executing road-network k-nearest-neighbor (kNN) search on multi-core machines. State-of-the-art kNN algorithms on road networks often involve elaborate index structures and complex computational logic. Moreover, most kNN algorithms are inherently sequential. These make the traditional approach of parallel programming very costly, laborious, and ineffective when they are applied to kNN algorithms. We propose the MPR (Multi-layer Partitioning-Replication) mechanism that orchestrates CPU cores and schedules kNN query and index update processes to run on the cores. The MPR mechanism performs workload analysis to determine the best arrangement of the cores with the objective of optimizing quality-of-service (QoS) measures, such as system throughput and query response time. We demonstrate the effectiveness of MPR by applying it to a number of state-of-the-art kNN indexing methods running on a multi-core machine. Our experiments show that multi-processing using our MPR approach requires minimal programming effort. It also leads to significant improvements in query response time and system throughput compared with other baseline parallelization methods.

Personalized PageRank (PPR) is a classic measure of the relevance among different nodes in a graph. Existing work on PPR has mainly focused on three general types of queries, namely, single-pair PPR, single-source PPR, and all-pair PPR. However, there are applications that rely on a new query type (referred to as batch one-hop PPR), which takes as input a set S of source nodes and, for each node s in S and each of s's neighbor v, asks for the PPR value of v with respect to s. None of the existing PPR algorithms is able to efficiently process batch one-hop queries, due to the inherent differences between batch one-hop PPR and the three general query types. To address the limitations of existing algorithms, this paper presents Baton, an algorithm for batch one-hop PPR that offers strong practical efficiency.

An Open Information Extraction (OIE) system processes textual data to extract assertions, which are structured data typically represented in the form of (subject;relation; object) triples. An Open Knowledge Base (OKB) is a collection of such assertions. We study the problem of canonicalizing an OKB, which is defined as the problem of mapping each name (a textual term such as "the rockies", "colorado rockies") to a canonical form (such as "rockies"). Galárraga et al. [18] proposed a hierarchical agglomerative clustering algorithm using canopy clustering to tackle the canonicalization problem. The algorithm was shown to be very effective. However, it is not efficient enough to practically handle large OKBs due to the large number of similarity score computations. We propose the FAC algorithm for solving the canonicalization problem. FAC employs pruning techniques to avoid unnecessary similarity computations, and bounding techniques to efficiently approximate and identify small similarities. In our experiments, FAC registers ordersof-magnitude speedups over other approaches.

We investigate the effectiveness of spectral methods in clustering multi-scale data, which is data whose clusters are of various sizes and densities. We review existing spectral methods that are designed to handle multi-scale data and propose an alternative approach that is orthogonal to existing methods. We put forward the algorithm ROSC, which computes an affinity matrix that takes into account both objects' feature similarity and reachability similarity. We perform extensive experiments comparing ROSC against 9 other methods on both real and synthetic datasets. Our results show that ROSC performs very well against the competitors. In particular, it is very robust in that it consistently performs well over all the datasets tested. Also, it outperforms others by wide margins for datasets that are highly multi-scale.

We study the classical kNN queries on road networks. Existing solutions mostly focus on reducing query processing time. In many applications, however, system throughput is a more important measure. We devise a mathematical model that describes throughput in terms of a number of system characteristics. We show that query time is only one of the many parameters that impact throughput. Others include update time and query/update arrival rates. We show that the traditional approach of improving query time alone is generally inadequate in optimizing throughput. Moreover, existing solutions lack flexibility in adapting to environments of different characteristics. We propose Toain, which is a very flexible algorithm that can be easily trained to adapt to a given environment for maximizing query throughput. We conduct extensive experiments on both real and synthetic data and show that Toain gives significantly higher throughput compared with existing solutions.

Spatial object search is prevalent in map services (e.g., Google Maps). To rent an apartment, for example, one will take into account its nearby facilities, such as supermarkets, hospitals, and subway stations. Traditional keyword search solutions, such as the nearby function in Google Maps, are insufficient in expressing the often complex attribute/spatial requirements of users. Those require- ments, however, are essential to reflect the user search intention. In this paper, we propose the Spatial Exemplar Query (SEQ), which allows the user to input a result example over an interface inside the map service. We then propose an effective similarity measure to evaluate the proximity between a candidate answer and the given example. We conduct a user study to validate the effectiveness of SEQ. Our result shows that more than 88% of users would like to have an example assisted search in map services. Moreover, SEQ gets a user satisfactory score of 4.3/5.0, which is more than 2 times higher than that of a baseline solution.

A heterogeneous information network (HIN) is one whose nodes model objects of different types and whose links model objects' relationships. In many applications, such as social networks and RDF-based knowledge bases, information can be modeled as HINs. To enrich its information content, objects (as represented by nodes) in an HIN are typically associated with additional attributes. We call such an HIN an Attributed HIN or AHIN. We study the problem of clustering objects in an AHIN, taking into account objects' similarities with respect to both object attribute values and their structural connectedness in the network. We show how supervision signal, expressed in the form of a must-link set and a cannot-link set, can be leveraged to improve clustering results. We put forward the SCHAIN algorithm to solve the clustering problem. We conduct extensive experiments comparing SCHAIN with other state-of-the-art clustering algorithms and show that SCHAIN outperforms the others in clustering quality.

A Sequence OLAP(S-OLAP) system provides a platform on which pattern-based aggregate (PBA) queries on a sequence database are evaluated. In its simplest form, a PBA query consists of a pattern template T and an aggregate function F. A pattern template is a sequence of variables, each is defined over a domain. Each variable is instantiated with all possible values in its corresponding domain to derive all possible patterns of the template. Sequences are grouped based on the patterns they possess. The answer to a PBA query is a sequence cuboid (s-cuboid), which is a multidimensional array of cells. Each cell is associated with a pattern instantiated from the query's pattern template. The value of each s-cuboid cell is obtained by applying the aggregate function F to the set of data sequences that belong to that cell. Since a pattern template can involve many variables and can be arbitrarily long, the induced s-cuboid for a PBA query can be huge. For most analytical tasks, however, only iceberg cells with very large aggregate values are of interest. This paper proposes an efficient approach to identifying and evaluating iceberg cells of s-cuboids. Experimental results show that our algorithms are orders of magnitude faster than existing approaches.

A heterogeneous information network (HIN) is used to model objects of different types and their relationships. Objects are often associated with properties such as labels. In many applications, such as curated knowledge bases for which object labels are manually given, only a small fraction of the objects are labeled. Studies have shown that transductive classification is an effective way to classify and to deduce labels of objects, and a number of transductive classifiers have been put forward to classify objects in an HIN. We study the performance of a few representative transductive classification algorithms on HINs. We identify two fundamental properties, namely, cohesiveness and connectedness, of an HIN that greatly influence the effectiveness of transductive classifiers. We define metrics that measure the two properties. Through experiments, we show that the two properties serve as very effective indicators that predict the accuracy of transductive classifiers. Based on cohesiveness and connectedness we derive (1) a black-box tester that evaluates whether transductive classifiers should be applied for a given classification task and (2) an active learning algorithm that identifies the objects in an HIN whose labels should be sought in order to improve classification accuracy.

A heterogeneous information network (HIN) is used to model objects of different types and their relationships. Meta-paths are sequences of object types. They are used to represent complex relationships between objects beyond what links in a homogeneous network capture. We study the problem of classifying objects in an HIN. We propose class-level meta-paths and study how they can be used to (1) build more accurate classifiers and (2) improve active learning in identifying objects for which training labels should be obtained. We show that class-level meta-paths and object classification exhibit interesting synergy. Our experimental results show that the use of class-level meta-paths results in very effective active learning and good classification performance in HINs.

A knowledge-based question-answering system (KB-QA) is one that answers natural language questions with information stored in a large-scale knowledge base (KB). Existing KB-QA systems are either powered by curated KBs in which factual knowledge is encoded in entities and relations with well-structured schemas, or by open KBs, which contain assertions represented in the form of triples (e.g., subject; relation phrase; argument). We show that both approaches fall short in answering questions with complex prepositional or adverbial constraints. We propose using n-tuple assertions, which are assertions with an arbitrary number of arguments, and n-tuple open KB (nOKB), which is an open knowledge base of n-tuple assertions. We present TAQA, a novel KB-QA system that is based on an nOKB and illustrate via experiments how TAQA can effectively answer complex questions with rich semantic constraints. Our work also results in a new open KB containing 120M n-tuple assertions and a collection of 300 labeled complex questions, which is made publicly available for further research.

Efficient main-memory index structures are crucial to main-memory database systems. Adaptive Radix Tree (ART) is the most recent in-memory index structure. ART is designed to avoid cache miss, leverage SIMD data parallelism, minimize branch mis-prediction, and have small memory footprint. When an in-memory index structure like ART has significantly few cache misses and branch mis-predictions, it is natural to question whether misses in Translation Lookaside Buffer (TLB) matters. In this paper, we try to confirm whether this is the case and if the answer is positive, what are the measures that we can take to alleviate that and how effective they are.

Scan and lookup are two core operations in main memory column stores. A scan operation scans a column and returns a result bit vector that indicates which records satisfy a filter. Once a column scan is completed, the result bit vector is converted into a list of record numbers, which is then used to look up values from other columns of interest for a query. Recently there are several in-memory data layout proposals that aim to improve the performance of in-memory data processing. However, these solutions all stand at either end of a trade-off --- each is either good in lookup performance or good in scan performance, but not both. In this paper we present ByteSlice, a new main memory storage layout that supports both highly efficient scans and lookups. ByteSlice is a byte-level columnar layout that fully leverages SIMD data-parallelism. Micro-benchmark experiments show that ByteSlice achieves a data scan speed at less than 0.5 processor cycle per column value --- a new limit of main memory data scan, without sacrificing lookup performance. Our experiments on TPC-H data and real data show that ByteSlice offers significant performance improvement over all state-of-the-art approaches.

We address security issues in a cloud database system which employs the DBaaS model --- a data owner (DO) exports data to a cloud database service provider (SP). To provide data security, sensitive data is encrypted by the DO before it is uploaded to the SP. Compared to existing secure query processing systems like CryptDB [7] and MONOMI [8], in which data operations (e.g., comparison or addition) are supported by specialized encryption schemes, our demo system, SDB, is implemented based on a set of data-interoperable secure operators, i.e., the output of an operator can be used as input of another operator. As a result, SDB can support a wide range of complex queries (e.g., all TPC-H queries) efficiently. In this demonstration, we show how our SDB prototype supports secure query processing on complex workload like TPC-H. We also demonstrate how our system protects sensitive information from malicious attackers.

In many applications, entities and their relationships are represented by graphs. Examples include the WWW (web pages and hyperlinks) and bibliographic networks (authors and co-authorship). A graph can be conveniently modeled by a matrix from which various quantitative measures are derived. Some example measures include PageRank and SALSA (which measure nodes ’ importance), and Personal-ized PageRank and Random Walk with Restart (which mea-sure proximities between nodes). To compute these mea-sures, linear systems of the form Ax = b, where A is a ma-trix that captures a graph’s structure, need to be solved. To facilitate solving the linear system, the matrix A is often de-composed into two triangular matrices (L and U). In a dy-namic world, the graph that models it changes with time and thus is the matrix A that represents the graph. We consider a sequence of evolving graphs and its associated sequence of evolving matrices. We study how LU-decomposition should be done over the sequence so that (1) the decomposition is efficient and (2) the resulting LU matrices best preserve the sparsity of the matrices A’s (i.e., the number of extra non-zero entries introduced in L and U are minimized.) We propose a cluster-based algorithm CLUDE for solving the problem. Through an experimental study, we show that CLUDE is about an order of magnitude faster than the traditional incremental update algorithm. The number of extra non-zero entries introduced by CLUDE is also about an order of magnitude fewer than that of the traditional algorithm. CLUDE is thus an efficient algorithm for LU de-composition that produces high-quality LU matrices over an evolving matrix sequence. 1.

The discounted hitting time (DHT), which is a random-walk similarity measure for graph node pairs, is useful in various applications, including link prediction, collaborative recommendation, and reputation ranking. We examine a novel query, called the multi-way join (or n-way join), on DHT scores. Given a graph and n sets of nodes, the n-way join retrieves a set of n-tuples with the k highest scores, according to some aggregation function of DHT values. This query enables analysis and prediction of complex relationship among n sets of nodes. Since an n-way join is expensive to compute, we develop the Partial Join algorithm (or PJ). This solution decomposes an n-way join into a number of top-m 2-way joins, and combines their results to construct the answer of the n-way join. Since PJ may necessitate the computation of top-(m+ 1) 2-way joins, we study an incremental solution, which allows the top-(m+ 1) 2-way join to be derived quickly from the top-m 2-way join results earlier computed. We further examine fast processing and pruning algorithms for 2-way joins. An extensive evaluation on three real datasets shows that PJ accurately evaluates n-way joins, and is four orders of magnitude faster than basic solutions.

With the rapid growth of Web 2.0, a variety of content sharing services, such as Flickr, YouTube, Blogger, and TripAdvisor etc, have become extremely popular over the last decade. On these websites, users have created and shared with each other various kinds of resources, such as photos, video, and travel blogs. The sheer amount of user-generated content varies greatly in quality, which calls for a principled method to identify a set of authorities, who created high-quality resources, from a massive number of contributors of content. Since most previous studies only infer global authoritativeness of a user, there is no way to differentiate the authoritativeness in different aspects of life (topics).

We address security issues in a cloud database system which employs the DBaaS model. In such a model, a data owner (DO) exports its data to a cloud database service provider (SP). To provide data security, sensitive data is encrypted by the DO before it is uploaded to the SP. Existing encryption schemes, however, are only partially homomorphic in the sense that each of them was designed to allow one specific type of computation to be done on encrypted data. These existing schemes cannot be integrated to answer real practical queries that involve operations of different kinds. We propose and analyze a secure query processing system (SDB) on relational tables and a set of elementary operators on encrypted data that allow data interoperability, which allows a wide range of SQL queries to be processed by the SP on encrypted information. We prove that our encryption scheme is secure against two types of threats and that it is practically efficient.

A Sequence OLAP (S-OLAP) system provides a platform on which pattern-based aggregate (PBA) queries on a sequence database are evaluated. In its simplest form, a PBA query consists of a pattern template T and an aggregate function F. A pattern template is a sequence of variables, each is defined over a domain. For example, the template T = (X,Y ,Y ,X) consists of two variables X and Y . Each variable is instantiated with all possible values in its corresponding domain to derive all possible patterns of the template. Sequences are grouped based on the patterns they possess. The answer to a PBA query is a sequence cuboid (s-cuboid), which is a multidimensional array of cells. Each cell is associated with a pattern instantiated from the query's pattern template. The value of each s-cuboid cell is obtained by applying the aggregate function F to the set of data sequences that belong to that cell. Since a pattern template can involve many variables and can be arbitrarily long, the induced s-cuboid for a PBA query can be huge. For most analytical tasks, however, only iceberg cells with very large aggregate values are of interest. This paper proposes an efficient approach to identify and evaluate iceberg cells of s-cuboids. Experimental results show that our algorithms are orders of magnitude faster than existing approaches.

In a crowdsourcing system, Human Intelligence Tasks (HITs) (e.g., translating sentences, matching photos, tagging videos with keywords) can be conveniently specified. HITs are made available to a large pool of workers, who are paid upon completing the HITs they have selected. Since workers may have different capabilities, some difficult HITs may not be satisfactorily performed by a single worker. If more workers are employed to perform a HIT, the quality of the HIT's answer could be statistically improved. Given a set of HITs and a fixed "budget", we address the important problem of determining the number of workers (or plurality) of each HIT so that the overall answer quality is optimized. We propose a dynamic programming (DP) algorithm for solving the plurality assignment problem (PAP). We identify two interesting properties, namely, monotonicity and diminishing return, which are satisfied by a HIT if the quality of the HIT's answer increases monotonically at a decreasing rate with its plurality. We show for HITs that satisfy the two properties (e.g., multiple-choice-question HITs), the PAP is approximable. We propose an efficient greedy algorithm for such case. We conduct extensive experiments on synthetic and real datasets to evaluate our algorithms. Our experiments show that our greedy algorithm provides close-to-optimal solutions in practice.

A social tagging system, such as del.icio.us and Flickr, allows users to annotate resources (e.g., web pages and photos) with text descriptions called tags. Tags have proven to be invaluable information for searching, mining, and recommending resources. In practice, however, not all resources receive the same attention from users. As a result, while some highly-popular resources are over-tagged, most of the resources are under-tagged. Incomplete tagging on resources severely affects the effectiveness of all tag-based techniques and applications. We address an interesting question: if users are paid to tag specific resources, how can we allocate incentives to resources in a crowd-sourcing environment so as to maximize the tagging quality of resources? We address this question by observing that the tagging quality of a resource becomes stable after it has been tagged a sufficient number of times. We formalize the concepts of tagging quality (TQ) and tagging stability (TS) in measuring the quality of a resource's tag description. We propose a theoretically optimal algorithm given a fixed “budget” (i.e., the amount of money paid for tagging resources). This solution decides the amount of rewards that should be invested on each resource in order to maximize tagging stability. We further propose a few simple, practical, and efficient incentive allocation strategies. On a dataset from del.icio.us, our best strategy provides resources with a close-to-optimal gain in tagging stability.

Order-preserving submatrices (OPSM's) have been shown useful in capturing concurrent patterns in data when the relative magnitudes of data items are more important than their exact values. For instance, in analyzing gene expression profiles obtained from microarray experiments, the relative magnitudes are important both because they represent the change of gene activities across the experiments, and because there is typically a high level of noise in data that makes the exact values untrustable. To cope with data noise, repeated experiments are often conducted to collect multiple measurements. We propose and study a more robust version of OPSM, where each data item is represented by a set of values obtained from replicated experiments. We call the new problem OPSM-RM (OPSM with repeated measurements). We define OPSM-RM based on a number of practical requirements. We discuss the computational challenges of OPSM-RM and propose a generic mining algorithm. We further propose a series of techniques to speed up two time dominating components of the algorithm. We show the effectiveness and efficiency of our methods through a series of experiments conducted on real microarray data.

Statistical debugging is a technique that mines data obtained from software executions in order to identify the program statements that are relevant to program bugs. Specifically, program predicates are injected into the program during compilation and statistics about those predicates are collected during the program execution. When bugs are found but the developers have no clue where the bugs are, they may call such a statistical debugger for help. The debugger ranks the injected predicates according to their statistical relevancy to bugs and presents the suspicious ones to the developers. When a bug is found and fixed, but the updated program still contains (some other) bugs, the preceding procedure is iterated until all bugs are fixed. There are two types of predicate-based statistical debugger: one type returns only simple predicates, another type returns only complex predicates. We envision that the next wave of statistic debuggers should be able to return both, depending on the kinds of bugs manifested in the software. In this paper, we take the first step and study the metrics for evaluating the effectiveness of statistical debuggers that can return both types of predicate predictors (simple or complex).

In typical location-based services (LBS), moving objects (e.g., GPS-enabled mobile phones) report their locations through a wireless network. An LBS server can use the location information to answer various types of continuous queries. Due to hardware limitations, location data reported by the moving objects are often uncertain. In this paper, we study efficient methods for the execution of Continuous Possible Nearest Neighbor Query (CPoNNQ) that accesses imprecise location data. A CPoNNQ is a standing query (which is active during a period of time) such that, at any time point, all moving objects that have non-zero probabilities of being the nearest neighbor of a given query point are reported. To handle the continuous nature of a CPoNNQ, a simple solution is to require moving objects to continuously report their locations to the LBS server, which evaluates the query at every time step. To save communication bandwidth and mobile devices' batteries, we develop two filter-based protocols for CPoNNQ evaluation. Our protocols install "filter bounds" on moving objects, which suppress unnecessary location reporting and communication between the server and the moving objects. Through extensive experiments, we show that our protocols can effectively reduce communication costs while maintaining a high query quality.

Web search queries issued by casual users are often short and with limited expressiveness. Query recommendation is a popular technique employed by search engines to help users refine their queries. Traditional similarity-based methods, however, often result in redundant and monotonic recommendations. We identify five basic requirements of a query recommendation system. In particular, we focus on the requirements of redundancy-free and diversified recommendations. We propose the DQR framework, which mines a search log to achieve two goals: (1) It clusters search log queries to extract query concepts, based on which recommended queries are selected. (2) It employs a probabilistic model and a greedy heuristic algorithm to achieve recommendation diversification. Through a comprehensive user study we compare DQR against five other recommendation methods. Our experiment shows that DQR outperforms the other methods in terms of relevancy, diversity, and ranking performance of the recommendations.

In social tagging systems, resources such as images and videos are annotated with descriptive words called tags. It has been shown that tag-based resource searching and retrieval is much more effective than content-based retrieval. With the advances in mobile technology, many resources are also geo-tagged with location information. We observe that a traditional tag (word) can carry different semantics at different locations. We study how location information can be used to help distinguish the different semantics of a resource's tags and thus to improve retrieval accuracy. Given a search query, we propose a location-partitioning method that partitions all locations into regions such that the user query carries distinguishing semantics in each region. Based on the identified regions, we utilize location information in estimating the ranking scores of resources for the given query. These ranking scores are learned using the Bayesian Personalized Ranking (BPR) framework. Two algorithms, namely, LTD and LPITF, which apply Tucker Decomposition and Pairwise Interaction Tensor Factorization, respectively for modeling the ranking score tensor are proposed. Through experiments on real datasets, we show that LTD and LPITF outperform other tag-based resource retrieval methods.

Many kinds of real-life data exhibit logical ordering among their data items and are thus sequential in nature. In recent years, the concept of Sequence OLAP (S-OLAP) has been proposed. The biggest distinguishing feature of SOLAP from traditional OLAP is that data sequences managed by an S-OLAP system are characterized by the subsequence/substring patterns they possess. An S-OLAP system thus supports pattern-based grouping and aggregation. Conceptually, an S-OLAP system maintains a sequence data cube which is composed of sequence cuboids. Each sequence cuboid presents the answer of a pattern-based aggregate (PBA) query. This paper focuses on the I/O aspects of evaluating PBA queries. We study the problems of joining plan selection and execution planning, which are the core issues in the design of I/O-efficient cuboid materialization algorithms. Through an empirical study, we show that our algorithms lead to a very I/O-efficient strategy for sequence cuboid materialization.

In a social tagging system, resources (such as photos, video and web pages) are associated with tags. These tags allow the resources to be effectively searched through tag-based keyword matching using traditional IR techniques. We note that in many such systems, tags of a resource are often assigned by a diverse audience of causal users (taggers). This leads to two issues that gravely affect the effectiveness of resource retrieval: (1) Noise: tags are picked from an uncontrolled vocabulary and are assigned by untrained taggers. The tags are thus noisy features in resource retrieval. (2) A multitude of aspects: different taggers focus on different aspects of a resource. Representing a resource using a flattened bag of tags ignores this important diversity of taggers. To improve the effectiveness of resource retrieval in social tagging systems, we propose CubeLSI - a technique that extends traditional LSI to include taggers as another dimension of feature space of resources. We compare CubeLSI against a number of other tag-based retrieval models and show that CubeLSI significantly outperforms the other models in terms of retrieval accuracy. We also prove two interesting theorems that allow CubeLSI to be very efficiently computed despite the much enlarged feature space it employs.

Traditional decision tree classifiers work with data whose values are known and precise. We extend such classifiers to handle data with uncertain information. Value uncertainty arises in many applications during the data collection process. Example sources of uncertainty include measurement/quantization errors, data staleness, and multiple repeated measurements. With uncertainty, the value of a data item is often represented not by one single value, but by multiple values forming a probability distribution. Rather than abstracting uncertain data by statistical derivatives (such as mean and median), we discover that the accuracy of a decision tree classifier can be much improved if the "complete information” of a data item (taking into account the probability density function (pdf)) is utilized. We extend classical decision tree building algorithms to handle data tuples with uncertain values. Extensive experiments have been conducted which show that the resulting classifiers are more accurate than those using value averages. Since processing pdfs is computationally more costly than processing single values (e.g., averages), decision tree construction on uncertain data is more CPU demanding than that for certain data. To tackle this problem, we propose a series of pruning techniques that can greatly improve construction efficiency.

The Sequence OLAP (S-OLAP) system is a novel online analytical processing system for analyzing sequence data. S-OLAP supports "pattern-based" grouping and aggregation on sequence data - a very powerful concept and capability that is not supported by traditional OLAP systems. It also supports several new OLAP operations that are specific to sequence data analysis. The query processing techniques documented in [1] have been implemented in our S-OLAP engine for efficient query processing. The system also provides users with a friendly graphical interface for query construction and result visualization. Query parameters can be interactively refined and the results are updated in real-time so as to facilitate the exploratory analysis of sequence data.

Although there has been a considerable body of work on skyline evaluation in multidimensional data with totally ordered attribute domains, there are only a few methods that consider attributes with partially ordered domains. Existing work maps each partially ordered domain to a total order and then adapts algorithms for totally-ordered domains to solve the problem. Nevertheless these methods either use stronger notions of dominance, which generate false positives, or require expensive dominance checks. In this paper, we propose two new methods, which do not have these drawbacks. The first method uses an appropriate mapping of a partial order to a total order, inspired by the lattice theorem and an off-the-shelf skyline algorithm. The second technique uses an appropriate storage and indexing approach, inspired by column stores, which enables efficient verification of whether a pair of objects are incompatible. We demonstrate that both our methods are up to an order of magnitude more efficient than previous work and scale well with different problem parameters, such as complexity of partial orders.

The idea of allowing query users to relax their correctness requirements in order to improve performance of a data stream management system (e.g., location-based services and sensor networks) has been recently studied. By exploiting the maximum error (or tolerance) allowed in query answers, algorithms for reducing the use of system resources have been developed. In most of these works, however, query tolerance is expressed as a numerical value, which may be difficult to specify. We observe that in many situations, users may not be concerned with the actual value of an answer, but rather which object satisfies a query (e.g., "who is my nearest neighbor?¿). In particular, an entity-based query returns only the names of objects that satisfy the query. For these queries, it is possible to specify a tolerance that is "nonvalue-based.¿ In this paper, we study fraction-based tolerance, a type of nonvalue-based tolerance, where a user specifies the maximum fractions of a query answer that can be false positives and false negatives. We develop fraction-based tolerance for two major classes of entity-based queries: 1) nonrank-based query (e.g., range queries) and 2) rank-based query (e.g., k-nearest-neighbor queries). These definitions provide users with an alternative to specify the maximum tolerance allowed in their answers. We further investigate how these definitions can be exploited in a distributed stream environment. We design adaptive filter algorithms that allow updates be dropped conditionally at the data stream sources without affecting the overall query correctness. Extensive experimental results show that our protocols reduce the use of network and energy resources significantly.

Abstract-We study the problem of clustering uncertain objects whose locations are described by probability density functions (pdfs). We show that the UK-means algorithm, which generalizes the k-means algorithm to handle uncertain objects, is very inefficient. The inefficiency comes from the fact that UK-means computes expected distances (EDs) between objects and cluster representatives. For arbitrary pdfs, expected distances are computed by numerical integrations, which are costly operations. We propose pruning techniques that are based on Voronoi diagrams to reduce the number of expected distance calculations. These techniques are analytically proven to be more effective than the basic bounding-box-based technique previously known in the literature. We then introduce an R-tree index to organize the uncertain objects so as to reduce pruning overheads. We conduct experiments to evaluate the effectiveness of our novel techniques. We show that our techniques are additive and, when used in combination, significantly outperform previously known methods.

Social tagging systems which allow users to create, edit and share collections of internet resources associated with tags in a collaborative fashion are growing in popularity in recent years. The rapidly growing amount of shared data in these folksonomies, i.e., taxonomies created by the folk, presents new technical challenges involved with discovering resources which are likely of interest to the user. Social tags which reflect the meaning of resources from the user's points of view provide an opportunity to enhance the quality of retrieval. In this paper, we introduce a novel framework to search relevant resources to the user query by incorporating information obtained from folksonomies' underlying data structures consisting of a set of user/tag/resource triplets. In contrast to traditional retrieval and recommendation techniques which represent a collection by a matrix, we represent our data as a third-order tensor on which a novel Cube Latent Semantic Indexing (CubeLSI) technique is proposed to capture latent semantic associations between tags. With the latent semantic representation we show how to rank relevant resources according to their relevance to user queries. The excellent performance of the method is demonstrated by an experimental evaluation on the deli.cio.us dataset. Copyright 2009 ACM.

In recent years, the warehousing and online-analytical processing (OLAP) of archived sequence data have received growing attentions. In particular, the concept of sequence OLAP is recently proposed with the objective of evaluating various kinds of so-called Pattern-Based Aggregate (PBA) queries so that various kinds of data analytical tasks on sequence data can be carried out efficiently. This paper studies the evaluation of ranking PBA queries, which rank the results of PBA queries and return only the top-ranked ones to users. We discuss how ranking PBA queries drastically improve the usability of S-OLAP systems and present techniques that can evaluate various kinds of ranking PBA queries efficiently.

Traditional decision tree classifiers work with data whose values are known and precise. We extend such classifiers to handle data with uncertain information, which originates from measurement/quantisation errors, data staleness, multiple repeated measurements, etc. The value uncertainty is represented by multiple values forming a probability distribution function (pdf). We discover that the accuracy of a decision tree classifier can be much improved if the whole pdf, rather than a simple statistic, is taken into account. We extend classical decision tree building algorithms to handle data tuples with uncertain values. Since processing pdf's is computationally more costly, we propose a series of pruning techniques that can greatly improve the efficiency of the construction of decision trees.

Traditional machine learning algorithms assume that data are exact or precise. However, this assumption may not hold in some situations because of data uncertainty arising from measurement errors, data staleness, and repeated measurements, etc. With uncertainty, the value of each data item is represented by a probability distribution function (pdf). In this paper, we propose a novel naive Bayes classification algorithm for uncertain data with a pdf. Our key solution is to extend the class conditional probability estimation in the Bayes model to handle pdf’s. Extensive experiments on UCI datasets show that the accuracy of naive Bayes model can be improved by taking into account the uncertainty information.

Service providers like Google and Amazon are moving into the SaaS (Software as a Service) business. They turn their huge infrastructure into a cloud-computing environment and aggressively recruit businesses to run applications on their platforms. To enforce security and privacy on such a service model, we need to protect the data running on the platform. Unfortunately, traditional encryption methods that aim at providing "unbreakable" protection are often not adequate because they do not support the execution of applications such as database queries on the encrypted data. In this paper we discuss the general problem of secure computation on an encrypted database and propose a SCONEDB Secure Computation ON an Encrypted DataBase) model, which captures the execution and security requirements. As a case study, we focus on the problem of k-nearest neighbor (kNN) computation on an encrypted database. We develop a new asymmetric scalar-product-preserving encryption (ASPE) that preserves a special type of scalar product. We use APSE to construct two secure schemes that support kNN computation on encrypted data; each of these schemes is shown to resist practical attacks of a different background knowledge level, at a different overhead cost. Extensive performance studies are carried out to evaluate the overhead and the efficiency of the schemes.

Finding frequent itemsets is the most costly task in association rule mining. Outsourcing this task to a service provider brings several benefits to the data owner such as cost relief and a less commitment to storage and computational resources. Mining results, however, can be corrupted if the service provider (i) is honest but makes mistakes in the mining process, or (ii) is lazy and reduces costly computation, returning incomplete results, or (iii) is malicious and contaminates the mining results. We address the integrity issue in the outsourcing process, i.e., how the data owner verifies the correctness of the mining results. For this purpose, we propose and develop an audit environment , which consists of a database transformation method and a result verification method. The main component of our audit environment is an artificial itemset planting (AIP) technique. We provide a theoretical foundation on our technique by proving its appropriateness and showing probabilistic guarantees about the correctness of the verification process. Through analytical and experimental studies, we show that our technique is both effective and efficient.

Order-preserving submatrices (OPSM's) have been shown useful in capturing concurrent patterns in data when the relative magnitudes of data items are more important than their absolute values. To cope with data noise, repeated experiments are often conducted to collect multiple measurements. We propose and study a more robust version of OPSM, where each data item is represented by a set of values obtained from replicated experiments. We call the new problem OPSM-RM (OPSM with repeated measurements). We define OPSM-RM based on a number of practical requirements. We discuss the computational challenges of OPSM-RM and propose a generic mining algorithm. We further propose a series of techniques to speed up two time-dominating components of the algorithm. We clearly show the effectiveness of our methods through a series of experiments conducted on real microarray data.

We study the problem of clustering uncertain objects whose locations are described by probability density functions (pdf). We show that the UK-means algorithm, which generalises the k-means algorithm to handle uncertain objects, is very inefficient. The inefficiency comes from the fact that UK-means computes expected distances (ED) between objects and cluster representatives. For arbitrary pdf's, expected distances are computed by numerical integrations, which are costly operations. We propose pruning techniques that are based on Voronoi diagrams to reduce the number of expected distance calculation. These techniques are analytically proven to be more effective than the basic bounding-box-based technique previous known in the literature. We conduct experiments to evaluate the effectiveness of our pruning techniques and to show that our techniques significantly outperform previous methods.

Abstract. Many kinds of real-life data exhibit logical ordering among their data items and are thus sequential in nature. However, traditional online analytical processing (OLAP) systems and techniques were not designed for sequence data and they are incapable of supporting sequence data analysis. In this paper, we propose the concept of Sequence OLAP, or S-OLAP for short. The biggest distinction of S-OLAP from traditional OLAP is that a sequence can be characterized not only by the attributes ’ values of its constituting items, but also by the subsequence/substring patterns it possesses. This paper studies many aspects related to Sequence OLAP. The concepts of sequence cuboid and sequence data cube are introduced. A prototype S-OLAP system is built in order to validate the proposed concepts. The prototype is able to support “pattern-based ” grouping and aggregation, which is currently not supported by any OLAP system. The implementation details of the prototype system as well as experimental results are presented. 1

We investigate the problem of frequent itemset mining over a data stream with bursty traffic. In many modern applications, data arrives at a system as a continuous stream of transactions. In many cases, the arrival rate of transactions fluctuates wildly. Traditional stream mining algorithms, such as Lossy Counting (LC), were generally designed to handle data streams with steady data arrival rates. We show that LC suffers significant loss of accuracy when the data stream is bursty. We propose the Adaptive Frequency Counting algorithm (AFC) to handle bursty data. AFC has a feedback mechanism that dynamically adjusts the mining speed to cope with the changing data arrival rate. Through extensive experiments, we show that AFC outperforms LC under bursty traffics in terms of the accuracy of the set of frequent itemsets

We study a problem of mining frequently occurring periodic patterns with a gap requirement from sequences. Given a character sequenceSof lengthLand a patternPof lengthl, we considerPa frequently occurring pattern inSif the probability ofobservingPgiven a randomly picked length-lsubsequence ofSexceeds a certain threshold. In many applications, particularly those related to bioinformatics, interesting patterns areperiodicwith agap requirement. That is to say, the characters inPshould match subsequences ofSin such a way that the matching characters inSare separated by gaps of more or less the same size. We show the complexity of the mining problem and discuss why traditional mining algorithms are computationally infeasible. We propose practical algorithms for solving the problem and study their characteristics. We also present a case study in which we apply our algorithms on some DNA sequences. We discuss some interesting patterns obtained from the case study.

We study the problem of clustering data objects whose locations are uncertain. A data object is represented by an uncertainty region over which a probability density function (pdf) is defined. One method to cluster uncertain objects of this sort is to apply the UK-means algorithm, which is based on the traditional K-means algorithm. In UK-means, an object is assigned to the cluster whose representative has the smallest expected distance to the object. For arbitrary pdf, calculating the expected distance between an object and a cluster representative requires expensive integration computation. We study various pruning methods to avoid such expensive expected distance calculation.

We study a problem of mining frequently occurring periodic patterns with a gap requirement from sequences. Given a character sequence S of length L and a pattern P of length l, we consider P a frequently occurring pattern in S if the probability of observing P given a randomly picked length-l subsequence of S exceeds a certain threshold. In many applications, particularly those related to bioinformatics, interesting patterns are periodic with a gap requirement. That is to say, the characters in P should match subsequences of S in such a way that the matching characters in S are separated by gaps of more or less the same size. We show the complexity of the mining problem and discuss why traditional mining algorithms are computationally infeasible. We propose practical algorithms for solving the problem, and study their characteristics. We also present a case study in which we apply our algorithms on some DNA sequences. We discuss some interesting patterns obtained from the case study.

We introduce a new type of KDD patterns called emerging substrings. In a sequence database, an emerging substring (ES) of a data class is a substring which occurs more frequently in that class rather than in other classes. ESs are important to sequence classification as they capture significant contrasts between data classes and provide insights for the construction of sequence classifiers. We propose a suffix tree-based framework for mining ESs, and study the effectiveness of applying one or more pruning techniques in different stages of our ES mining algorithm. Experimental results show that if the target class is of a small population with respect to the whole database, which is the normal scenario in single-class ES mining, most of the pruning techniques would achieve considerable performance gain.

This paper introduces an efficient method for the maintenance of wavelet-based histograms built on partial sums. Wavelet-based histograms can be constructed from either raw data distributions or partial sums. The two construction methods have their own merits. Previous works have only focused on the maintenance of raw-data-based histograms. However it is highly inefficient to apply directly their techniques to partial-sum-based histograms because a single data update would trigger changes of multiple partial sums, which in turn, would trigger large amounts of computation on the changes of the wavelet decomposition. We present a novel technique to compute the effects of data updates on the wavelet decomposition of partial sums. Moreover, we point out some special features of the wavelet decomposition of partial sums and adapt a probabilistic counting technique for the maintenance of partial-sum-based histograms. Experimental results show that our maintenance method is efficient and its accuracy is robust to changing data distributions.

A real-time database system contains base data items which record and model a physical, real-world environment. For better decision support, base data items are summarized and correlated to derive views. These base data and views are accessed by application transactions to generate the ultimate actions taken by the system. As the environment changes, updates are applied to base data, which subsequently trigger view recomputations. There are thus three types of activities: base data update, view recomputation, and transaction execution. In a real-time database system, two timing constraints need to be enforced. We require that transactions meet their deadlines (transaction timeliness) and read fresh data (data timeliness). In this paper, we define the concept of absolute and relative temporal consistency from the perspective of transactions for discrete data objects. We address the important issue of transaction scheduling among the three types of activities such that the two timing requirements can be met. We also discuss how a real-time database system should be designed to enforce different levels of temporal consistency.

The problem of mining frequent sequences is to extract frequently occurring subsequences in a sequence database. Algorithms on this mining problem include GSP, MFS, and SPADE. The problem of incremental update of frequent sequences is to keep track of the set of frequent sequences as the underlying database changes. Previous studies have extended the traditional algorithms to efficiently solve the update problem. These incremental algorithms include ISM, GSP+ and MFS+. Each incremental algorithm has its own characteristics and they have been studied and evaluated separately under different scenarios. This paper presents a comprehensive study on the relative performance of the incremental algorithms as well as their non-incremental counterparts. Our goal is to provide guidelines on the choice of an algorithm for solving the incremental update problem given the various characteristics of a sequence database.

One major approach to music retrieval is to model music as a sequence of features, after which traditional information retrieval techniques are applied on the sequence. Because of the temporal nature of music and the inexactness of user queries, most effort on music retrieval systems focus on issues such as indexing and approximation match. In contrast, the processing of music before feature extraction, such as the identification of a melody track, were often considered easy or done. This may be the case in a controlled environment, such as one for musicology research, where the pieces are carefully analyzed by human beings before being submitted to the database. However, in an environment where large volumes of music is obtained from the Web, manual music analysis is impractical. Since many well-known musical features often pertain to the melody of musical pieces, and users often remember the melody of a song, algorithms that select the melody tracks of a piece are important for Web-based content-based retrieval systems. We describe a number of algorithms for automatic melody track selection in a music retrieval context. We also study the performance of the algorithms by comparing their answers to those judged by human beings.

MOTIVATION: The main goal in this paper is to develop accurate probabilistic models for important functional regions in DNA sequences (e.g. splice junctions that signal the beginning and end of transcription in human DNA). These methods can subsequently be utilized to improve the performance of gene-finding systems. The models built here attempt to model long-distance dependencies between non-adjacent bases. RESULTS: An efficient modeling method is described which models biological data more accurately than a first-order Markov model without increasing the number of parameters. Intuitively, a small number of parameters helps a learning system to avoid overfitting. Several experiments with the model are presented, which show a small improvement in the average accuracy as compared with a simple Markov model. These experiments suggest that single long distance dependencies do not help the recognition problem, thus confirming several previous studies which have used more heuristic modeling techniques. AVAILABILITY: This software is available for downloaded and as a web resource at http://www.ai.uic.edu/software CONTACT: [email protected]

Traditional World Wide Web search engines, such as AltaVista.com, index and recommend individual Web pages to assist users in locating relevant documents. As the Web grows, however, the number of matching pages increases at a tremendous rate. Users are often overwhelmed by the large answer set recommended by the search engines. Also, if a matching document is a hypertext, the document structure is destroyed and the individual pages that compose the document are returned instead. The logical starting point of the hyperdocument is thus hidden among the large basket of matching pages. Users need to spend a lot of effort browsing through the pages to locate the starting point, a very time consuming process. This paper studies the anchor point indexing problem. The set of anchor points of a given user query is a small set of key pages from which the larger set of documents that are relevant to the query can be easily reached. The use of anchor points helps solve the problems of huge answer set and low precision suffered by most search engines by considering the hyperlink structures of the relevant documents, and by providing a summary view of the result set.

On-line analytical processing (OLAP) requires efficient processing of complex decision support queries over very large databases. It is well accepted that pre-computed data cubes can help reduce the response time of such queries dramatically.Avery important design issue of an efficient OLAP system is therefore the choice of the right data cubes to materialize. We call this problem the data cube schema design problem. In this paper we show that the problem of finding an optimal data cube schema for an OLAP system with limited memory is NP-hard. As a more computationally efficient alternative, we propose a greedy approximation algorithm cMP and its variants. Algorithm cMP consists of two phases. In the first phase, an initial schema consisting of all the cubes required to efficiently answer the user queries is formed. In the second phase, cubes in the initial schema are selectively merged to satisfy the memory constraint. We show that cMP is very effective in prunning the search space for an optimal schema. This leads to a highly efficient algorithm. We report

A database system contains base data items which record and model a physical, real world environment. For better decision support, base data items are summarized and correlated to derive views. These base data and views are accessed by application transactions to generate the ultimate actions taken by the system. As the environment changes, updates are applied to the base data, which subsequently trigger view recomputations. There are thus three types of activities: base data update, view recomputation, and transaction execution. In a real-time system, two timing constrains need to be enforced. We require transactions meet their deadlines (transaction timeliness) and read fresh data (data timeliness). In this paper we define the concept of absolute and relative temporal consistency from the perspective of transactions. We address the important issue of transaction scheduling among the three types of activities such that the two timing requirements can be met. We also discuss how a real-time database system should be designed to enforce different levels of temporal consistency.

Traditional WWW search engines index and recommend individual Web pages to assist users in locating relevant documents. Users are often overwhelmed by the large answer set recommended by the search engines. The logical starting point of the hyper-document is thus hidden among the large basket of matching pages. Users need to spend a lot of effort browsing through the pages to locate the starting point, a very time consuming process. The paper studies the anchor point indexing problem. The anchor points of a given user query is a small set of key pages from which the larger set of documents that are relevant to the query can be easily reached. The use of anchor points help solve the problems of huge answer set and low precision suffered by most search engines by considering the hyper-link structures of the relevant documents, and by providing a summary view of the result set.

The Web connects people speaking more than twenty languages in more than one hundred countries. Search engines, which provide users with starting points to navigate and retrieve resources an the Web, should thus be able to handle documents in many languages. Moreover, with information being added and changed every minute on the Web, search engines should discover new index terms time-efficiently. The paper introduces an abstraction of viewing multilingual documents and a statistical analysis method so that search engines can index multilingual documents in a generic, efficient, and effective manner with minimal requirements of language-specific information.

Previous studies, e.g., Haritsa et al. (Haritsa, J.R., Livny, M., Carey, M., 1990. Proceedings of Ninth ACM Symposium on Principles of Database systems) have shown that optimistic concurrency control (OCC) generally performs better than lock-based protocols in disk-based real-time database systems (RTDBS). In this paper we compare the two concurrency control protocols in both disk-based and memory-resident multiprocessor RTDBS. Based on simulation, we analyze the intrinsic behaviors of the two protocols. The result of our performance evaluation experiments show that different characteristics of the two environments indeed have great impact on the protocols' performance. We identify such system characteristics and expose the weaknesses of traditional OCC and lock-based protocols. To improve performance, a new protocol, called Two Phase Locking-Lock Write All (2PL-LW), is proposed. We show that 2PL-LW performs better than the traditional protocols in meeting transaction deadlines in both disk-based and memory-resident RTDBS.

In a distributed environment, tasks often have processing demands at multiple different sites. A distributed task is usually divided into several subtasks, each to be executed in order at some site. In a real-time system, an overall deadline is usually specified by an application designer indicating when a distributed task is to be finished. In this paper, we present and analyze techniques for automatically translating the overall deadline into deadlines for the individual subtasks.

In this paper, we consider soft real-time systems with redundant off-the-shelf processing components (e.g., CPU, disk, network), and show how applications can exploit the redundancy to improve the system's ability of meeting response time goals (soft deadlines). We consider two scheduling policies, one that evenly distributes load (Balance), and one that partitions load according to job slackness (Chop). We evaluate the effectiveness of these policies through analysis and simulation. Our results show that by intelligently distributing jobs by their slackness amount the servers, Chop can significantly improve real-time performance.

Many papers have examined how to efficiently export a materialized view but to our knowledge none have studied how to efficiently import one. To import a view, i.e., to install a stream of updates, a real-time database system must process new updates in a timely fashion to keep the database "fresh," but at the same time must process transactions and ensure they meet their time constraints. In this paper, we discuss the various properties of updates and views (including staleness) that affect this tradeoff. We also examine, through simulation, four algorithms for scheduling transactions and installing updates in a soft real-time database.

Complex distributed tasks often involve parallel execution of subtasks at different nodes. To meet the deadline of a global task, all of its parallel subtasks have to be finished on time. Compared to a local task (which involves execution at only one node), a global task may have a much harder time making its deadline because it is fairly likely that at least one of its subtasks run into an overloaded node. Another problem with complex distributed tasks occurs when a global task consists of a number of serially executing subtasks. In this case, we have the problem of dividing up the end-to-end deadline of the global task and assigning them to the intermediate subtasks. In this paper, we study both of these problems. Different algorithms for assigning deadlines to subtasks are presented and evaluated.>

Real-time scheduling algorithms are usually only available in the kernels of real-time operating systems, and not in more general purpose operating systems, like Unix. For some soft real-time problems, a traditional operating system may be the development platform of choice. This paper addresses methods of emulating real-time scheduling algorithms on top of standard time-share schedulers. We examine (through simulations) three strategies for priority assignment within a traditional multi-tasking environment. The results show that the emulation algorithms are comparable in performance to the real-time algorithms and in some instances outperform them.>

Fault-tolerant computer systems have redundant paths connecting their components. Given these paths, it is possible to use aggressive techniques to reduce the average value and variability of the response time for short, critical messages. One technique is to send a copy of a packet over an alternate path before it is known whether the first copy failed or was delayed. A second technique is to split a single stream of packets over multiple paths. The authors analyze both approaches and show that they can provide significant improvements over conventional, conservative mechanisms.>

In a distributed environment, tasks often have processing demands on multiple sites. A distributed task is usually divided up into several subtasks, each one to be executed at some site in order. In a real-time system, an overall deadline is usually specified by an application designer indicating when a distributed task is to be finished. To study the subtask deadline assignment problem a simple model of the system and tasks is postulated. The focus is on soft real-time systems. In such systems, it is very difficult to guarantee that all deadlines will be met, and hence one tries to minimize the number of deadlines that are missed. The authors examine (through simulations) four strategies for subtask deadline assignment in a distributed soft real-time environment.>

The Gold Mailer, a system that provides users with an integrated way to send and receive messages using different media, efficiently store and retrieve these messages, and access a variety of sources of other useful information, is described. The mailer solves the problems of information overload, organization of messages and multiple interfaces. By providing good storage and retrieval facilities, it can be used as a powerful information processing engine covering a range of useful office information. The Gold Mailer's query language, indexing engine, file organization, data structures, and support of mail message data and multimedia documents are discussed.>

In a 1976 book, “Algorithms + Data Structures = Programs” [15], Niklaus Wirth defined programs to be algorithms and data structures. Of course, by now we know that man does not live from programs alone, and that there is a second fundamental computer science equation: “Programs + Databases = Information Systems.”

Fault-tolerant computer systems have redundant paths connecting their components. Given these paths, it is possible to use aggressive techniques to reduce the average value and variability of the response time for critical messages. One technique is to send a copy of a packet over an alternate path before it is known if the first copy failed or was delayed. A second technique is to split a single stream of packets over multiple paths. The authors analyze both approaches and show that these techniques can provide significant improvements over conventional, conservative mechanisms.>