Heterogeneous graph embedding has recently advanced with two primary directions: self-supervised learning to address the scarcity of labeled data and metapath-free approaches to eliminate reliance on handcrafted, domain-specific metapaths. While metapath-free methods capture structural and semantic information directly from neighbor relationships, their performance often remains unstable across datasets and varying label scales, as they still rely on semi-supervised training. Conversely, self-supervised learning techniques, such as contrastive learning and masked graph autoencoders, achieve remarkable results by extracting information intrinsically from the data. However, these approaches often depend on predefined metapaths, which limit their ability to generalize and constrain representation quality. This paper proposes Relation-aware Heterogeneous Masked Graph Autoencoder (RHMGA), a novel metapathfree self-supervised learning framework that bridges these gaps. Our method integrates a semantic relation network with masked graph autoencoders to jointly reconstruct node features and relation-specific adjacency matrices. By capturing both structural and semantic information during aggregation and reconstruction, RHMGA enables robust self-supervised pretraining without predefined metapaths. Extensive experiments demonstrate that our approach consistently outperforms state-of-the-art methods, achieving superior performance and adaptability across diverse heterogeneous graph datasets.

Existing algorithms for mining high utility patterns over a data stream are two-phase algorithms that are not scalable due to the large number of candidates generation in the first phase, particularly when the minimum utility threshold is low. Moreover, in the second phase, the algorithm needs to scan the database again to find out actual utility for candidates. In this article, we propose one-phase algorithm SOHUPDS \(+\) to mine high utility itemsets in the current sliding window of the data stream with respect to absolute or relative minimum utility threshold. To facilitate SOHUPDS \(+\) , we propose a data structure IUDataListSW \(+\) , which stores and maintains utility and upper-bound values of the items in the current sliding window when sliding window advances. In addition, we propose a transaction merging strategy, called BitmapTransactionMerging , which saves execution time for utility and upper-bound values computations in denser datasets. Moreover, we propose update strategies to utilize mined high utility patterns from the previous sliding window to update high utility patterns in the current sliding window. The results of experiments illustrate that SOHUPDS \(+\) is more efficient than the state-of-the-art algorithms in terms of execution time as well as memory usage in most of the experiments on various datasets.

Despite considerable research into information diffusion, most models focus on static networks. Networks in the real world change over time and information is lost when a node disappears. We sought to resolve the problem of information loss by defining information and information loss in general terms. Experiments demonstrate the efficacy of the proposed scheme in resolving problems of information loss in real-world networks using only a few edges.

Natural language understanding models often make use of surface patterns or idiosyncratic biases in a given dataset to make predictions pertaining to natural language inference (NLI) tasks. Unfortunately, this renders the resulting model vulnerable to out-of-distribution datasets to which the identified features are inapplicable, thereby leading to erroneous results. Many of the methods developed for out-of-distribution datasets have proven effective; however, they also tend to impose a tradeoff in performance when applied to in-distribution datasets. In this paper, we use a teacher model providing knowledge for the student ensemble model as basic information for training. The student ensemble model then integrates information of deep and shallow representations to extend learning performance to a wide range of examples. The evaluation demonstrates that the proposed model outperformed state-of-the-art models when applied to in-distribution as well as out-of-distribution datasets.

Most previous research on influence maximization has focused on static social networks, despite the dynamic nature of networks in the real world. The computational cost imposed by recalculating results in response to dynamic changes precludes the use of conventional updating algorithms when dealing with large-scale networks. In this study, we developed a novel approach to estimating the influence of nodes through the creation of Influence SubGraphs. We also developed methods by which to update Influence SubGraphs to overcome the problem of influence maximization in dynamic social networks. Experiment results demonstrated the efficacy of the proposed scheme, in achieving influence propagation performance comparable to that of state-of-the-art methods with far lower memory requirements and far shorter execution times.

Most sequential recommendation systems, including those that employ a variety of features and state-of-the-art network models, tend to favor items that are the most popular or of greatest relevance to the historic behavior of the user. Recommendations made under these conditions tend to be repetitive; i.e., many options that might be of interest to users are entirely disregarded. This paper presents a novel algorithm that assigns a novelty score to potential recommendation items. We also present an architecture by which to incorporate this functionality in existing recommendation systems. In experiments, the proposed NASM system outperformed state-of-the-art sequential recommender systems, thereby verifying that the inclusion of novelty score can indeed improve recommendation performance.

Social network-based team formation problem has been widely studied from different aspects. However, the skills in earlier works were treated equally, and cannot be substituted by other related or similar skills. In addition, assigning experts who possess alternative skills for a required skill is not allowed. To better fit real world scenarios, we propose a novel hierarchical skill model to let skills interchangeable. By considering the communication cost and the personnel cost, we develop an optimization framework under the hierarchical skill model to deal with the trade-off between communication and personnel cost. The experiments show that our proposed framework and the hierarchical skill model is reasonable and has better performance than earlier works.

Recommendation systems which are designed to understand and predict user interest based on user preferences play an important role in the era of information explosion. We propose the item influence embedding which adopts the social influence diffusion concept to model the item relations. We can learn the activation paths in items-item relation graph. In addition, for generating top-k items, most of recommendation systems calculate the similarity between user embedding and embedding of all items. The calculation costs too much time when number of users and items are huge. Therefore, we propose the User Preference Translation Model (UPTM) to recommend the Top-k items based on the language translation technology. UPTM directly generates the recommendation items based on translating the user preference. We can avoid to calculate the similarity of user embedding and item embedding. From the experimental results, UPTM not only outperforms the compared methods but also save the time in real large datasets.

Influence maximization refers to the process of identifying a predefined number of nodes within a given social network with the aim of maximizing the spread of influence. Most previous work has focused on unsigned networks, which means the existence of polarity relationships has largely been disregarded. In this work, we define a Sign-aware Influence Maximization (SIM) problem, which involves identification of the seed set that would simultaneously maximize positive influence and minimize negative influence. We begin by considered competitive influence under various dominance mechanisms on SCIC model, which extends the classic Independent Cascade (IC) model by incorporating binary opinions and signed relationships. We then proved that the influence of SIM under the SCIC model is non-monotonic and non-submodular, which implies that simple greedy hill-climbing would be unable to achieve an approximation ratio of 1-1/e in seeking to resolve the SIM problem. We then developed a simulation-based algorithm called Sign-aware Competitive Maximum Influence Arborescence (S-CMIA) to simulate the propagation of influence within a local region. Experiment results demonstrate the superiority of the proposed algorithm over existing methods in resolving the SIM problem in terms of reward.

In influence maximization problem, we would like to find an initial subset of nodes in a given graph, which maximizes the final number of affected nodes through "word of mouth" propagation. Measuring the influence spread of set of seed nodes and gradually selecting the node with largest marginal increase is one of the main approaches of existing algorithms. In this paper, we try to solve this problem from a different strategy - in an improvement perspective. A more complex condition is depicted as both positive and negative opinions are propagating in the social network. The objective function considers the maximization of the positive influence and minimizes the negative opinion spreading simultaneously. We propose IDR (Influence Distribution Redirection) algorithm to define initial seed nodes of influence diffusion based on redirecting the influence distribution of nodes to maximize the objective function. The influence distribution of nodes shows the potential influence trend of nodes during the influence diffusion process. The key strategy is reducing the positive influence nearby the steady nodes and increasing in the vacillate region. From the experimental results, IDR outperforms the compared method on the objective function. In addition, IDR also improves the performance of increasing the number of positive active nodes and decreasing the number of negative activated nodes respectively.

Sequential pattern mining is used to find frequent data sequences over time. When sequential patterns are generated, the newly arriving patterns may not be identified as frequent sequential patterns due to the existence of old data and sequences. Progressive sequential pattern mining aims to find the most up-to-date sequential patterns given that obsolete items will be deleted from the sequences. When sequences come with multiple data streams, it is difficult to maintain and update the current sequential patterns. Even worse, when we consider the sequences across multiple streams, previous methods cannot efficiently compute the frequent sequential patterns. In this work, we propose an efficient algorithm PSP-AMS to address this problem. PSP-AMS uses a novel data structure PSP-MS-tree to insert new items, update current items, and delete obsolete items. By maintaining a PSP-MS-tree, PSP-AMS efficiently finds the frequent sequential patterns across multiple streams. The experimental results show that PSP-AMS significantly outperforms previous algorithms for mining of progressive sequential patterns across multiple streams on synthetic data as well as real data.

The influence maximization problem has been considered a vital problem when companies provide similar products or services. Since there are limited resources, companies must determine a strategy to occupy as much market share as possible. In this paper, we propose a general Deterministic Network Adaptive (DNA) framework to solve the multi-round multi-party influence maximization problem. To obtain the most market share, using one single strategy to determine seed nodes is not sufficient in the long term. The reason is that the network status changes during the multi-round procedure. The strategies of selecting seed nodes in each round should depend on the current status of influence diffusion in the network. DNA framework leverages the concept of reinforcement learning to maximize the expected cumulative influence. In addition, the learning process is deterministic, so that it does not take time to explore the spaces that are less important. We further design a similarity function to measure the similarity between two networks. DNA framework can avoid redundant computation when the similar networks have been trained before. Moreover, we propose the method to make the policy decision to maximize the influence spread in coopetition scenario based on DNA framework. The proposed framework is evaluated with synthetic data and real-world data. From the experimental results, DNA framework outperforms the existing works in influence maximization problems. The coopetition policy which is generated by DNA has the best performance in most cases.

Previous works of subsequence search in time interval-based events sequences have focused on the relations among events without considering the duration of each event. However, the same event with different time duration may lead to different results. In this work, we propose an index structure referred to as Endpoint Index based on the concept of inverted index to efficiently extract the Time Interval-based Event with Duration, TIED, subsequence. TIED subsequence search considers duration of event intervals in conjunction with relation among events. This makes the results of TIED subsequence search more accurate than those obtained using traditional search methods. In addition, we propose an algorithm SSD, Subsequence Search with Duration, incorporating a pruning strategy to search TIED subsequences efficiently. For the performance evaluation, we modify previous algorithms to compare with our proposed methods SSD_Endpoint and SSD_ER. The experimental results demonstrate that SSD_Endpoint and SSD_ER are more efficient than state-of-the-art algorithms.

Since the tremendous success of social networking websites, the related analytical research has been widely studied. Among these studies, social influence has been a significant and popular topic. We rely on the social influence model to predict and learn the influence diffusion process. However, traditional models only categorize nodes into two types of states, active and inactive. In addition, most previous models have only taken positive influences into account. Moreover, if inactive nodes are influenced successfully and turn into active nodes, these nodes cannot change their states forever. In this work, we not only break the above limitations but also propose a novel propagation method in our model. We proposes five states to represent the multiple states of influence. According to the new propagation method, the strength of the social influence may be reduced over time. Eventually, we utilize the measurement of precisions to compare with related models. The proposed multi-state model outperforms other two-state models in precisions of prediction. The experimental results show the superiority of multiple states.

The disappearance of important nodes which are prominent characters in a social network may lead the social network to a broken structure. Many previous works have discussed reconstructing such networks using the network topology to devise an approach that finds a substitute node for a deleted node and generates appropriate links to avoid a fragmentation of the network. A common used property in finding substitute node is centrality, but calculating some kinds of centrality may spend too much time on re-scanning the graph. Thus, we propose a local approach, CLOMADE, standing for Choosing LOcal MAximum DEgree. We only need to scan the whole graph once for calculating degree. We choose a node with local maximum degree to be the substitute node and generate new links from the substitute node to other nodes. The experiments show that CLOMADE outperforms previous works in execution time.

Time interval-based pattern mining is proposed to improve the lack of the information of time intervals by sequential pattern mining. Previous works of time interval-based pattern mining focused on the relations between events without considering the duration of each event. However, the same event with different time durations will cause definitely different results. For example, if some people cough for one week, they may get a cold for a while. In contrast, if some patients cough for one year, they may get pneumonia in the future. In this work, we propose two algorithms, SARA and SARS, to extract the frequent Time Interval-based Event with Duration, TIED, patterns. TIED patterns not only keep the relations between two events but also reveal the time periods when each event happens and ends. In the experiments, we propose a naive algorithm and modify a previous algorithm to compare the performances with SARA and SARS. The experimental results show that SARA and SARS are more efficient in execution time and memory usage than other two algorithms.

Instead of finding clusters in the full feature space, subspace clustering is an emergent task which aims at detecting clusters embedded in subspaces. Most of previous works in the literature are density-based approaches, where a cluster is regarded as a high-density region in a subspace. However, the identification of dense regions in previous works lacks of considering a critical problem, called "the density divergence problemrdquo in this paper, which refers to the phenomenon that the region densities vary in different subspace cardinalities. Without considering this problem, previous works utilize a density threshold to discover the dense regions in all subspaces, which incurs the serious loss of clustering accuracy (either recall or precision of the resulting clusters) in different subspace cardinalities. To tackle the density divergence problem, in this paper, we devise a novel subspace clustering model to discover the clusters based on the relative region densities in the subspaces, where the clusters are regarded as regions whose densities are relatively high as compared to the region densities in a subspace. Based on this idea, different density thresholds are adaptively determined to discover the clusters in different subspace cardinalities. Due to the infeasibility of applying previous techniques in this novel clustering model, we also devise an innovative algorithm, referred to as DENCOS (density conscious subspace clustering), to adopt a divide-and-conquer scheme to efficiently discover clusters satisfying different density thresholds in different subspace cardinalities. As validated by our extensive experiments on various data sets, DENCOS can discover the clusters in all subspaces with high quality, and the efficiency of DENCOS outperformes previous works.

Although there have been many recent studies on the mining of sequential patterns in a static database and in a database with increasing data, these works, in general, do not fully explore the effect of deleting old data from the sequences in the database. When sequential patterns are generated, the newly arriving patterns may not be identified as frequent sequential patterns due to the existence of old data and sequences. Even worse, the obsolete sequential patterns that are not frequent recently may stay in the reported results. In practice, users are usually more interested in the recent data than the old ones. To capture the dynamic nature of data addition and deletion, we propose a general model of sequential pattern mining with a progressive database while the data in the database may be static, inserted, or deleted. In addition, we present a progressive algorithm Pisa, which stands for progressive mining of sequential patterns, to progressively discover sequential patterns in defined time period of interest (POI). The POI is a sliding window continuously advancing as the time goes by. Pisa utilizes a progressive sequential tree to efficiently maintain the latest data sequences, discover the complete set of up-to-date sequential patterns, and delete obsolete data and patterns accordingly. The height of the sequential pattern tree proposed is bounded by the length of POI, thereby effectively limiting the memory space required by Pisa that is significantly smaller than the memory needed by the alternative method, direct appending (DirApp). Note that the sequential pattern mining with a static database and with an incremental database are special cases of the progressive sequential pattern mining. By changing start time and end time of the POI, Pisa can easily deal with a static database or an incremental database as well. Complexity of algorithms proposed is analyzed. The experimental results show that Pisa not only significantly outperforms the prior methods in execution time by orders of magnitude but also possesses graceful scalability.

Generally speaking, to implement Apriori-based association rule mining in hardware, one has to load candidate itemsets and a database into the hardware. Since the capacity of the hardware architecture is fixed, if the number of candidate itemsets or the number of items in the database is larger than the hardware capacity, the items are loaded into the hardware separately. The time complexity of those steps that need to load candidate itemsets or database items into the hardware is in proportion to the number of candidate itemsets multiplied by the number of items in the database. Too many candidate itemsets and a large database would create a performance bottleneck. In this paper, we propose a HAsh-based and Pipelined (abbreviated as HAPPI) architecture for hardware- enhanced association rule mining. We apply the pipeline methodology in the HAPPI architecture to compare itemsets with the database and collect useful information for reducing the number of candidate itemsets and items in the database simultaneously. When the database is fed into the hardware, candidate itemsets are compared with the items in the database to find frequent itemsets. At the same time, trimming information is collected from each transaction. In addition, itemsets are generated from transactions and hashed into a hash table. The useful trimming information and the hash table enable us to reduce the number of items in the database and the number of candidate itemsets. Therefore, we can effectively reduce the frequency of loading the database into the hardware. As such, HAPPI solves the bottleneck problem in a priori-based hardware schemes. We also derive some properties to investigate the performance of this hardware implementation. As shown by the experiment results, HAPPI significantly outperforms the previous hardware approach and the software algorithm in terms of execution time.

We investigate the general model of mining associations in a temporal database, where the exhibition periods of items are allowed to be different from one to another. The database is divided into partitions according to the time granularity imposed. Such temporal association rules allow us to observe short-term but interesting patterns that are absent when the whole range of the database is evaluated altogether. Prior work may omit some temporal association rules and thus have limited practicability. To remedy this and to give more precise frequent exhibition periods of frequent temporal itemsets, we devise an efficient algorithm Twain (standing for TWo end AssocIation miNer .) Twain not only generates frequent patterns with more precise frequent exhibition periods, but also discovers more interesting frequent patterns. Twain employs Start time and End time of each item to provide precise frequent exhibition period while progressively handling itemsets from one partition to another. Along with one scan of the database, Twain can generate frequent 2-itemsets directly according to the cumulative filtering threshold. Then, Twain adopts the scan reduction technique to generate all frequent k -itemsets ( k > 2) from the generated frequent 2-itemsets. Theoretical properties of Twain are derived as well in this article. The experimental results show that Twain outperforms the prior works in the quality of frequent patterns, execution time, I/O cost, CPU overhead and scalability.

In this paper, a problem, called "the density divergence problem" is explored. This problem is related to the phenomenon that the densities of the clusters vary in different subspace cardinalities. We take the densities into consideration in subspace clustering and explore an algorithm to adaptively determine different density thresholds to discover clusters in different subspace cardinalities.

When sequential patterns are generated, the newly arriving patterns may not be identified as frequent sequential patterns due to the existence of old data and sequences. In practice, users are usually more interested in the recent data than the old ones. To capture the dynamic nature of data addition and deletion, we propose a general model of sequential pattern mining with a progressive database. In addition, we present a progressive concept to progressively discover sequential patterns in recent time period of interest.

In the data stream environment, the patterns generated at different time instances are different due to data evolution. As time progresses, the behavior and members of clusters usually change. Hence, clustering continuous data streams allows us to observe the changes of group behavior. In order to support flexible clustering requirements, we devise in this paper a Clustering on Demand framework, abbreviated as COD framework, to dynamically cluster multiple data streams. While providing a general framework of clustering on multiple data streams, the COD framework has two advantageous features, namely, one data scan for online statistics collection and compact multiresolution approximations, which are designed to address, respectively, the time and the space constraints in a data stream environment. The COD framework consists of two phases, i.e., the online maintenance phase and the offline clustering phase. The online maintenance phase provides an efficient mechanism to maintain summary hierarchies of data streams with multiple resolutions in time linear in both the number of streams and the number of data points in each stream. On the other hand, an adaptive clustering algorithm is devised for the offline phase to retrieve approximations of desired substreams from summary hierarchies according to clustering queries. We propose two summarization techniques, based on wavelet and regression analyses, to construct the summary hierarchies. The regression-based summary hierarchy approximates the data stream more precisely and provides better clustering results, at the cost of slightly longer time than and twice the storage space as the wavelet-based one. An adaptive version of COD framework is designed to make a selection between a wavelet-based model and a regression-based model for building the summary hierarchy. By the adaptive COD, we can obtain clustering results with almost the same quality as the regression-based COD while using much less storage space for the summary hierarchy. As shown in the complexity analyses and also validated by our empirical studies, the COD framework performs very efficiently in the data stream environment while producing clustering results of very high quality.

In the data stream environment, the patterns generated by the mining techniques are usually distinct at different time because of the evolution of data. In order to deal with various types of multiple data streams and to support flexible mining requirements, we devise in this paper a clustering on demand framework, abbreviated as COD framework, to dynamically cluster multiple data streams. While providing a general framework of clustering on multiple data streams, the COD framework has two major features, namely one data scan for online statistics collection and compact multiresolution approximations, which are designed to address, respectively, the time and the space constraints in a data stream environment. Furthermore, with the multiresolution approximations of data streams, flexible clustering demands can be supported.