Meta Paths Research Articles

Developing a BERT based triple classification model using knowledge graph embedding for question answering system

The current BERT-based question answering systems use a question and a contextual text to find the answer. This causes the systems to return wrong answers or nothing if the text contains irrelevant contents with the input question. Besides, the systems haven’t answered yes-no and aggregate questions yet. Besides that, the systems only concentrate on the contents of text regardless of the relationship between entities in the corpus. This systems cannot validate the answer. In this paper, we presented a solution to solve these issues by using the BERT model and the knowledge graph to enhance a question answering system. We combined content-based and linked-based information for knowledge graph representation learning and classified triples into one of three classes such as base class, derived class, or non-existent class. We then used the BERT model to build two classifiers: BERT-based text classification for content information and BERT-based triple classification for link information. The former was able to make a contextual embedding vector for representing triples that were used to classify into the three above classes. The latter generated all path instances from all meta paths of a large heterogeneous information network by running the Motif Search method of Apache Spark on a distributed environment. After creating the path instances, we produced triples from these path instances. We made content-based information by converting triples into natural language text with labels and considered them as a text classification problem. Our proposed solution outperformed other embedding methods with an average accuracy of 92.34% on benchmark datasets and the Motif Finding algorithm with an average executive time improvement of 37% on the distributed environment.

A computational investigation of the selectivity and mechanism of the Lewis acid catalyzed oxa‐Diels–Alder cycloaddition of substituted diene with benzaldehyde

The selectivity and the mechanism of the uncatalyzed and AlCl3 catalyzed hetero-Diels-Alder reaction (HDR) between ([E]-4-methylpenta-2,4-dienyloxy)(tert-butyl)dimethylsilane 1 and benzaldehyde 2 have been studied using density functional theory at the MPWB1K/6-31G(d) level of theory. The uncatalyzed HDR between diene 1 and alkene 2 is characterized by a polar character and proceeds via an asynchronous one-step mechanism for the meta paths and synchronous for the ortho ones. In the presence of AlCl3 catalyst, the mechanism changes to be stepwise, while the first step is the rate-determining step. The activation energies widely decrease, and the polar character increases dramatically. A large analysis of the mechanism is performed using the activation strain model/energy decomposition analysis (ASM/EDA) model, the natural bond orbital (NBO) and state specific dual descriptors (SSDDs). The obtained results indicate that the combined interaction energy associated with the distortion of the reactants in these HDR are at the origin of the observed kinetics. NBO analyses were applied to estimate the Lewis-acid catalyst donor-acceptor interaction with the molecular system. The SSDD analysis shed light into the orientation effects on the reaction kinetics by providing important information about charge transfer interactions during the chemical reaction. It indicates that the more favorable HDR pathway have the lowest excitation energies, facilitating the interaction between diene 1 and benzaldehyde 2 moieties. Non-covalent interaction (NCI) and QTAIM analyses of the meta-endo structure indicate that the presence of several weak NCIs formed at this approach is at the origin of the meta-endo selectivity.

Uniting Heterogeneity, Inductiveness, and Efficiency for Graph Representation Learning

Recently, bearing the message passing paradigm, graph neural networks(GNNs) have greatly advanced the performance of node representation learning on graphs. However, a majority class of GNNs are only designed for homogeneous graphs, leading to inferior adaptivity to the more informative heterogeneous graphs with various types of nodes and edges. Also, despite the necessity of inductively producing representations for completely new nodes (e.g., in streaming scenarios), few heterogeneous GNNs can bypass the transductive learning scheme where all nodes must be known during training. Furthermore, the training efficiency of most heterogeneous GNNs has been hindered by their sophisticated designs for extracting the semantics associated with each meta path or relation. In this paper, we propose a wide and deep message passing network (WIDEN) to cope with the aforementioned problems about heterogeneity, inductiveness, and efficiency that are rarely investigated together in graph representation learning. In WIDEN, we propose a novel inductive, meta path-free message passing scheme that packs up heterogeneous node features with their associated edges from both low- and high-order neighbor nodes. To further improve the training efficiency, we innovatively present an active down-sampling strategy that drops unimportant neighbor nodes to facilitate faster information propagation.

Distributed Relationship Mining over Big Scholar Data

In this paper, we propose a system infrastructure to construct the big scholar data as a large knowledge graph, discover the meta paths between the entities and calculate the relevancy between entities in the graph. The core infrastructure is established on the secured and private Amazon Elastic Compute Cloud(Amazon EC2) platform. The infrastructure maintains the data evenly across the repositories, processes the data parallel by utilizing open source Spark framework, manages computing resources optimally by utilizing YARN and Hadoop HDFS, and discovers the relationship distributedly between different types of entities. We incorporate four relationship discovery tasks including citation recommendation, potential collaborator discovery, similar venue measurement and paper to venue recommendation on top of this infrastructure. For relationship mining tasks, we propose a mixed and weighted meta path (MWMP) method to explore the potential relationship between different types of entities. To verify the accuracy and measure parallelization speedup of our algorithm, we set up clusters through Amazon EC2 platform.

Activity recommendations based on heterogeneous social networks

EBSN (event-based Social Network) is a Heterogeneous Network, namely, H-EBSN (Heterogeneous event-based Social Network), which contains both online interactive information and offline interactive information. How to effectively use the interaction between online and offline to recommend activities is one of the hot issues at present. The following work has been done for the activity recommendation of H-EBSN:(1) Construct H-EBSN, extract all link-roads and obtain user interest through link-roads’ characteristics.(2) Analyze H-EBSN meta path, take the path number of extracted link-roads as feature vector, establish user interest model, and use generalized regression neural network to solve the model.(3) The training set was constructed by using the movie-watching activity data set, and the generalized regression neural network was used to solve the model. Experiments show that the activity recommendation method based on H-EBSN proposed in this paper has a good recommendation effect.

Mining Similar Aspects for Gene Similarity Explanation Based on Gene Information Network.

Analysis of gene similarity not only can provide information on the understanding of the biological roles and functions of a gene, but may also reveal the relationships among various genes. In this paper, we introduce a novel idea of mining similar aspects from a gene information network, i.e., for a given gene pair, we want to know in which aspects (meta paths) they are most similar from the perspective of the gene information network. We defined a similarity metric based on the set of meta paths connecting the query genes in the gene information network and used the rank of similarity of a gene pair in a meta path set to measure the similarity significance in that aspect. A minimal set of gene meta paths where the query gene pair ranks the highest is a similar aspect, and the similar aspect of a query gene pair is far from trivial. We proposed a novel method, SCENARIO, to investigate minimal similar aspects. Our empirical study on the gene information network, constructed from six public gene-related databases, verified that our proposed method is effective, efficient, and useful.

Leveraging consumer behaviors for product recommendation: an approach based on heterogeneous network

Consumer behaviors (e.g., clicking products, adding products to favorites, adding products to carts, and purchasing products) play important roles in inferring consumers’ interests for product recommendation. Although studies have been conducted to incorporate the consumer behaviors for product recommendation, the heterogeneity of the behaviors and their composites were seldom explored for product recommendation. There is a need to capture the heterogeneity of the consumer behaviors and reveal their importance in the product recommendation because the behaviors indicate different consumer preferences for products. To bridge the gap, this research proposes a heterogeneous network-based approach to leverage the consumer behaviors for product recommendation. The proposed approach represents consumers and products as different types of nodes and behaviors as different types of edges. Meta paths that describe behavioral relationships between the consumers and products are used to calculate their similarities, which are further used to generate recommendations. To select informative meta paths for product recommendation, a heuristic selection mechanism is proposed. Besides, the research uses a non-negative matrix factorization method to learn the weights of the selected meta paths and then makes personalized recommendations for consumers. Experimental results based on real-world data demonstrate that the proposed approach not only helps to understand the roles of different consumer behaviors in product recommendation, but also achieves better recommendation performance than several baseline methods.

A novel community answer matching approach based on phrase fusion heterogeneous information network

Community Question Answering (CQA) allows users to ask or answer questions in a social way, so it is becoming the primary means for people acquiring knowledge. However, the asker must wait until a satisfactory answer appears, which reduces user activity. In this paper, we propose an innovative answering method that matches the most relevant answers for the new issue automatically. Firstly, we utilize phrases to represent the semantic of the posts (answers/questions) and construct a Phrase Fusion Heterogeneous Information Network, called PFHIN, to represent complex entity relationships in CQA. So, the answer selection is regarded as the related entity retrieval task. Then, we define the distance between entities in PFHIN, which is independent of the meta path. Finally, the Type-constrained Top-k Similarity Entity Finding Algorithm (TTSEF) is proposed for finding the nearest entities according to the known start entity and end-entity type, which can match the most relevant answers automatically.To the best of our knowledge, it is the first work to define the phrase information network for answer selection and provide a novel idea for the heterogeneous information network fusion. Experimental results on three large-scale datasets (Stack Overflow, Super User, and Mathematics) from Stack Exchange demonstrate that our proposed approaches significantly outperform the state-of-the-art answer retrieval methods. Moreover, we conduct an in-depth analysis of the meta path to the optimal answer and reveal the critical role of phrases in community answer matching.

Entity set expansion in knowledge graph: a heterogeneous information network perspective

Entity set expansion (ESE) aims to expand an entity seed set to obtain more entities which have common properties. ESE is important for many applications such as dictionary construction and query suggestion. Traditional ESE methods relied heavily on the text and Web information of entities. Recently, some ESE methods employed knowledge graphs (KGs) to extend entities. However, they failed to effectively and efficiently utilize the rich semantics contained in a KG and ignored the text information of entities in Wikipedia. In this paper, we model a KG as a heterogeneous information network (HIN) containing multiple types of objects and relations. Fine-grained multi-type meta paths are proposed to capture the hidden relation among seed entities in a KG and thus to retrieve candidate entities. Then we rank the entities according to the meta path based structural similarity. Furthermore, to utilize the text description of entities in Wikipedia, we propose an extended model CoMeSE++ which combines both structural information revealed by a KG and text information in Wikipedia for ESE. Extensive experiments on real-world datasets demonstrate that our model achieves better performance by combining structural and textual information of entities.

NEDD: a network embedding based method for predicting drug-disease associations

BackgroundDrug discovery is known for the large amount of money and time it consumes and the high risk it takes. Drug repositioning has, therefore, become a popular approach to save time and cost by finding novel indications for approved drugs. In order to distinguish these novel indications accurately in a great many of latent associations between drugs and diseases, it is necessary to exploit abundant heterogeneous information about drugs and diseases.ResultsIn this article, we propose a meta-path-based computational method called NEDD to predict novel associations between drugs and diseases using heterogeneous information. First, we construct a heterogeneous network as an undirected graph by integrating drug-drug similarity, disease-disease similarity, and known drug-disease associations. NEDD uses meta paths of different lengths to explicitly capture the indirect relationships, or high order proximity, within drugs and diseases, by which the low dimensional representation vectors of drugs and diseases are obtained. NEDD then uses a random forest classifier to predict novel associations between drugs and diseases.ConclusionsThe experiments on a gold standard dataset which contains 1933 validated drug–disease associations show that NEDD produces superior prediction results compared with the state-of-the-art approaches.

Knowledge-driven drug repurposing using a comprehensive drug knowledge graph.

Due to the huge costs associated with new drug discovery and development, drug repurposing has become an important complement to the traditional de novo approach. With the increasing number of public databases and the rapid development of analytical methodologies, computational approaches have gained great momentum in the field of drug repurposing. In this study, we introduce an approach to knowledge-driven drug repurposing based on a comprehensive drug knowledge graph. We design and develop a drug knowledge graph by systematically integrating multiple drug knowledge bases. We describe path- and embedding-based data representation methods of transforming information in the drug knowledge graph into valuable inputs to allow machine learning models to predict drug repurposing candidates. The evaluation demonstrates that the knowledge-driven approach can produce high predictive results for known diabetes mellitus treatments by only using treatment information on other diseases. In addition, this approach supports exploratory investigation through the review of meta paths that connect drugs with diseases. This knowledge-driven approach is an effective drug repurposing strategy supporting large-scale prediction and the investigation of case studies.

Anomaly detection in bitcoin information networks with multi-constrained meta path

Abstract As the most popular digital currency, Bitcoin has a high economic value, and its security has been paid more and more attention. Anomaly detection of Bitcoin has become a problem that must be solved. The existing Bitcoin anomaly detection methods only use static network models, and only the simple structural features such as node attributes and in/out-degree are considered to measure the similarities between nodes. Therefore, we propose a series of constrained anomaly detection algorithms for Bitcoin data. In our algorithms, we first construct a temporal Bitcoin network model for Bitcoin data. Then, combining time constraints, attribute constraints and structure constraints, a multi-constrained meta path is proposed on the basis of the meta path to specify the candidate sets, reference sets and similarity measurement strategies and detect local abnormal users and transactions that are of interest to users from static and dynamic angles with lower space-time overhead. Experiments on real-world Bitcoin data show that the constrained algorithms have certain improvements in recall, precision and F2 score when compared to the algorithms that only considers simple structural features such as node attributes and in/out-degree.

Detecting Outlier Patterns With Query-Based Artificially Generated Searching Conditions

In the age of social computing, finding interesting network patterns or motifs is significant and critical for various areas, such as decision intelligence, intrusion detection, medical diagnosis, social network analysis, fake news identification, and national security. However, subgraph matching remains a computationally challenging problem, let alone identifying special motifs among them. This is especially the case in large heterogeneous real-world networks. In this article, we propose an efficient solution for discovering and ranking human behavior patterns based on network motifs by exploring a user's query in an intelligent way. Our method takes advantage of the semantics provided by a user's query, which in turn provides the mathematical constraint that is crucial for faster detection. We propose an approach to generate query conditions based on the user's query. In particular, we use meta paths between the nodes to define target patterns as well as their similarities, leading to efficient motif discovery and ranking at the same time. The proposed method is examined in a real-world academic network using different similarity measures between the nodes. The experiment result demonstrates that our method can identify interesting motifs and is robust to the choice of similarity measures.

Efficient Ridesharing Framework for Ride-matching via Heterogeneous Network Embedding

Ridesharing has attracted increasing attention in recent years, and combines the flexibility and speed of private cars with the reduced cost of fixed-line systems to benefit alleviating traffic pressure. A major issue in ridesharing is the accurate assignment of passengers to drivers, and how to maximize the number of rides shared between people being assigned to different drivers has become an increasingly popular research topic. There are two major challenges facing ride-matching: scalability and sparsity. Here, we show that network embedding drives the optimal matches between drivers and riders. Contrary to existing approaches that merely depend on the proximity between passengers and drivers, we employ a heterogeneous network to learn the latent semantics from different choices in two types of ridesharing, and extract features in terms of user trajectories and sentiment. A novel framework for ridesharing, RShareForm, which encodes not only the objects but also a variety of semantic relationships between them, is proposed. This article extends the existing skip-gram model to incorporate meta-paths over a proposed heterogeneous network. It allows diverse features to be used to search for similar participants and then ranks them to improve the quality of ride-matching. Extensive experiments on a large-scale dataset from DiDi in Chengdu, China show that by leveraging heterogeneous network embedding with meta paths, RShareForm can significantly improve the accuracy of identifying the participants for ridesharing over existing methods, including both meta-path guided similarity search methods and variants of embedding methods.

Structural hole detection based on weighted meta path in heterogeneous networks

With the rapid development of online social networks, the detection of structural holes, i.e. identifying the key nodes that can bridge with individuals or groups without direct relationship in social networks, has attracted more attention of a large number of researches. The existing researches mainly focus on the influence of a homogeneous network structure, ignoring the importance of node types and different edges in online social networks. In this paper, an algorithm based on weighted meta paths for detecting structural hole in heterogeneous networks (SH_WMP) is proposed. SH_WMP not only flexibly integrates rich semantic information of heterogeneous networks, but also utilizes edge weight and potential link information to improve the performance. Experimental results show that the proposed method outperforms the comparison methods.

WMGCN: Weighted Meta-Graph Based Graph Convolutional Networks for Representation Learning in Heterogeneous Networks

Network embedding has been an effective tool to analyze heterogeneous networks (HNs) by representing nodes in a low-dimensional space. Although many recent methods have been proposed for representation learning of HNs, there is still much room for improvement. Random walks based methods are currently popular methods to learn network embedding; however, they are random and limited by the length of sampled walks, and have difficulty capturing network structural information. Some recent researches proposed using meta paths to express the sample relationship in HNs. Another popular graph learning model, the graph convolutional network (GCN) is known to be capable of better exploitation of network topology, but the current design of GCN is intended for homogenous networks. This paper proposes a novel combination of meta-graph and graph convolution, the meta-graph based graph convolutional networks (MGCN). To fully capture the complex long semantic information, MGCN utilizes different meta-graphs in HNs. As different meta-graphs express different semantic relationships, MGCN learns the weights of different meta-graphs to make up for the loss of semantics when applying GCN. In addition, we improve the current convolution design by adding node self-significance. To validate our model in learning feature representation, we present comprehensive experiments on four real-world datasets and two representation tasks: classification and link prediction. WMGCN's representations can improve accuracy scores by up to around 10% in comparison to other popular representation learning models. What's more, WMGCN'feature learning outperforms other popular baselines. The experimental results clearly show our model is superior over other state-of-the-art representation learning algorithms.

Effective and Efficient Discovery of Top-k Meta Paths in Heterogeneous Information Networks

<i>Heterogeneous information networks (HINs)</i> , which are typed graphs with labeled nodes and edges, have attracted tremendous interest from academia and industry. Given two HIN nodes <inline-formula><tex-math notation="LaTeX">$s$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$t$</tex-math></inline-formula> , and a natural number <inline-formula><tex-math notation="LaTeX">$k$</tex-math></inline-formula> , we study the discovery of the <inline-formula><tex-math notation="LaTeX">$k$</tex-math></inline-formula> most important meta paths in real time, which can be used to support friend search, product recommendation, anomaly detection, and graph clustering. In this work, we argue that the shortest path between <inline-formula><tex-math notation="LaTeX">$s$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$t$</tex-math></inline-formula> may not necessarily be the most important path. As such, we combine several ranking functions, which are based on <i>frequency</i> and <i>rarity</i> , to redefine the unified importance function of the meta paths between <inline-formula><tex-math notation="LaTeX">$s$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$t$</tex-math></inline-formula> . Although this importance function can capture more information, it is very time-consuming to find top- <inline-formula><tex-math notation="LaTeX">$k$</tex-math></inline-formula> meta paths using this importance function. Therefore, we integrate this importance function into a multi-step framework, which can efficiently filter some impossible meta paths between <inline-formula><tex-math notation="LaTeX">$s$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$t$</tex-math></inline-formula> . In addition, we combine bidirectional searching algorithm with this framework to further boost the efficiency performance. The experiment on different datasets shows that our proposed method outperforms state-of-the-art algorithms in terms of effectiveness with reasonable response time.

Weighted meta paths and networking embedding for patent technology trade recommendations among subjects

Most patent technology recommendations are based on link prediction of a homogeneous trade network and multiple-attribute matching. We constructed a heterogeneous information network (HIN) with four types of nodes and seven types of relations; designed a heterogeneous relation traversal algorithm based on the meta paths and meta structures inspired by the depth first search (DFS) strategy; obtained subject-relation sequences; and then calculated the weight of each meta path and meta structure through logistic regression. Using the relation sequence corpus of the weighted meta paths and meta structures among subjects, the patent technology trade recommendation model based on network embedding (PSR-vec) was proposed. The model was trained by using the Skip-gram method to obtain a vector-space representation for all subjects. Finally, the recommendation target was achieved by measuring the cosine similarity of the subject vectors. Through empirical research on the electronic information patent data, we observed that the PSR-vec model with weighted meta paths and meta structures was more precise than that with a single meta path or meta structure, which indicated that the patent technology trade was influenced by multiple factors. Second, the PSR-vec model combining weighted meta paths and meta structures was more precise than the unweighted model, which reflected more differences in multiple factors affecting trade. Third, compared to Deep Walk, Node2vec, Metapath2vec, and GraphSAGE methods, the PSR-vec model had a higher precision of up to 80%. Eventually, the recommendation subjects of PSR-vec included the holding relation, the supply relation, and the loose relation, which increased the diversity of the recommendation results. Our research thus provided a decision-making method for effective docking among patent technology trade subjects.

Churn modeling with probabilistic meta paths-based representation learning

Finding structural and efficient ways of leveraging available data is not an easy task, especially when dealing with network data, as is the case in telco churn prediction. Several previous works have made advancements in this direction both from the perspective of churn prediction, by proposing augmented call graph architectures, and from the perspective of graph featurization, by proposing different graph representation learning methods, frequently exploiting random walks. However, both graph augmentation as well as representation learning-based featurization face drawbacks. In this work, we first shift the focus from a homogeneous to a heterogeneous perspective, by defining different probabilistic meta paths on augmented call graphs. Secondly, we focus on solutions for the usually significant number of random walks that graph representation learning methods require. To this end, we propose a sampling method for random walks based on a combination of most suitable random walk generation strategies, which we determine with the help of corresponding Markov models. In our experimental evaluation, we demonstrate the benefits of probabilistic meta path-based walk generation in terms of predictive power. In addition, this paper provides promising insights regarding the interplay of the type of meta path and the predictive outcome, as well as the potential of sampling random walks based on the meta path structure in order to alleviate the computational requirements of representation learning by reducing typically sizable required data input.

Efficient heterogeneous proximity preserving network embedding model

We study the problem of representation learning in heterogeneous information networks. Its unique challenges come from the existence of multiple types of vertices and edges. Existing proximity-based network embedding techniques ignore the type information when evaluating the proximity and limits their usage in heterogeneous scenario. In this paper, we propose a heterogeneous proximity preserving network embedding model via meta path guided random walk, which is capable of capturing the high-order proximity between vertices specified by the given path. To improve the learning efficiency, we introduce a sampling based learning strategy which can incrementally learn representations. We conduct experiments on two real world heterogeneous information networks. Experimental results on several mining tasks prove the effectiveness of our approach over many competitive baselines. The model is very efficient and is able to learn embeddings for large networks both in offline and online scenarios. Besides, for expert system, our approach can be applied to improve the representation of knowledge entities by depicting the knowledge base as a heterogeneous information network.