ABLE: Meta-Path Prediction in Heterogeneous Information Networks

Given a heterogeneous information network (HIN) H, a head node h , a meta-path P, and a tail node t , the meta-path prediction aims at predicting whether h can be linked to t by an instance of P. Most existing solutions either require predefined meta-paths, which limits their scalability to schema-rich HINs and long meta-paths, or do not aim at predicting the existence of an instance of P. To address these issues, in this article, we propose a novel prediction model, called ABLE, by exploiting the A ttention mechanism and B i L STM for E mbedding. Particularly, we present a concatenation node embedding method by considering the node types and a dynamic meta-path embedding method that carefully considers the importance and positions of edge types in the meta-paths by the Attention mechanism and BiLSTM model, respectively. A triplet embedding is then derived to complete the prediction. We conduct extensive experiments on four real datasets. The empirical results show that ABLE outperforms the state-of-the-art methods by up to 20% and 22% of improvement of AUC and AP scores, respectively.

HinPhish: An Effective Phishing Detection Approach Based on Heterogeneous Information Networks

Internet users have suffered from phishing attacks for a long time. Attackers deceive users through malicious constructed phishing websites to steal sensitive information, such as bank account numbers, website usernames, and passwords. In recent years, many phishing detection solutions have been proposed, which mainly leverage whitelists or blacklists, website content, or side channel-based techniques. However, with the continuous improvement of phishing technology, current methods have difficulty in achieving effective detection. Hence, in this paper, we propose an effective phishing website detection approach, which we call HinPhish. HinPhish extracts various link relationships from webpages and uses domains and resource objects to construct a heterogeneous information network. HinPhish applies a modified algorithm to leverage the characteristics of different link types in order to calculate the phish-score of the target domain on the webpage. Moreover, HinPhish not only improves the accuracy of detection, but also can increase the phishing cost for attackers. Extensive experimental results demonstrate that HinPhish can achieve an accuracy of 0.9856 and F1-score of 0.9858 .

Incorporating heterogeneous information in deep learning with informative meta-paths for community recommendations

Communities of interest promote knowledge sharing and discovery in social network platforms. However, platform users face difficulties of finding suitable communities, given their increasing number. Although recommendations have been proposed to help users find communities of interest, these methods ignore or exclude heterogeneous interactions between users and communities. In addition, widely used meta-paths help capture the complex semantic relation among entities but heavily rely on domain knowledge. In this study, we propose a novel recommendation model based on informative meta-path discovery in heterogeneous information networks and deep learning. Users, communities, relevant items and their relations are considered as entities in a heterogeneous information network, from where informative meta-paths are extracted on the basis of information theory to measure user-community similarities. Finally, similarities are incorporated in a deep learning model to predict whether target users join candidate communities. The proposed recommendation model is evaluated and compared against baseline methods using two data sets. Results demonstrate the superior performance of the present model in terms of precision, recall and F score.

Commodities trend link prediction on heterogeneous information networks

Events can be defined as an action or a series of actions that have a determined theme, time, and place. Event analysis tasks for knowledge extraction from news and social media have been explored in recent years. However, there are still few studies that aim to enrich predictive models using event data. In particular, agribusiness events have multiple components to be considered for a successful prediction model. For example, price trend predictions for commodities can be performed through time series analysis of prices, but we can also consider events that represent knowledge about external factors during the training step of predictive models. In this paper, we present a method for integrating events into trend prediction tasks. First, we propose to model events and time-series information through heterogeneous information networks (HIN) that allow multiple components to be directly modeled through multi-type nodes and edges. Second, we learn features from HIN through network embedding methods, i.e., network nodes are mapped to a dense vector of features. In particular, we propose a network embedding method that propagates the semantic of the pre-trained neural language models to a heterogeneous information network and evaluates its performance in a trend link prediction. We show that the use of our proposed model language-based embedding propagation is competitive with state-of-art network embeddings algorithms. Moreover, our proposal performs network embedding incrementally, thereby allowing new events to be inserted in the same semantic space without rebuilding the entire network embedding.

HINRL4Rec: A Novel Integrated Heterogeneous Network Embedding with Reinforcement Learning for Recommendation

Recent KG-oriented recommendation techniques mainly focus on the direct interaction between entities in the given KGs as the rich information sources for leveraging the quality of recommendation outputs. However, they are still hindered by the heterogeneity, type-varied entities and their relationships in knowledge graphs (KG) as the heterogeneous information networks (HIN). This limitation seems challenging to build up an effective approach for the KG-based recommendation system in both semantic path-based exploitation and heterogeneous information extraction. To meet these challenges, we proposed a novel integrated HIN embedding with reinforcement learning (RL)-based feature engineering for recommendation, called as: HINRL4Rec. First of all, we apply the combined textual meta-path-based embedding approach for learning multiple-rich-schematic representations of user/item and their associated entities. Then, these extracted multi-typed embeddings of user and item entities are fused into the unified embedding spaces during the KG embedding process. Finally, the unified representations of users and items are then used to facilitate the RL-based policy-driven searching process in the next steps for performing the recommendation task. Extensive experiments in real-world datasets demonstrate the effectiveness of our proposed model in comparing with recent state-of-the-art recommendation baselines.