Graph is an important data representation which appears in a wide diversity of real-world scenarios. Effective graph analytics provides users deeper understanding what behind the data, and thus can benefit lot useful applications such as node classification, recommendation, link prediction, etc. However, most methods suffer high computation space cost. embedding effective yet efficient way to solve problem. It converts into low dimensional structural information properties are maximumly...

With the recent surge of location based social networks (LBSNs), activity data millions users has become attainable. This contains not only spatial and temporal stamps user activity, but also its semantic information. LBSNs can help to understand mobile users' preference (STAP), which enable a wide range ubiquitous applications, such as personalized context-aware recommendation group-oriented advertisement. However, modeling user-specific STAP needs tackle high-dimensional data, i.e.,...

With the increasing popularity of location-based services, such as tour guide and social network, we now have accumulated many location data on Web. In this paper, show that, by using based GPS users' comments at various locations, can discover interesting locations possible activities that be performed there for recommendations. Our research is highlighted in following location-related queries our daily life: 1) if want to do something sightseeing or food-hunting a large city Beijing, where...

Most machine-learning methods focus on classifying instances whose classes have already been seen in training. In practice, many applications require not previously. Zero-shot learning is a powerful and promising paradigm, which the covered by training we aim to classify are disjoint. this paper, provide comprehensive survey of zero-shot learning. First all, an overview According data utilized model optimization, into three settings. Second, describe different semantic spaces adopted...

In this paper, we study an important yet largely under-explored setting of graph embedding, i.e., embedding communities instead each individual nodes. We find that community is not only useful for community-level applications such as visualization, but also beneficial to both detection and node classification. To learn our insight hinges upon a closed loop among embedding. On the one hand, can help improve detection, which outputs good fitting better other be used optimize by introducing...

With the increasing popularity of location tracking services such as GPS, more and mobile data are being accumulated. Based on data, a potentially useful service is to make timely targeted recommendations for users places where they might be interested go activities that likely conduct. For example, user arriving in Beijing wonder visit what she can do around Forbidden City. A key challenge recommendation problems we have each individual very limited, while accurate recommendations, need...

Motivated by the vast applications of knowledge graph and increasing demand in education domain, we propose a system, called KnowEdu, to automatically construct for education. By leveraging on heterogeneous data (e.g., pedagogical learning assessment data) from this system first extracts concepts subjects or courses then identifies educational relations between concepts. More specifically, it adopts neural sequence labeling algorithm extract instructional employs probabilistic association...

In this paper, we study the problem of using representation learning to assist information diffusion prediction on graphs. particular, aim at estimating probability an inactive node be activated next in a cascade. Despite success recent deep methods for diffusion, find that they often underexplore cascade structure. We consider as not merely sequence nodes ordered by their activation time stamps; instead, it has richer structure indicating process over data graph. As result, introduce new...

Knowledge tracing serves as the key technique in computer supported education environment (e.g., intelligent tutoring systems) to model student's knowledge states. While Bayesian and deep models have been developed, sparseness of exercise data still limits tracing's performance applications. In order address this issue, we advocate for propose incorporate structure information, especially prerequisite relations between pedagogical concepts, into model. Specifically, by considering how...

Human activity recognition plays an important role in people's daily life. However, it is often expensive and time-consuming to acquire sufficient labeled data. To solve this problem, transfer learning leverages the samples from source domain annotate target which has few or none labels. Unfortunately, when there are several domains available, difficult select right for transfer. The means that most similar properties with domain, thus their similarity higher, can facilitate learning....

Given ubiquitous graph data such as the Web and social networks, proximity search on graphs has been an active research topic. The task boils down to measuring between two nodes a graph. Although most earlier studies deal with homogeneous or bipartite only, many real-world are heterogeneous objects of various types, giving rise different semantic classes proximity. For instance, network users can be close for reasons, being classmates family members, which represent distinct Thus, it becomes...

In this paper, we introduce a new setting for graph embedding, which considers embedding communities instead of individual nodes. We find that community is not only useful community-level applications such as visualization but also provide an exciting opportunity to improve detection and node classification. Specifically, consider the interaction between closed loop, through embedding. On one hand, can since detected are used fit other be optimize by introducing community-aware high-order...

Knowledge graph embedding plays an important role in knowledge representation, reasoning, and data mining applications. However, for multiple cross-domain graphs, state-of-the-art models cannot make full use of the from different domains while preserving privacy exchanged data. In addition, centralized model may not scale to extensive real-world graphs. Therefore, we propose a novel decentralized scalable learning framework, \emph{Federated Graphs Embedding} (FKGE), where embeddings graphs...

In activity recognition, one major challenge is huge manual effort in labeling when a new domain of activities to be tested. this paper, we ask an interesting question: can transfer the available labeled data from set existing help recognize another different but related domain? Our answer "yes", provided that sensor two domains are some way. We develop bridge between by learning similarity function via Web search, under condition same feature space. Based on learned measures, our algorithm...

This department presents the results of first Data Mining Contest held at 2007 International Conference on Mining.

Chain businesses have been dominating the market in many parts of world. It is important to identify optimal locations for a new chain store. Recently, numerous studies done on store location recommendation. These typically learn model based features existing stores city and then predict what other sites are suitable running one. However, these models do not work when enterprise wants open business where there enough data about this To solve cold-start problem, we propose CityTransfer, which...

With the growing popularity of location-based social networks, numerous location visiting records (e.g., check-ins) continue to accumulate over time. The more these are collected, better we can understand users’ mobility patterns and accurately predict their future locations. However, due personality trait neophilia, people also show propensities novelty seeking in human mobility, that is, exploring unvisited but tailored locations for them visit. As such, existing prediction algorithms,...

In fashion recommender systems, each product usually consists of multiple semantic attributes (e.g., sleeves, collar, etc). When making cloth decisions, people show preferences for different the clothes with v-neck collar). Nevertheless, most previous recommendation models comprehend clothing images a global content representation and lack detailed understanding users' preferences, which leads to inferior performance. To bridge this gap, we propose novel Semantic Attribute Explainable...

Many real-world networks have a rich collection of objects. The semantics these objects allows us to capture different classes proximities, thus enabling an important task semantic proximity search. As the core search, we measure on heterogeneous graph, whose nodes are various types Most existing methods rely engineering features about graph structure between two their proximity. With recent development embedding, see good chance avoid feature for There is very little work using embedding We...

Due to privacy and security constraints, directly sharing user data between parties is undesired. Such decentralized silo issues commonly exist in recommender systems. In general, systems are data-driven. The more it uses, the better performance obtains. a severe limitation of recommender's performance. Federated learning an emerging technology, which bridges silos builds machine models without compromising security. We design system based on federated learning. It known as system....