Posts by Collection



Discussions about The Application and Analysis of Computer Network on E-Commerce

Published in Digital Communication World, 2015

At present, e-commerce has become the main shopping method for young people, but e-commerce in its development shows various problems, especially network security. Therefore, computer network technology is particularly important. Based on the application status of computer network technology in e-commerce, this paper analyzes the computer network technology commonly used in e-commerce. Starting from the actual situation, we study the problems of the use of e-commerce in computer network technology, and put forward feasible strategies to contribute to the application of computer network technology.

Fast Inference of Diffusion Networks without Infection Temporal Information

Published in NDBC, 2018

The communication and influence among people form complex diffusion networks which consider person as nodes. Diffusion network inference can help us understand these diffusion networks and the related diffusion processes more intuitively and accurately, so as to better predict, promote or prevent future diffusion events. Most of the existing network inference algorithms require the infection temporal information of nodes. But in many real-world diffusion processes, the exact infection temporal information is often unavailable, resulting in a limited availability for existing methods. This paper aims at an effective, efficient and infection temporal information-free approach for diffusion network inference, and studies how to infer the influence relationships and transmission rates between the nodes based on only the final infection statuses of the nodes observed in a number of diffusion processes. To this end, the approach proposed in this paper first utilizes the mutual information of infection statuses among the nodes to quantify the correlation of the infections and reveal the underlying influence relationships between the nodes. Then, the proposed approach constructs the log-likelihood function of the observed infection statuses with transmission rates as the variables, and adopts the Expectation-Maximization algorithm to maximize the function and approximate the optimal transmission rates. Extensive experimental results demonstrate that compared against the existing approaches, the proposed approach has a reasonably better accuracy performance on diffusion network inference, and significantly reduces runtime.

Inferring Diffusion Networks without Timestamps (UNDER REVIEW)

Published in ICDE (UNDER REVIEW), 2018

To infer the structures of diffusion networks, most existing approaches require not only the final infection statuses of network nodes, but also the timestamps that pinpoint the exact time when infections of the nodes occur. Nonetheless, in many natural diffusion processes such as the spread of epidemics, monitoring the exact infection timestamps is often difficult and expensive. In this work, we investigate the problem of inferring the diffusion network structures without infection timestamps, using only the final infection statuses of nodes. Instead of utilizing the sequence of timestamps to determine the potential parent-child influence relationships between the nodes, we design a probabilistic generative model of the final infection statuses to quantitatively measure the likelihood of potential structures of the objective diffusion network, taking into account network complexity. Then, for each node in the network, we can infer an appropriate number of most probable parent nodes. Furthermore, in order to reduce redundant computation during the inference, we preclude a few insignificant candidate parent nodes from being considered in the inference, if their infections have very weak correlations with the infections of the corresponding child nodes. Extensive experiments on both synthetic and real-world networks are conducted, and the results verify the effectiveness and efficiency of our approach.

From Appearances to Essence: Learning Diffusion Networks from Final Infection Results (UNDER REVIEW)

Published in IJCAI (UNDER REVIEW), 2019

To infer the structures of diffusion networks, most of the existing approaches require not only the final infection statuses of the nodes in the network, but also the temporal information that pinpoints the exact time when infections occur. Nonetheless, in many natural diffusion processes such as the spread of epidemics, monitoring the exact infection time is usually difficult and expensive. In this paper, we investigate the problem of inferring the diffusion network structures based on only the final infection statuses of nodes. Instead of utilizing the sequence of infection time to determine the potential parent-child influence relationships between the nodes, we present a conditional entropy-based method to infer the candidate parent nodes for each node in the network. Furthermore, to refine the inferred diffusion network structure, we disqualify the insignificant candidate parent nodes whose infections have very weak correlations with the infections of the corresponding child nodes. Experimental results on both synthetic and real-world networks verify the effectiveness and efficiency of our approach.


Location Based Searching Algorithm


Location-based services (LBS) emerge with the rapid rise of big data and the Internet, which provides a new service and business model for Internet enterprises, and also brings users a more pleasant and convenient experience of using the Internet. Internet service platforms for taxis and other vehicles provide positioning-based “taxi” services for users. On the surface, the user only needs to open the software and click on the destination to “call” the nearest idle taxi according to the current location; in fact, after receiving the client’s request and the client’s current location report, the server can find the nearest taxi according to a specific algorithm and return its information to the client (usually mobile devices such as mobile phones).

Research on Arbitrarily Shaped Clustering


Arbitrary shaped clustering is an important research topic in the domain of data mining. Most of the existing approaches cannot ensure good clustering results and scalability on large high-dimensional datasets at the same time. To meet the clustering requirement on large high-dimensional datasets, some researchers proposed that we could extract the backbone of each cluster before conducting clustering on the backbone. In this way, we can reduce time to some extent, however the time reduced depends on the proportion of representatives forming the backstone to the whole dataset, which varies with the dataset. In this paper, we put forward an efficient clustering method based on representatives sampling and boundary similarity. It includes three steps: firstly, conduct representatives sampling, which makes represemtatives distribute evenly and continuously; secondly, adjust position of representatives iteratively to make them get closer to corresponding k nearest neighbors; finally, conduct agglomerative clustering based on boundary similarity. We conduct extensive experiments on synthetic and real-world datasets, and contrast experiments with other methods. The experimental results prove the validity and efficiency of our method.

Research on Diffusion Network Inference without Temporal Information


In previous works, the approaches mainly focused on network inference with temporal data. However, we have to doubt that whether the temporal information is precise in real world. Many networks do not have any records of precise time of infection like social networks do. For example, in a network of disease spreads, it’s quite time consuming to retract the exact infection time and even if we have the temporal info, it is probably incorrect and misleading for the reason that people react differently to the same disease, and the time they seek for medical help can be greatly influenced by subjective factors. The same condition also happens in other kinds of diffusion networks. The second defect is that they are quite time consuming if the algorithms are performed on lap-tops, usually several days. If they are applied, a high-performance computer or server is necessary.

Research on Recommender Systems for Further Education


Recently, many researches have been done on multiple types of recommender systems, including job recommending, movie recommending, commodity recommending etc. However, as a kind of recommender system needed by many students, further education-oriented recommender systems are rarely mentioned. In this paper, we propose a preliminary recommender system based on profile comparison applied in further education, which can recommend professors to students. In order to compare the profiles of student and professors, we proposed a distance function using word2vec and doc2vec. Moreover, regarding the amount of history data collected during operation, we separate the model into two stages. In the two stages, different methods of giving recommendations are applied.


Crowdsourcing Service Platform

Project, Wuhan University, 2017

This is a national-level college student innovative project and I am the leader in it. This project plans to develop and operate a trading platform for on-campus services (crowdsourcing) and second-hand goods. Currently in our university, we do not have enough platform to share and acquire sufficient service provided by others. We aim at designing a platform to balance these resources. After much investigation through offline and online survey of what functions we should provide, we separate designed the application as follows. There are two main parts. In part one, users can post their request or need on our platform. The other part is specifically designed for second-hand trading. We have already implemented the application and trying to make it into real use.

Patent Information Analyzing Platform

Project, Wuhan University, 2017

At present, China vigorously encourages the public to start businesses and innovate the future. As an important achievement of innovation, patent technology in intellectual property rights has been paid more and more attention by enterprises and individuals. The “Creative Spring” patent map is a more intuitive map by collecting and integrating patented technology data. It provides prediction and evaluation services for development teams, avoids the risk of waste of resources caused by repeated research and development, and provides relevant legal suggestions, such as the guidance of the state in this field. At present, the development of patent map in China is lagging behind, basically blank. By providing forecasting and analysis services for enterprises and government departments, “Creative Spring” patent map can collect service fees from it, and constantly improve data sources, expand profits, and achieve the purpose of entrepreneurship.

Embedded Smart Monitoring System

Project, Wuhan University, 2017

This project aimed at designing a smart monitoring system, providing basic functions and some advanced functions. Detailed functions are as follows.

  • Video recording
  • Picture recording
  • Online live
  • Family facial recognition
  • Unlisted face alarming
  • Gesture control

For the facial recognition and gesture recognition, local recognition programs and FACE++ APIs were applied to provide more accurate recognition result due to the performance limit of our embedded system – Raspberry Pi.

National College Student Information Security Contest

Contest, Wuhan University, 2017

In recent years, with the vigorous development of the Internet, social networks have become a necessity for every one of us. We often use instant messaging tools such as WeChat to make video calls and use platforms such as Weibo and Friendship Circle to record our lives. But as we enjoy the fun and convenience of social networks and instant messaging devices, our personal information is slowly leaking through these modern communication tools. Because of the openness of social networks and video communication platforms, when we use videos or photos containing personal portrait information on the network, we cannot ensure that this facial information is not abused, nor can we rely entirely on the Internet platform to protect our information.

At present, many face-based authentication systems still use 2D devices to identify users, which means that our photos and videos and other 2D portrait information can be easily used in these authentication systems. Taking Android mobile phone as an example, there are few devices using 3D depth camera to obtain relatively safe three-dimensional facial data. Most of them still use ordinary cameras for facial recognition, which facilitates the abuse of portrait information.

In order to solve this problem, we are committed to propose a solution to protect the face information flooded in social networks and video communications. The face processing of real-time video calls or photos greatly reduces the possibility of the image being verified by the identity authentication system directly and does not affect the view of the photos or videos themselves. In view of the rapid development of mobile devices and applications, our solution will use Android applications as a platform to provide a facial information protection tool for the vast number of social media users.

In order to achieve this product, we choose to add noise to the video or photo to make the direct 2D face recognition fail, which cannot be verified by the face recognition system, and reduce the impact of noise to the unaware of the naked eye (see figure 2.6.2). The basis of face recognition is neural network training, as long as the face part of the photo cannot be recognized by the general neural network. On the other hand, for the noisy images needed in image processing, we train them by deep learning. By building the training model of noise image neural network and using the existing universal face data set as data input, the available noise image is generated according to the corresponding algorithm.

MCM/ICM 2018

Contest, Wuhan University, 2018

In MCM/ICM 2018, we were awarded as Meritorious Winner. In this contest, we chose the problem D, organizing a charging network for electric vehicles. During our preperation, we proposed a top-down model for network design optimization. The following is the summary of our paper.

Optimizing network of charging stations: A top-down approach

Keywords: Electric Vehicle; Station Planning; Top-down Approach; Network

We construct a top-down model on planning the charging station for electric vehicles with the target of developing an all-electrical vehicle traffic system gradually.

The problem contains three main points: the position, optimal amount, and the sequence of building them. We separate it into three levels: Country-level: determine the maximum number of charging stations; Network-level: determine the sequence of edge (road) selection; City-level: determine the accurate position of each charging station within city scale. On country-level, we develop Charging Stations Amount Estimation Sub-model (CSAE), which is under ideal assumptions to estimate the number of destination charging stations needed in each city and average density of supercharging stations needed on each road if everyone switched to all-electric. As for network-level, depending partly on the Connectivity-Freedom Degree (CFD) to evaluate the benefit to graph of an edge, Top-level Network Sequence Selection, Sub-model (TNSS) can effectively and efficiently find out the sequence of roads along which we build charging stations. Finally, on City-level, City-scale Charging Station Planning Sub-model (CCSP) is designed to plan accurate positions for every charging stations, given the maximum number of charging stations within a city. CCSP takes population density, traffic flow within a district and the capacity of a charging station into account. In this paper, models are introduced in a top-down order: CSAE, TNSS and CCSP.

To evaluate the efficacy of every model, we perform experiments on both synthetic and real data. Results show that our model both meets the requirement of this problem and accurately generates an optimal solution. Additionally, the model we proposed is compatible with further extension and has a high execution efficiency.