Postdoctoral Fellow on Machine Learning and Text Mining with supervisor Prof. Qiang Yang.

In the past year, I worked in a startup company to do text analysis and mining for the large scale resume data. In this company, I worked on how to embed the large scale resume data into vector space, and extract resume topics using topic models for similar resume analysis and retrieval, since the resumes can be treated as one type of SSDs. We also did some research works on resume embedding by word2vector model. Moreover, I got involved in company extraction by named entity recognition technology and resume text extraction. Our works play an important role in resume data analysis and mining.

I am working on the research about Semi-Structured Documents (SSDs) modeling with topic models using the ideas of Tag-Weighted. These SSDs contain both unstructured features (e.g., plain text) and metadata (e.g., tags). This type of data is common in web applications, such as e-commerce, blogs, news, and so on. We proposed a tag-weighted topic model framework to handle the traditional SSDs, and this work was accepted by IJCAI2013. For hybrid SSDs, we presented a extended model called tag-weighted Dirichlet allocation, which was accepted by ICDM2013. Recently, I am working on how to model the correlations among the tags. Meanwhile, I am in charge of other research works about SSDs modeling using topic models.

I worked in Microsoft research Asia to do the research on Semi-Structured Documents modeling with deep learning methods, such as deep belief networks and deep Boltzmann machine. In this period, my colleagues and I proposed a deep neural network to handle SSDs, which called Compensatory Deep Belief Network model. This part of the work is still continuing. In the future, we will develop the model with more flexible deep structures.

I developed a platform for “Internet Public Opinion Mining”, which is designed for large scale business data analysis. This system is built on Hadoop, which consists of three parts: a) Business data crawler. We designed and developed a distributed crawler to collect business data from different information sources. b) Business data processing and analysis. I built the data processing and analysis platform to preprocessing, indexing, information extraction and analysis for the business data. c) A management system for information view. This system is online used by a company.

I worked in Institute of China Telecom on large scale images retrieval based on MapReduce. We developed a search engine platform for more than one hundred million images using Hadoop. I was in charge of clustering large scale images to speed up image searching.

I completed the development of 3D Radar image interpolation algorithm, and then developed Radar Image fast template matching algorithm in Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen. This part of works is for weather prediction using real space Radar image data, when I was an intern in Chinese Academy of Sciences.