1. Build Smart Choice System, a recommendation system to suggest goods to sell to WeChat OfficialAccounts based on their articles.2. Rank a product based on the keywords matched between this product and the articles. Use the topic model to avoid the ambiguity of keywords.3. Now it serves as the backend algorithm for Weizoom company.

1. Work on project User Profiling on E-commerce Websites from Online Review.2. Use weakly-supervised learning to predict demographic attribute based on demographic mentions.3. For example, if someone said: “I bought my wife iPhone 6s, and it worked well". Then the consumer is probable to be a married male.

1. Develop Semantic Cloud Platform, a system to detect the vicissitude of Chinese semantic in last ten years.2. Learn word embeddings in each year and take the nearest neighbors as the possible semantics. Cluster potential semantics with AP clustering. Regard the clustering results as the multiple semantics of each word.3. Result reveals influences from subculture (new language), big company (like “Apple”), and so on.