104 Job match recommendation using deep learning

摘要：

In this talk, I will share the experience of 104 hackathon contest which includes data exploration, data cleaning, model selection, fine-tuning by hyper-parameter and I will also go through each ML model used in the contest.

In this contest, our team (Tobacco AI) use deep learning to predict whether user will apply the certain job or not. In this contest, 104 organizer provides us two kinds of data, user behavior data and job information database. For user behavior data, we extract it to generate the action sequence and apply convolutional neural network (CNN) and long short-term memory neural network (LSTM) to find user behavior pattern. For job information database, we use variational auto-encoder (VAE) to find job-hidden vector of every job and form the job-hidden space. Due to user de-identification, we can use the job-hidden vector to calculate job-averaging feature to be one of user features and extract user behavior data to get user-job feature. Then, we concatenate these different hidden output vectors from CNN/LSTM/VAE to train the model. As a result, we are the second place on leaderboard and get the merit award and AWS award.

講者簡介：

Chris Lin (Tsung-Fu Lin) is Sr. software engineer of TrendMicro. He is enthusiast in big data and cloud computing technologies, such as Spark and Hadoop. Recently, he is responsible for graph mining and deep learning on threat research, such as URL classification, email writing style identification, etc.

• ← Making Elephants Fly: Hadoop on Cloud
• 使用 SMACK 開發小型 Hybrid Database 系統時踩過的坑 →