Event Source Page Discovery via Policy-based RL with Multi-Task Neural Sequence Model

摘要

The problem of finding event announcement pages for any given website is called event source page discovery. In this paper, we show a policy-based deep reinforcement learning (RL) model for the event source page discovery agent. We use two stages to train our agent, pre-training and fine-tuning. In the pre-training phase, the model is trained with limited labeled data, where each episode has a fixed number of steps. In the fine-tuning phase, the agent is trained using unlabeled data and a reward system based on an event source page classifier. The agent learns whether to continue exploring or stop exploring through an adaptive threshold. The proposed agent achieves 74% precision with a 1.28 unit cost (the average number of clicks for each event source page) on the real word data set.