auniquesun / project

implementations of the framework described in paper Presenting Problems and Methods under Weak Supervision

Introduction

This repository published the source code of paper Presenting Problems and Methods under Weak Supervision. In this paper, we put forward a framework that can present the problem and the method of a given paper automatically by means of sequence labeling. The framework consists of four steps and three algorithms. Extensive experiments on three public datasets which are AMiner, Microsoft Academic Graph (MAG) and Paper with Code (PWC) show the effectiveness of our framework. Details are provided in the code.

Framework

We separate our framework into four parts, let we put them together. The figure below presents the flowchart of the framework we put forward. At first, we develop Algorithm 1 to find new pattern words in step 1 from massive text corpus in the settings of weak supervision. Step 2 involves in pattern-based phrase extraction implemented by Algorithm 2. Then we generate data with labels in step 3 according to the overall pattern words and phrases returned by previous two steps, and this step corresponds to Algorithm 3. At last, it turns to step 4, various classical sequence labeling models are trained on labeled data to summarize or present the problem and the method with a prediction sequence of tags.

Contact

Anyone who is confronted with problems can open issues under this project. I am looking forward to hearing your voices and will reply you in time.