This the the repository for the CAiSE-2020 accepted long paper “An Approach for Process Model Extraction By Multi-Grained Text Classification“. In this paper, we formalize the PME task into the multi-grained text classification problem, and propose a neural network and the coarse-to-fine (grained) learning mechanism to effectively extract multi-grained procedural knowledge.

• code/
  This directory contains the source code of our approach.
• data/
  This directory contains two datasets used for evaluation.

The dataset named as X-Y denotes a text classification task Y on data source X. For two sentence-level tasks — SC (sentence classification) and SSR (sentence semantics recognition), an example <x,y> in each line denotes a sentence x and its label y. each of them can be solely used to evaluate single-sentence classification tasks. For the word-level task — SRL (semantic role labeling), an example <x,i,y> in each line denotes a sentence x, a subordinate word index i and a corresponding label y. It can be further used to evaluate sequential-text classification tasks. We will keep updating them to provide more reliable version(s), including correcting wrongly-annotated labels and adding more training/testing examples. The up-to-date version can be directly downloaded from this repository. In summary:

• COR for cooking recipes and MAM for maintenance manuals.
• COR-SC.txt is the dataset for sentence-level classification (ST1) to identify whether a sentence is describing an action or a statement. The format is <Sentence, Label>.
• COR-SSR.txt is the dataset for sentence-level semantics recognition (ST2) to recognize the semantics of a Statement sentence to control the execution of following actions. The format is <Sentence, Label, Y/N> in which Y denotes the sentence belongs to the label while N not (so, you could ignore all examples with N notations, just focus on the examples with Y notations).
• COR-SRL.txt is the dataset for word-level semantic role labeling to assign semantic roles to words in an Action sentence. The format is <Word, Word Postion in the Sentence, Label>.

• Python (3.6.8 or 3.7.3)
• PyTorch (1.0.1)
• Word2Vec. Note that the embedding layer can be replaced by other models such as BERT. If necessary, kindly replace corresponding lines of the code to do so.

If you find this project helps, please kindly consider citing this paper as:

@inproceedings{MGTC,
  title = {An Approach for Process Model Extraction By Multi-Grained Text Classification},
  author = {Chen Qian and Lijie Wen and Akhil Kumar and Leilei Lin and Li Lin and Zan Zong and Shuang Li and Jianmin Wang},
  booktitle = {Proceedings of The 32nd International Conference on Advanced Information Systems Engineering (CAiSE)},
  year = {2020},
  pages = {268-282}
}