Community-Based Service Ecosystem Evolution Analysis

Data Description

In data dictionary, we release our data used in this paper. The release dataset collection and processing process can refer:

@misc{liu2020datadriven,
    title={A Data-driven Approach for Constructing Multilayer Network-based Service Ecosystem Models},
    author={Mingyi Liu and Zhiying Tu and Xiaofei Xu and Zhongjie Wang},
    year={2020},
    eprint={2004.10383},
    archivePrefix={arXiv},
    primaryClass={cs.CY}
}

Related code be found in:

• MSEM-EventExtraction
• MSEM-EvolutionaryRelationGenerate

data/nodes.json: Each line represents a node in the json format. Each node contains following fields:

field	description	example	note
id	node unique id	ofo	-
label	node label	ofo	for service and stakeholder label is same as id
type	node type	Stakeholder	In this paper, we only use Stakeholder and Service

data/edges.json: Each line represents an edge in the json format. Each edge contains following fields:

field	description	example	note
id	edge unique id	....	-
source	source node id	9	-
target	target node id	HappyCycle	-
timestamp	when the edge was created	2019-08-12	Structure relation will be set to 1999-01-01
type	the edge type	evolutionary	HasX, structural or evolutionary
r	the edge semantics type	BelongTo	this attribute will be used to aging function
generated_from	where the edge generated from	9	this attribute only for evolutionary type edge

Service ecosystem snapshots generation, static community detection, nodes' social position calculation are time consuming. So we provide a copy of our intermediate results, which have been stored in data/*.pkl. You can use this data to save your time.

Source code

• generate_snapshots.py: this script mainly contains aging function and Service Ecosystem snapshots steps
• community_features.py: this script contains some common used community features.
• community_options.py: this script contains community detection, meta community identifier, community evolutionary event detection, community features vector construction, etc.
• model_operations.py: this script contains how to train the service community evolutionary event prediction model. And apply the SHAP explainer on the model.
• report.py: this script mainly contains how to visualize the analysis results.
• test.py: this script gives an example of whole process. (Notebook version will come soon)

Report

Generated figures are stored in figure folder with .html format.

• summary_{evolution_evnt}.html: the summary report of evolution_evnt shows the importance of different features (merged) to the event.
• dependency_{feature}_{evolution_evnt}.html: the dependency report gives much detailed information.

Please refer to the paper to get more details.

Appendix

Evaluate on different classifiers

In our paper, we chose random forest. You can chose other prediction mode. All you need to do is moddify the function train_prediction_model in model_operations.py.

if model_name == "random_forest":
    model = RandomForestClassifier(n_estimators=50)
elif model_name == "xxxx": #modify here, and change params value
    model = XxxxClassifier(mmmm=xxx)