guoshengkang / NAFM

This repository is the Python implementation with tensorflow 1.4.0 for the ICWS 2020 accepted paper:

Guosheng Kang, Jianxun Liu, Buqing Cao, Manliang Cao. "NAFM: Neural and Attentional Factorization Machine for Web API Recommendation". IEEE International Conference on Web Services. 2020, pp. 330-337.

1. Characteristics of the dataset

Item	Values
Number of Mashups	6206
Number of APIs	12919
Number of invocations	13107
Average number of invocations per Mashup	2.11
Number of called APIs	940
Called proportion of APIs	7.28%
Number of interactions	9297
Number of labeled Mashup tags	18601
Number of Mashup tags	403
Number of labeled API tags	44891
Number of API tags	473
Average length of Mashup description	27.63
Average length of API description	68.98
Number of Mashups with included APIs	5341
Sparsity of Mashup-API matrix	99.81%
Number of API category	383
Number of Mashup category	312
Number of API/Mashup categories	427

Run programme dataset_characteristics.py to generate the above statistical data.

2. Integrating FM Model with Deep Learning Models

Model	Reference	Code
basicFM	S. Rendle, “Factorization Machines,” 2010 IEEE International Conference on Data Mining, 2010, pp. 995-1000.	code
FNN	W. Zhang, T. Du, and J. Wang, “Deep learning over multi-field categorical data,” European conference on information retrieval, 2016, pp. 45-57.	code
Wide&Deep	H.-T. Cheng, L. Koc, J. Harmsen, T. Shaked, T. Chandra, H. Aradhye, G. Anderson, G. Corrado, W. Chai, and M. Ispir, “Wide & deep learning for recommender systems,” the 1st workshop on deep learning for recommender systems, 2016, pp. 7-10.	code
Deep&Cross	Y. Shan, T. R. Hoens, J. Jiao, H. Wang, D. Yu, and J. Mao, “Deep crossing: Web-scale modeling without manually crafted combinatorial features,” the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, 2016, pp. 255-262.	code
DeepFM	H. Guo, R. Tang, Y. Ye, Z. Li, and X. He, “DeepFM: A Factorization-Machine based Neural Network for CTR Prediction,” International Joint Conference on Artificial Intelligence, 2017, pp. 1725-1731.	code
NFM	X. He, and T.-S. Chua, “Neural factorization machines for sparse predictive analytics,” International ACM SIGIR conference on Research and Development in Information Retrieval, 2017, pp. 355-364.	code
AFM	J. Xiao, H. Ye, X. He, H. Zhang, F. Wu, and T.-S. Chua, “Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks,” International Joint Conference on Artificial Intelligence, 2017, pp. 3119-3125.	code

3. FM Model based Applications for Web API Recommendation

• basicFM

  • B. Cao, B. Li, J. Liu, M. Tang, and Y. Liu, “Web APIs Recommendation for Mashup Development based on Hierarchical Dirichlet Process and Factorization Machines,” International Conference on Collaborative Computing: Networking, Applications and Worksharing, 2016, pp. 3-15.
  • M. M. Rahman, X. Liu, and B. Cao, “Web API Recommendation for Mashup Development Using Matrix Factorization on Integrated Content and Network-based Service Clustering,” 2017 IEEE International Conference on Services Computing (SCC), 2017, pp. 225-232.
  • B. Cao, B. Li, J. Liu, M. Tang, Y. Liu, and Y. Li, “Mobile Service Recommendation via Combining Enhanced Hierarchical Dirichlet Process and Factorization Machines,” Mobile Information Systems, vol. 2019, 2019.

• DeepFM

  • X. Zhang, J. Liu, B. Cao, Q. Xiao, and Y. Wen, “Web Service Recommendation via Combining Doc2Vec-based Functionality Clustering and DeepFM-Based Score Prediction,” 2018 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Ubiquitous Computing & Communications, Big Data & Cloud Computing, Social Computing & Networking, Sustainable Computing & Communications, 2018, pp. 509-516.

• AFM:

  • Y. Cao, J. Liu, M. Shi, B. Cao, T. Chen, and Y. Wen, “Service Recommendation Based on Attentional Factorization Machine,” International Conference on Services Computing, Year, pp. 189-196.

4. NAFM: Neural and Attentional Factorization Machine for Web API Recommendation

4.1 Motivation

The state-of-the-art models for Web API recommendation, they either model factorized interactions with the same weight or neglect the the non-linear and complex inherent structure of real-world data. In real-world applications, different predictor variables usually have different predictive power, and not all features contain useful signal for estimating the target. Moreover, complex and non-linear structure are usually underlied in real-world data. Therefore, we propose a hybrid factorization machine model with a novel neural network architecture named NAFM by integrating neural network (i.e., neural component) to capture the non-linear feature interactions and attention network (i.e., attention component) to capture the different importance of feature interactions. We implemented NAFM using tensorflow 1.4.0 based on deepctr package. NAFM code

4.2 Data preprocessing

• Step1: Train Doc2Vec model with all Mashup and API description information, and selected 100 most popular APIs and the associated 1993 Mashups as our experimental data
  • Code: get_samples.py
  • Input: the raw dataset
  • Output: doc2vec.model, samples.pickle
• Step2: Process the experimental data and transform them into input data for models
  • Code: get_input_data.py
  • Input: doc2vec.model, samples.pickle
  • Output: input_data.csv
• Step3: Split the input data into two parts with 80% trainging data and 20% testing data
  • Code: process_input_data.py
  • Input: input_data.csv
  • Output: input_data.pickle
• Step4: Parameter optimization
  • Code: parameter_optimization.py
  • Input: input_data.pickle
  • Output: print the defaulf best parameter values

4.3 Evaluation

• Performance comparison
  • Code: run_input_data.py
  • Input: input_data.csv
  • Output: evaluation_results.csv
• Impact of parameters
  • Code: impact_of_parameters.py
  • Input: input_data.pickle
  • Output: Logloss_impact_of_parameters.csv, AUC_impact_of_parameters.csv
• Plot figures for impact of parameters
  • Code: plot_impact_of_parameters.py
  • Input: Logloss_impact_of_parameters.csv, AUC_impact_of_parameters.csv
  • Output: impact_of_parameters.png

5. Neural Network Architechtures and Comparison for FM based Models in Web API Recommendation

5.1 Neural network architechtures for FM based Models in Web API recommendation

Models	Architechtures
basicFM
DeepFM
AFM
NAFM

5.2 Comparison of FM based models for Web API recommendation

Model	Order-1 Features	Order-2 Features	High-order Feature Interactions	Discriminate the Importance of Order-2 Feature Interactions	High-order Input
basicFM	√	√	×	×	None
DeepFM	√	√	√	×	Embedding Vectors
AFM	√	√	×	√	None
NAFM	√	√	√	√	Order-2 Feature Interactions