tanya34fish / fpmc

Factorizing Personalized Markov Chains for Next-Basket Recommendation (C++ version)

Basketrec

Next-Basket Recommendation

• This project implemented a BPR-based recommendation. It extends the FPMC model by considering 2 previous baskets to predict next basket.
• Origin paper: Factorizing personalized Markov chains for next-basket recommendation (WWW, 2010)

Implementation

• This project is modified from codes of Tag Recommender which is also proposed by Steffen Rendle.
• please see this website for details.
• http://www.informatik.uni-konstanz.de/rendle/software/tag-recommender/
• Reference: Steffen Rendle, Lars Schmidt-Thieme (2010): Pairwise Interaction Tensor Factorization for Personalized Tag Recommendation, in Proceedings of the Third ACM International Conference on Web Search and Data Mining (WSDM 2010), ACM.

Usage

• cd trigramrec-1.2.src
• make
• ./run_cv.sh 1 ipad # This will run the code that consider 2 previous basket to predict next basket.

Dataset

• cross_validation/train contains train files and cross_validation/test contains test files
• Format: userId sequenceId sequenceLength app_0 app_1 app_2 .... app_n

More Details of Implementation.

• Please see the following.

• src/basketrec/basketrec.cpp:

  1. this is the main function
  2. load data (use src/basketrec/src/Data.h)
  3. instantiate a NextBasketRecommender object. (use src/basketrec/src/NextBasketRecommender.h)
  4. instantiate a NextBasketRecommenderFPMC object.
  5. call "NextBasketRecommenderFPMC::train" to set up parameters and training.
  6. call "NextBasketRecommender::evaluate" for evaluation on test data.
  7. call "NextBasketRecommender::savePredict" to output for the prediction on test data.

• NextBasketRecommender

  • "NextBasketRecommender" is the base class. It contains basic function you will need.
  • You can extend this class when you want to implement other learning methods other than FPMC.
  • NextBasketRecommender::evaluate - This evaluates the results on the test data. We use MRR for evaluation metric.
  • NextBasketRecommender::predictTopItems - given user and basket, predict scores for every possible items.
  • NextBasketRecommender::testpredict - use for output the predict result.
  • NextBasketRecommender::savePrediction - call NextBasketRecommender::testpredict and output to file.

• basket_rec_fpmc.h:

  • "NextBasketRecommenderFPMC" is a derived class from "NextBasketRecommender".
  • It contains important parameters and functions you will need during the learning process.
  • NextBasketRecommenderFPMC::train - This will instantiate a "BasketLearnerBPR" object.
  • NextBasketRecommenderFPMC::init - assign the parameters and set up the latent matrices we need to learn.
  • NextBasketRecommenderFPMC::predict - given user, basket and an item, predict the score.
  • NextBasketRecommenderFPMC::learn - the main learning process.

• BasketLearnerBPR:

  • "BasketLearnerBPR" is a class used only for BPR setting.
  • BasketLearnerBPR::train - calculate how many baskets from training data. Implement the outer iteration and sampling for BPR.
  • on line 111, it will call "NextBasketRecommenderFPMC::learn" for the learning process.
  • on line 118, it will call "NextBasketRecommender::evaluate" for the evaluation.
  • BasketLearnerBPR::drawNextItemNeg - sample a negative item (used for BPR)