Jayveersinh Raj

BS20-DS-01

j.raj@innopolis.university

The project is Neural Collaborative filtering on movie lens 100k dataset.

Training

The training notebook can be found under notebooks. The model is trained on 100k samples of movie lens dataset

Evaluation

Metrics

Hit Rate (HR) : Hit Rate is a binary metric that measures the proportion of correct recommendations in a given recommendation list. It answers the question: "Did the user find at least one relevant item in the top-k recommendations?"

Normalized Discounted Cumulative Gain (nDCG): NDCG is a ranking metric that evaluates the quality of the entire ranked list of recommendations. It considers both the relevance of the items and their positions in the list.

Formulas for the both can be found under the figures directory of reports. Following are the numbers (Mean) over test set of 610 unique users not present in the training.

HR	nDCG	layer for MLP	Epochs
0.784	0.545	[64,32,16]	10

Example output

More details

For further details please refer to the report under the reports directory.

Reference

The project is based on: Neural Collaborative filtering

Citation

@article{DBLP:journals/corr/abs-1708-05031,
  author       = {Xiangnan He and
              Lizi Liao and
              Hanwang Zhang and
              Liqiang Nie and
              Xia Hu and
              Tat{-}Seng Chua},
  title        = {Neural Collaborative Filtering},
  journal      = {CoRR},
  volume       = {abs/1708.05031},
  year         = {2017},
  url          = {http://arxiv.org/abs/1708.05031},
  eprinttype    = {arXiv},
  eprint       = {1708.05031},
  timestamp    = {Mon, 13 Aug 2018 16:49:05 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-1708-05031.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}