--
Personalized
Collaborating Filtering
Memory Based
User
Item
Model Based
Matrix Factorization Based
Funk Matrix Factorization
SVD/SVD++
PMF/Bayesian PMF/ Non Linear PMF
None Negative MF
ALS
FM
FFM
RankFM
LTFM
BMF
Slope One
RMF
NPCA
IMF/BPR
Content Based Filtering
Item-Centric
User-Centric
Hybrid Methods
Meta-level
Feature combination
Feature augmentation
Mixed hybridization
Cascade hybridization
Switching hybridization
Weighted hybridization
Deep Learning Based
LLM Empowerd
Non-personalized
Most Popular Items
Recommendation
LightFM
Hybrid matrix factorization
Outperforms both collaborative and content-based models in cold-start or sparse interaction data scenarios
Semantic features encoding
can be used for related recommendation tasks such as tag recommendations
Provide several evaluation metrics
auc_score
precision_at_k
recall_at_k
reciprocal_rank
Implicit feedback recommender
Implements WARP (Weighted Approximate-Rank Pairwise) loss
Paralellized via HOGWILD SGD.
Battle tested by many developers and is very very fast
Different loss functions
logistic: useful when both positive (1) and negative (-1) interactions are present.
BPR: Bayesian Personalised Ranking 1 pairwise loss. Maximises the prediction difference between a positive example and a randomly chosen negative example.
WARP: Weighted Approximate-Rank Pairwise 2 loss. Maximises the rank of positive examples by repeatedly sampling negative examples until rank violating one is found. Useful when only positive interactions are present and optimising the top of the recommendation list (precision@k) is desired.
k-OS WARP: k-th order statistic loss 3. A modification of WARP that uses the k-th positive example for any given user as a basis for pairwise updates.
Deep Learning
Multi-Layer Perceptron Based Recommendation
NCF
DeepFM
Autoencoder Based Recommendation
AutoRec
Multi-VAE and Multi-DAE
CNN Based Recommendation
RNN Based Recommendation
GRU4Rec
Restricted Boltzmann Machines Based Recommendation
Neural Attention Models Based Recommendation
Attentive Collaborative Filtering
Hashtag Recommendation with
Topical Attention-Based LSTM
Neural AutoRegressive Based Recommendation
Deep Reinforcement Learning Based Recommendation
DRN
DEERS
Adversarial Networks Based Recommendation
LLMs Efficient
Methods
Model-Centric
Efficient Methods
Efficient Pre-Training
Model Compression
Efficient Fine-Tuning
Parameter-Efficient
Fine-Tuning
Adapter-based Tuning
Reparametrazation
Prefix Tuning
Prompt Tuning
Memory-Efficient Fine-Tuning
Efficient Inference
Algorithm-Level Inference Acceleration
System-Level Inference Acceleration
Efficient Architecture
Efficient Attention
- Sharing-based Attention
- Feature Information Reduction
- Kernelization or Low-Rank
- Fixed Pattern Strategies
- Learnable Pattern Strategies
- Hardware-Assisted Attention
Mixture of Experts (MoE)
Transformer-Alternative Architectures
PEFT Methods
Additive
Fine-Tunning
Adapter-based
Adapter Design
Serial Adapter, Parallel Adapter, CoDA
Multi-task Adaptation
AdapterFusion , AdaMix , AdapterSoup , MerA , Hyperformer
Soft Prompt-based
Soft Prompt Design
Prefix-tuning , p-tuning , Prompt-tuning , Xprompt, APrompt
Training Speedup
Reparameterized
Fine-Tuning
Low-rank
Decomposition
LoRA , Compacter KronA , VeRA , DoRA
LoRA Derivatives
Dynamic Rank
DyLoRA , AdaLoRA , SoRA , CapaBoost , AutoLoRA
LoRA Improvement
Laplace-LoRA , LoRA Dropout , LoRA+
Multiple LoRA
LoRAHub, MoLORA , MoA, MoLE, MixLoRA
Selective
Fine-Tunning
Unstructural Masking
U-Diff pruning , FishMask , Fish-Dip , AM
Structural Masking
S-Diff pruning ,S-BitFit , FAR, Bitfit, SPT